Balancing Efficiency and Engagement: AI-Assisted Content for Research Communications in the RECOVER Initiative

Balancing Efficiency and Engagement: AI-Assisted Content for Research Communications in the RECOVER Initiative | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article Balancing Efficiency and Engagement: AI-Assisted Content for Research Communications in the RECOVER Initiative Zoe Lewczak, Praveen Mudumbi, Janelle Linton, Maika Mitchell, and 5 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7660686/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Introduction The growing availability of AI tools is transforming health and science communication by streamlining content creation and promotion. This study investigates the impact of AI-assisted research summaries on user engagement with the NIH-funded RECOVER program's website and evaluates the efficiency and readability of the content. Methods We analyzed Google Analytics 4 data from two distinct periods: one with entirely human-generated content and a second with AI-assisted content. We measured changes in page views, active users, and average engagement time, and assessed the review time and readability of the AI-enhanced summaries. Results There was no significant change in page views or active users between the two periods. However, average engagement time increased by 4.37 seconds (P = .0461), suggesting AI-assisted content may be more compelling. Human review of AI-drafts averaged 19.88 changes, and readability improved, with the mean Flesch-Kincaid grade level decreasing from 12.28 to 11.56. Conclusion This study demonstrates that AI can be a valuable tool for accelerating the creation of accessible and engaging content. Our findings highlight a crucial balance: while AI can save effort and reduce cost in public engagement efforts, human oversight remains essential to ensure the accuracy, clarity, and accessibility of vital health communications. Health sciences/Medical research/Translational research Scientific community and society/Scientific community/Publishing/Media Artificial Intelligence Plain Language Summaries Lay Language Research Health Communications Science Communications Community Engagement Media Promotion Research Promotion Figures Figure 1 Figure 2 Introduction Communication is a cornerstone of any successful research initiative, and for health and science organizations, it is critical for establishing credibility and engaging the public. Effective scientific communication goes beyond simply reporting findings; it involves making complex information accessible, balanced, and respectful to a diverse audience. 1 Technical manuscripts are like a professional chef's intricate recipe—full of jargon and complex techniques that are inaccessible to most. Effective communication is the skilled cookbook author who translates that recipe into simple, step-by-step instructions so anyone can prepare and enjoy the result. The rapid advancement of Artificial Intelligence (AI) has revolutionized this field, offering new ways to streamline and enhance communication and media promotion efforts. 2 A recent survey by the Institute for Public Relations found that 45% of organizations now use AI for various tasks, including content creation, ideation, and market research. 3 For communication professionals, AI has become a powerful tool to synthesize vast amounts of information, identify key themes, and inform strategic messaging. 4 This paper explores how the Researching COVID to Enhance Recovery (RECOVER) program, funded by the National Institutes of Health (NIH), has strategically integrated AI into its communication workflows. We focus on its use in drafting plain-language summaries and video scripts, which are essential for broad media promotion and public engagement. RECOVER Communications Teams using AI NYU's Clinical Science Core (CSC) for NIH RECOVER serves as the scientific leadership and operational backbone of the RECOVER Initiative, set up to better understand post-acute sequelae of SARS-CoV-2 infection, known as Long COVID. CSC has led all aspects of developing and conducting studies, including the dissemination of results for the RECOVER cohort studies at > 200 enrolling sites located in all 50 states and Puerto Rico. Since their inception in 2021, the RECOVER cohort studies enrolled a diverse population of nearly 30,000 adult and pediatric participants, collected > 50,000 biospecimens, and have over 140 reports published or under review. To ensure this research reaches the widest possible audience, the CSC employs three distinct communication teams: Community Engagement, External Affairs, and internal Communications. These teams work in tandem to transform complex scientific findings into accessible formats for various media channels. A core principle of RECOVER is to ensure patient and community perspectives are integrated throughout the research and reflected in all communications. Historically, scientific publications are written by and for the scientific community, making them difficult for non-specialists to understand. This creates a significant barrier to the public accessing potentially life-changing research. 5 To bridge this gap, RECOVER's communication teams have embraced Plain Language Summaries (PLS) and videos as crucial tools. A PLS translates complex research into easy-to-understand language, a process that is now being optimized with AI. 6 The team leverages NYUChatGPT , an AI tool developed by New York University (NYU) that leverages OpenAI's GPT-3 technology to assist with a variety of tasks, including generating text-based content to help the NYU community. 7 A finalized, peer-reviewed manuscript serves as the input for NYUChatGPT and is guided by a prompt. This carefully crafted prompt, aligned with our established plain language standards guide, is provided to NYUChatGPT (See Fig. 1 ). As a result, a plain language summary of the manuscript is returned. This use of AI allows the team to accelerate the creation of accessible content, thereby speeding up media promotion efforts and the overall dissemination of critical health information. This prompt instructs NYUChatGPT to generate a PLS that adheres to specific readability levels (6th grade), avoids jargon, and effectively communicates the key findings and implications of the research to a non-scientific audience. This NYUChatGPT-produced initial draft serves as a foundation for the team to further refine and polish, ensuring accuracy, clarity, and accessibility for the target audience. The final draft is sent to the manuscript’s authors to ensure accuracy prior to publishing on the RECOVER website. Digital media allows for user-friendly presentations of research findings: it puts evidence into practice, allowing health professionals and the public to better understand key research findings. 8 Images and video can also be integral aspects of data collection, analysis, and reporting research studies. 9 Digital media content has become a powerful platform for public engagement with science. By offering PLS in video format and incorporating infographics, we can more widely disseminate our findings. The Discover RECOVER series, launched by the Community Engagement (CE) team in collaboration with researchers and Patient, Caregiver, and Community Representatives, communicates manuscript findings in brief videos. The CE team utilizes NYUChatGPT in creating plain-language scripts for Discover RECOVER. AI is prompted to synthesize and summarize the key points of an uploaded manuscript in simple language. The CE team used the key points to draft the script in English. The script is then reviewed by the authors and the Patient, Caregiver, and Community Representatives, ensuring that the AI-generated content accurately reflects the key findings of the manuscript. In alignment with RECOVER’s commitment to equal access in research, Discover RECOVER video scripts and infographics are translated into Spanish for improved access to findings. The CE team utilizes NYUChatGPT to draft the first version of Spanish subtitles and translated infographics for the videos. The CE team inputs the final English version of the Discover RECOVER script into NYUChatGPT and instructs the AI to generate a plain language script at a 6th grade reading level with active voice utilizing a Spanish Translations Plain Language Living Word Bank. This Spanish Translation Word Bank is a project managed by the CSC and Administrative Coordinating Center (ACC) Communications team that includes standard translation preferences for RECOVER deliverables. The AI-generated script is then shared with CSC Communications for quality assurance. The Communications team reviews the script to ensure the translation is in accordance with the preferred terms for RECOVER. This is an important step in the process as generative AI often makes translation errors, being unable to capture the complexity of healthcare information. Methods To evaluate the effect of generative-AI enhanced content on web traffic and engagement for the RECOVER website, recovercovid.org, we used Google Analytics 4 (GA4) to compare user interactions from two distinct time periods: December 1, 2023 to September 10, 2024, when all content was human-generated, and September 10, 2024 to January 20, 2025, when NYUChatGPT was deployed to assist with all research summaries, written and video, in English and Spanish. GA4 was configured to collect data on the following metrics: Views. The number of mobile app screens or web pages that users saw. Repeated views of a single screen or page are counted. Active Users. The number of distinct users who visited the recovercovid.org website. Average engagement time per active user. The average time that recovercovid.org was in focus in an active user’s browser or app, expressed in seconds. We examined traffic and engagement metrics on: The “Research Summaries” page, which contains the PLS of various RECOVER studies and hyperlinks to the externally published full studies. The "Videos” page, which hosts “Discover RECOVER” and other video assets. The “Publications” page, which hosts a separate page for every RECOVER publication, links to their plain language deliverables (summaries and/or videos), and to the externally published article. Paired t-tests were conducted to determine whether observed differences in website traffic and engagement between the two time periods were statistically significant. To assess the human oversight required to develop resources with the assistance of generative-AI, we performed a raw count of the number of changes made by human reviewers from the original AI-generated draft to the finalized resource. This assessment was performed on all AI-assisted video scripts and written PLS published between September 10, 2024 and January 20, 2025. To ensure readability of the AI-enhanced PLS and English-language video scripts, we employed the Flesch-Kincaid Readability Test. This established and widely used metric calculates a grade level equivalent for written text, effectively estimating the years of education required for comprehension. 10 Flesch-Kincaid Readability levels are a composite of the number of words, sentences, and syllables in a piece of text. 11 A lower Flesch-Kincaid grade level indicates greater readability and broader accessibility. Given the importance of clear communication in our PLS, we targeted a 6th-grade reading level. While NYUGPT was prompted to generate content at this target level, the initial AI-generated drafts exhibited a significantly higher average reading level of 12th grade. Therefore, the Flesch-Kincaid test served as an objective measure of the discrepancy between the intended and actual complexity of the AI-generated text. Specifically, we assessed the Flesch-Kincaid level for both the initial AI-generated draft and the final, human-revised version of each PLS and English-language video script. This two-stage assessment allowed us to quantify the impact of human editing on improving readability and achieving our target grade level, thereby ensuring the PLS and video scripts are accessible to our audience. Data collection was performed continuously throughout both time periods, and all data were anonymized and aggregated to maintain privacy and ensure compliance with data protection regulations. Results Between the two time periods, the change in views and active users was not statistically significant. However, there was an average increase in the engagement time of 4.37 seconds (95% CI 0.076s – 8.67s)( P = .0461) per active user. Between the two time periods, there was a general decrease in views and active users on the RECOVER homepage, publications page, and the research summaries page (see Table 1 ). However, views and active users on the videos page continued to increase between the two time periods, 25% increase in views and 15% increase in active users, suggesting the video content continued to drive interest even as general interest declined. Table 1 Change in views, active users, and average engagement time Views Active Users Average engagement time per active user (seconds) Page P1 P2 % Change P1 P2 % Change P1 P2 % Change* Homepage 87883 31855 -63.753 56104 20761 -62.996 29.304 22.991 -21.543 Publications 17939 8397 -53.191 5179 2604 -49.720 78.215 57.871 -26.010 Research Summaries 7891 3317 -57.965 3220 1210 -62.422 74.428 83.799 12.591 Videos 536 673 25.560 358 410 14.525 35.824 42.510 18.663 P1 = The time period from December 1, 2023 to September 10, 2024 P2 = The time period from September 10, 2024 to January 20, 2025 *Statistically significant, P = 0.046. All AI-enhanced deliverables required extensive human revision to ensure that they were clear, concise, accurate, and in plain language. We performed a count of the number of material changes from the initial AI-generated draft to the final product. The mean number of material changes for all AI-enhanced deliverables (n = 17) was 19.875 (SD = 15.046, 95% CI [11.858, 27.892]). We analyzed the Flesch-Kincaid Readability (FKR) grade level, ranging from 0 to 18 (Fig. 2 ), for the initial AI-generated draft of the content and the final, human-reviewed product. For all AI-enhanced deliverables, the mean FKR level of the initial AI-draft was 12.276 (SD = 1.810, 95% CI [11.346, 13.207]) while the mean FKR level for the final products was 11.56 (SD = 2.129, 95% CI [10.464, 12.654]). While both the original and final products still carry a mean FKR level equivalent to a 12th grade reading level, we observed a 6.02% decrease in the FKR level after human review across all AI-enhanced products (Table 2 ). This decrease in readability level represents a simplification of the text after human-review. Table 2 Human-reviewer changes and Flesch-Kincaid Grade Levels for all AI-enhanced deliverables # of Changes Final product word count AI-Draft Grade Level* Final Draft Grade Level* All AI-enhanced deliverables (n = 17) Mean 19.875 283.176 12.276 11.559 SD 15.046 164.101 1.810 2.129 95% CI [11.858, 27.892] [198.803, 367.549] [11.346, 13.207] [10.464, 12.654] Videos (n = 5) Mean 39.600 502.600 14.040 13.740 SD 7.987 139.581 0.789 1.141 95% CI [29.682, 49.518] [329.288, 675.912] [13.06, 15.02] [12.323, 15.157] PLS (n = 12) Mean 10.909 191.750 11.542 10.650 SD 5.576 32.886 1.592 1.749 95% CI [7.163, 14.655] [170.855, 212.645] [10.53, 12.553] [9.539, 11.761] *Grade Level = Flesch-Kincaid Readability Grade Level. Figure 2 : The Flesch-Kincaid Grade Level (0–18) assesses text readability, with lower scores indicating easier comprehension. This figure shows the reading level spectrum, from "basic" (0–5) to "advanced" (13–18), with the "average" range ( 6 – 12 ) considered ideal for Plain Language. RECOVER targets a 6th-grade reading level to maximize accessibility for diverse audiences, including patients, caregivers, and the public. Each Publications page is devoted to a different scientific publication, and thus different scientific topics. We used the increase in average user engagement time for each publications page as a surrogate for public interest in various research topics. To determine topics of interest to the end user, we calculated the percent change in average active user engagement time for each publication’s webpage between the two time periods and then evaluated the subject matter of those pages that showed an increase in average engagement time. Based on this analysis, we were able to identify 10 general research topics that continue to garner user engagement (see Table 3 ). The complete list of topics that saw increased engagement time between the two time periods can be found in S_1: Research topics and subtopics that continue to see increased average user engagement time. Table 3 Research topics that continue to see increased average user engagement time Topic # of webpages dedicated to this topic with increased average user engagement time Cardiopulmonary issues 18 Neuropsychiatric issues 17 PASC Characteristics 11 Infectious Disease issues 9 Investigating techniques 8 Miscellaneous 7 Sleep/Fatigue issues 7 Gastrointestinal issues 5 Investigating Lab measures 4 ENT issues 3 OB/GYN issues 2 Genitourinary issues 1 Discussion In RECOVER, AI is leveraged to hasten distillation of research for PLS and video presentations. By automating these initial drafts, our teams save nearly 4 hours per project, significantly reducing the typical 10 + hour process. With an anticipated 25–30 manuscripts annually, this translates to a 200–240 hour reduction in workload for PLS development and video scripts. This time saving allows communication teams to dedicate more resources to other projects, e.g. media promotion. This faster turn-around improves web user engagement through timely dissemination of information. Our analysis revealed the crucial role of human review and quality control in developing effective scientific and health communications. By prioritizing readability, human review can enhance accessibility, facilitating a broader understanding of scientific and health-related information. This improved comprehension is essential for informed decision-making regarding personal health and engagement with scientific advancements. A focus on readability is not merely a stylistic preference, but a critical component of responsible scientific communication and health education. 12 Our results indicate that AI-generated output often requires substantial editing to meet desired readability targets. Nine out of the twelve AI-generated PLS had a noticeably higher Flesch-Kincaid grade level than the final, human-revised versions. This discrepancy underscores the current limitations of AI in independently producing summaries that consistently adhere to plain language principles. However, with appropriate human oversight, AI tools enable speedier dissemination, improved readability, and greater team efficiency. By implementing a strategic approach to AI utilization, we aimed to enhance the quality and accessibility of our research summaries and video scripts. This refined approach allowed us to produce summaries and video scripts that were not only accurate but also more engaging and understandable to a broader audience. Subsequently increasing average engagement time, validating the effectiveness of these AI-assisted strategies. Optimizing for readability and relevance, we have successfully captured and maintained public interest in key research topics. This data-driven approach empowers our teams to make informed decisions about future content development, ensuring that our deliverables continue to meet the evolving needs and interests of our audience. Looking to the future, we would further streamline the communication pipeline between investigators and the public. Information directly from investigators can be challenging to translate, leading to delays and potential misunderstandings. AI has the potential to not only reduce wait times in this process but also expedite the creation of understandable content. Our goal is to achieve consistently simple PLS and video scripts that are highly accessible in multiple languages, ensuring a wider reach. Limitations This study, while demonstrating the potential of AI in generating PLS for scientific research, also highlights important limitations that warrant consideration. A primary limitation is the reliance on a relatively small sample size. Only twelve PLS and five videos (English and Spanish) were generated with AI assistance, limiting the generalizability of our findings. Future research with a larger and more diverse dataset of scientific articles, engagement data, and target reading levels is necessary to validate these initial observations. While NYUChatGPT shows promise for content generation and translation of technical documents into accessible language, its capabilities may not fully capture the nuance required to communicate complex chronic diseases, particularly as it relates to patients’ lived experiences. Furthermore, it may generate responses with hallucinated research material. 13 The risk of misrepresenting research findings or presenting them in a distorted fashion trivializing the experiences of marginalized groups remains a significant concern. AI systems may perpetuate erroneous information— amplifying demographic health disparities or supporting since-discredited psychosomatic models of disease, thus preventing patients from getting the help they need. 14 These concerns are especially relevant for communication aimed at informing decision-making, particularly regarding health outcomes. Similarly, our framework also has methodological limitations in measuring impact. While viewership metrics can capture surface-level engagements, tracking changes at the level of policy decisions or clinical practice remains significantly more challenging. Our current system effectively measures engagement metrics, such as views, active users, and engagements over time. However, it doesn't directly measure whether those activities lead to real-world impact, such as improved health or effective policy decisions. As a result, we may overemphasize quantitative metrics at the expense of qualitative real-world impact. Engagement and web traffic metrics alone cannot provide a complete picture of the effect of AI-generated content on communicating findings. These metrics cannot account for confounding factors such as the general interest in science and Long COVID, the limitations in the granularity of the available data, and the inherent limitations in using Google Analytics data. During our study period, views on the RECOVER homepage dropped 64%, which could indicate a general decreased interest in Long COVID. However, these metrics cannot explain why this decrease occurred. They cannot account for the changing socio-political climate that may impact user engagement. Similarly, we are limited by the granularity of the data available. Our data does not indicate whether a Publication’s summary was interacted with, only that the page was accessed. It is possible that not all users were exposed to AI-generated content when they viewed each page. Our analysis is also subject to the limitations inherent in using Google Analytics data. For example, a user who visits the RECOVER website, clears their browser cookies, and returns to the site, is considered a new user, potentially overestimating the number of unique users. Conversely, if two users visit the RECOVER website on the same device (e.g. a shared computer), they are recorded as a single user, potentially underestimating the number of unique users. Our work provides empirical evidence of the current capabilities and limitations of AI in a specific context – generating PLS, video scripts, and synthesizing evidence for communications. While AI holds promise, it is not yet a perfect solution and requires careful oversight. The potential for biases in AI algorithms calls for greater transparency in AI-generated content requiring further investigation. 15 Our findings demonstrate that human input remains essential to refine AI-generated text, ensuring clarity, conciseness, and appropriate readability levels for target audiences. This necessity for human intervention has implications for the efficiency gains promised by AI tools and highlights the importance of incorporating human review into any AI-assisted workflow. Just as translating science is crucial for informing research, it's equally relevant to ensure the research reaches the people it affects. Communications teams can leverage AI to synthesize information from diverse sources to develop informed and accurate messaging for various audiences. 16 Members of these teams are expected to make highly technical research accessible. The use of AI posits itself as a useful tool for making research digestible, however, the quality and reliability of AI-generated summaries depend heavily on the data it's trained on and the algorithms used. 17 Once again, human oversight remains critical to ensure accuracy, avoid misinterpretations, and address potential biases that could inadvertently shape communication. The current discourse surrounding AI development reflects a tension between rapid advancement and cautious implementation. 18 As institutions seek to integrate AI into established practices like scientific and strategic communications, future research should explore the ethical dimensions to ensure responsible and beneficial use of AI in disseminating scientific knowledge and shaping public discourse. 19 Finally, this study used the Flesch-Kincaid Grade Level as a metric of readability. 20 Future work should explore other readability metrics and qualitative assessments of PLS and video scripts to provide a more comprehensive evaluation of AI-generated content. 21 Conclusion Integrating AI into scientific communication for the NIH RECOVER program offers opportunities for efficiency in content creation, as demonstrated by NYUChatGPT's effectiveness in generating PLS without reducing engagement. However, our findings underscore the necessity of human oversight to ensure accuracy, clarity, and accessibility, as AI-generated content often requires revision. While AI modestly increased user engagement, it struggles with the nuances of scientific research and patient experiences, highlighting the risk of misrepresentation in sensitive areas like health disparities. Despite AI's potential to synthesize and translate information, it carries risks of bias and misinterpretation. Responsible AI implementation in scientific communication demands careful human review, ongoing refinement of AI tools, and a focus on both accuracy and audience sensitivity. Declarations Sources of Funding: National Institutes of Health (NIH) Other Transactional Authority Agreements OT2HL161847. https://www.nih.gov/ Disclaimer: The content is solely the responsibility of the authors and does not necessarily represent the official views of the RECOVER Program, the NIH or other funders Conflict of Interest: NJ reports participation on the Advisory Board for the RECOVER Vital Clinical Trial. All other authors report no conflicts of interest. Reporting Guideline: SQUIRE Research Question: What is the effect of deploying AI-enhanced summaries of research findings on user engagement with the NIH-funded Researching COVID to Enhance Recovery (RECOVER) program’s website? Funding This research was funded by National Institutes of Health (NIH) Agreement OTA OT2HL161847 as part of the Researching COVID to Enhance Recovery (RECOVER) research program. Disclosures Authorship has been determined according to ICMJE recommendations. Data Availability Statement The data that supports the findings of this study are available in the supplementary material of this article. Acknowled gements We wish to thank the National Community Engagement Group (NCEG), all Patient, Caregiver and Community Representatives, and all the participants enrolled in the RECOVER Initiative, with special thanks to Juan Lewis for their support. References Jucan M, Jucan C (2014) The power of science communication. Procedia - Soc Behav Sci 149:288–292. 10.1016/j.sbspro.2014.08.288 Panda G, Upadhyay AK, Khandelwal K (2019) Artificial intelligence: A strategic disruption in public relations. J Creat Commun 14(3):196–213. 10.1177/0973258619866585 Davis-Gilbert OK, Fajardo O (2024) The impact of AI on the bottom line in PR. Institute for Public Relations . January 29, Available from: https://www.instituteforpr.org/the-impact-of-ai-on-the-bottom-line-in-pr/ Biswal SK (2020) The space of artificial intelligence in public relations: The way forward. In: Kulkarni A, Satapathy S (eds) Optimization in machine learning and applications: Algorithms for intelligent systems. Springer, Singapore, pp 169–176. doi: 10.1007/978-981-15-0994-0_11 . Barreto JOM, de Melo RC, da Silva LALB, de Araújo BC, de Freitas Oliveira C, Toma TS, de Bortoli MC, Demaio PN, Kuchenmüller T (2024) Research evidence communication for policymakers: A rapid scoping review on frameworks, guidance and tools, and barriers and facilitators. Health Res Policy Syst 22(1):99. 10.1186/s12961-024-01169-9 PMID: 39118156; PMCID: PMC11312384 Dormer L, Schindler T, Williams LA et al (2022) A practical ‘how-to’ guide to plain language summaries (PLS) of peer-reviewed scientific publications: Results of a multi-stakeholder initiative utilizing co-creation methodology. Res Involv Engagem 8:23. 10.1186/s40900-022-00358-6 New York University IT, Generative AI services [Internet]. New York: New York University; [cited 2024 Jan 4]. Available from: https://www.nyu.edu/life/information-technology/artificial-intelligence-at-nyu/generative-ai-services.html#private Buljan I, Malički M, Wager E, Puljak L, Hren D, Kellie F, West H, Alfirević Ž, Marušić A (2018) No difference in knowledge obtained from infographic or plain language summary of a Cochrane systematic review: Three randomized controlled trials. J Clin Epidemiol . ;97:86–94. 10.1016/j.jclinepi.2017.12.003 . PMID: 29269021 Walker EB, Boyer DM (2018) Research as storytelling: the use of video for mixed methods research. Video J Educ Pedag 3:8. 10.1186/s40990-018-0020-4 Flesch R (1948) A new readability yardstick. J Appl Psychol. ;32(3):221 – 33. 10.1037/h0057532 . PMID: 18867058 Flesch R (2016) How to write plain English [Internet]. University of Canterbury; Jul 12 [cited 2025 Feb 12]. Available from: https://web.archive.org/web/20160712094308/http://www.mang.canterbury.ac.nz/writing_guide/writing/flesch.shtml Redish J (2000) Readability formulas have even more limitations than Klare discusses. ACM J Comput Doc 24(3):132–137. 10.1145/344599.344637 Footnotes Jucan M, Jucan C. The power of science communication. Procedia - Soc Behav Sci . 2014;149:288 − 92. doi: 10.1016/j.sbspro.2014.08.288 . Panda G, Upadhyay AK, Khandelwal K. Artificial intelligence: A strategic disruption in public relations. J Creat Commun . 2019;14(3):196–199. doi: 10.1177/0973258619866585 . Davis-Gilbert OK, Fajardo O. The impact of AI on the bottom line in PR. Institute for Public Relations . January 29, 2024. Available from: https://www.instituteforpr.org/the-impact-of-ai-on-the-bottom-line-in-pr/ . Biswal SK. The space of artificial intelligence in public relations: The way forward. In: Kulkarni A, Satapathy S, editors. Optimization in machine learning and applications: Algorithms for intelligent systems . Singapore: Springer; 2020. p. 169–172. doi: 10.1007/978-981-15-0994-0_11 . Dormer L, Schindler T, Williams LA, et al. A practical ‘How-To’ Guide to plain language summaries (PLS) of peer-reviewed scientific publications: results of a multi-stakeholder initiative utilizing co-creation methodology. Res Involv Engagem. 2022;8:23. https://doi.org/10.1186/s40900-022-00358-6 Barreto JOM, de Melo RC, da Silva LALB, de Araújo BC, de Freitas Oliveira C, Toma TS, de Bortoli MC, Demaio PN, Kuchenmüller T. Research evidence communication for policymakers: A rapid scoping review on frameworks, guidance and tools, and barriers and facilitators. Health Res Policy Syst . 2024 Aug 8;22(1):99. doi: 10.1186/s12961-024-01169-9 . PMID: 39118156; PMCID: PMC11312384. New York University IT. Generative AI services [Internet]. New York: New York University; [cited 2024 Jan 4]. Available from: https://www.nyu.edu/life/information-technology/artificial-intelligence-at-nyu/generative-ai-services.html#private . Buljan I, Malički M, Wager E, Puljak L, Hren D, Kellie F, West H, Alfirević Ž, Marušić A. No difference in knowledge obtained from infographic or plain language summary of a Cochrane systematic review: Three randomized controlled trials. J Clin Epidemiol. 2018 May;97:86–94. doi: 10.1016/j.jclinepi.2017.12.003 . PMID: 29269021. Walker EB, Boyer DM. Research as storytelling: the use of video for mixed methods research. Video J Educ Pedag. 2018;3:8. https://doi.org/10.1186/s40990-018-0020-4 Flesch R. A new readability yardstick. J Appl Psychol. 1948 Jun;32(3):221 − 33. doi: 10.1037/h0057532 . PMID: 18867058. Flesch R. How to Write Plain English [Internet]. University of Canterbury; 2016 Jul 12 [cited 2025 Feb 12]. Available from: https://web.archive.org/web/20160712094308/http://www.mang.canterbury.ac.nz/writing_guide/writing/flesch.shtml Jucan M, Jucan C. The Power of Science Communication. Procedia - Soc Behav Sci. 2014;149:288–292. https://doi.org/10.1016/j.sbspro.2014.08.288 New York University IT. Generative AI services [Internet]. New York: New York University; [cited 2024 Jan 4]. Available from: https://www.nyu.edu/life/information-technology/artificial-intelligence-at-nyu/generative-ai-services.html#private . Panda G, Upadhyay AK, Khandelwal K. Artificial intelligence: A strategic disruption in public relations. J Creat Commun . 2019;14(3):208–210. doi: 10.1177/0973258619866585 . Panda G, Upadhyay AK, Khandelwal K. Artificial intelligence: A strategic disruption in public relations. J Creat Commun . 2019;14(3):208–210. doi: 10.1177/0973258619866585 . Panda G, Upadhyay AK, Khandelwal K. Artificial intelligence: A strategic disruption in public relations. J Creat Commun . 2019;14(3):208–210. doi: 10.1177/0973258619866585 . Dormer L, Schindler T, Williams LA, et al. A practical ‘how-to’ guide to plain language summaries (PLS) of peer-reviewed scientific publications: Results of a multi-stakeholder initiative utilizing co-creation methodology. Res Involv Engagem . 2022;8:23. doi: 10.1186/s40900-022-00358-6 . Davis-Gilbert OK, Fajardo O. The impact of AI on the bottom line in PR. Institute for Public Relations . January 29, 2024. Available from: https://www.instituteforpr.org/the-impact-of-ai-on-the-bottom-line-in-pr/ . Dormer L, Schindler T, Williams LA, et al. A practical ‘How-To’ Guide to plain language summaries (PLS) of peer-reviewed scientific publications: results of a multi-stakeholder initiative utilizing co-creation methodology. Res Involv Engagem. 2022;8:23. https://doi.org/10.1186/s40900-022-00358-6 . Flesch R. How to write plain English [Internet]. University of Canterbury; 2016 Jul 12 [cited 2025 Feb 12]. Available from: https://web.archive.org/web/20160712094308/http://www.mang.canterbury.ac.nz/writing_guide/writing/flesch.shtml . Redish J. Readability formulas have even more limitations than Klare discusses. ACM J Comput Doc . 2000;24(3):132-5. doi: 10.1145/344599.344637 . Additional Declarations Yes there is potential Competing Interest. Nita Jain describes the relevant competing interest: advisory board member for the RECOVER Vital Clinical Trial. Supplementary Files SupplementaryTable12.docx Supplementary Table 1 Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-7660686","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":517807918,"identity":"fb4765fb-edbd-42d0-aaee-7818355c35b6","order_by":0,"name":"Zoe Lewczak","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAABHElEQVRIie2QMUvEMBSAXwm0S0rXHCfXv5AiKHKDfyXlIF0KKi6CcihCuwRc9V84dXKIBDKd3HrQ5aAQlxsOXBTkzno3iSl4m0M+eEl44Xvv8QAcjv/KfHP6XvN90W0OtcG6le2Xj/Z3Vvz+n5S4vJXALlQcl5pfnuVjOAxeDPl4GkIU5NSmeEIzYBOV3Aiu64dKwZHIDnrCZNATC6uCSFsqLWqGICvqsJJAJUcUSwV0Zu/ik5MlpKua+dFrcR5WY6BTg5LPVjnuUDDJAdLrmmHCNQor1BbnXrPpQuwKwZoC0+vknphRP6wUpjPjNXsyw2RiTjs2NoflFY/jO568tYMN6JTD80IOB1E5erRuGSB4/zHpr4fD4XA4ducLje9cndtMzzQAAAAASUVORK5CYII=","orcid":"https://orcid.org/0009-0001-8391-9383","institution":"Harvard Medical School","correspondingAuthor":true,"prefix":"","firstName":"Zoe","middleName":"","lastName":"Lewczak","suffix":""},{"id":517807919,"identity":"d6b102ef-972e-4b73-b94b-37efec888a3b","order_by":1,"name":"Praveen Mudumbi","email":"","orcid":"https://orcid.org/0000-0003-3792-1789","institution":"New York University Grossman School of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Praveen","middleName":"","lastName":"Mudumbi","suffix":""},{"id":517807920,"identity":"fa31b497-5cfe-4e0c-a4c7-bc51fa8636b7","order_by":2,"name":"Janelle Linton","email":"","orcid":"","institution":"NYU Grossman School of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Janelle","middleName":"","lastName":"Linton","suffix":""},{"id":517807921,"identity":"c36dfaa1-30ba-4a18-b42d-599e38a81d5c","order_by":3,"name":"Maika Mitchell","email":"","orcid":"","institution":"NYU Grossman School of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Maika","middleName":"","lastName":"Mitchell","suffix":""},{"id":517807922,"identity":"9080b07e-4816-4745-9aef-cf8e7e0ba474","order_by":4,"name":"Jasmine Briscoe","email":"","orcid":"","institution":"NYU Grossman School of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Jasmine","middleName":"","lastName":"Briscoe","suffix":""},{"id":517807923,"identity":"6f29f79b-2053-4ff8-86e0-38f59ab93507","order_by":5,"name":"Pricilla Short","email":"","orcid":"","institution":"NYU Grossman School of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Pricilla","middleName":"","lastName":"Short","suffix":""},{"id":517807924,"identity":"c57a4d15-0cdf-43e3-aa65-e877b173f9a9","order_by":6,"name":"Nita Jain","email":"","orcid":"https://orcid.org/0000-0002-4197-937X","institution":"Timeless Biosciences","correspondingAuthor":false,"prefix":"","firstName":"Nita","middleName":"","lastName":"Jain","suffix":""},{"id":517807925,"identity":"791fbc0e-32ed-43e9-9d09-7edefda4d2df","order_by":7,"name":"Anisha Sekar","email":"","orcid":"","institution":"RECOVER Patient, Caregiver, or Caregiver Representative","correspondingAuthor":false,"prefix":"","firstName":"Anisha","middleName":"","lastName":"Sekar","suffix":""},{"id":517807926,"identity":"51d03df2-3e63-41ab-81b0-cbdf9bfb30b0","order_by":8,"name":"Alicia Chung","email":"","orcid":"https://orcid.org/0000-0001-9330-1416","institution":"NYU Grossman School of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Alicia","middleName":"","lastName":"Chung","suffix":""}],"badges":[],"createdAt":"2025-09-19 17:50:34","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-7660686/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7660686/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":92229412,"identity":"c6f31532-5f96-446a-b889-c888717f5287","added_by":"auto","created_at":"2025-09-26 06:09:06","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":258644,"visible":true,"origin":"","legend":"","description":"","filename":"BalancingEfficiencyandEngagement.docx","url":"https://assets-eu.researchsquare.com/files/rs-7660686/v1/bdd0e39d51716b26da501941.docx"},{"id":92229413,"identity":"330bbaa8-5032-4afe-b690-6348e39af427","added_by":"auto","created_at":"2025-09-26 06:09:06","extension":"pdf","order_by":2,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":175906,"visible":true,"origin":"","legend":"","description":"","filename":"Figure2.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7660686/v1/56f75ff1cf35c7dea037b25e.pdf"},{"id":92229414,"identity":"7bd6c394-5b4f-437e-b9a4-21999f5927a8","added_by":"auto","created_at":"2025-09-26 06:09:06","extension":"json","order_by":3,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":9867,"visible":true,"origin":"","legend":"","description":"","filename":"NCOMMS2574553.json","url":"https://assets-eu.researchsquare.com/files/rs-7660686/v1/c0935c3e84662db4e2f7af7d.json"},{"id":92229415,"identity":"f43ed4d9-6c4f-481e-b19d-12ee517f20db","added_by":"auto","created_at":"2025-09-26 06:09:06","extension":"xml","order_by":4,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":76276,"visible":true,"origin":"","legend":"","description":"","filename":"NCOMMS25745530enriched.xml","url":"https://assets-eu.researchsquare.com/files/rs-7660686/v1/6b6fea578d9d6b0da58d14b0.xml"},{"id":92229418,"identity":"45d09d86-cf55-4fde-ad2a-fab488cc64f9","added_by":"auto","created_at":"2025-09-26 06:09:06","extension":"pdf","order_by":5,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":37110,"visible":true,"origin":"","legend":"","description":"","filename":"Figure1.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7660686/v1/d9e6173279570b48b937e3c7.pdf"},{"id":92229825,"identity":"1fa1e66a-ce4a-4273-92e2-e28d6417d614","added_by":"auto","created_at":"2025-09-26 06:17:06","extension":"pdf","order_by":6,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":175906,"visible":true,"origin":"","legend":"","description":"","filename":"Figure2.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7660686/v1/07b23935de733417cc5354f2.pdf"},{"id":92229416,"identity":"a0e91865-976b-4705-9cb2-aaa081b75031","added_by":"auto","created_at":"2025-09-26 06:09:06","extension":"png","order_by":7,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":111790,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-7660686/v1/8ce40f7feec981084c336ddd.png"},{"id":92229410,"identity":"114f9f85-1443-4c42-a1c1-9f186caedc3f","added_by":"auto","created_at":"2025-09-26 06:09:06","extension":"png","order_by":8,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":47526,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-7660686/v1/e92f45e8d0a8a23fa1c31b2c.png"},{"id":92229422,"identity":"4ed80937-20d2-4cac-b2f4-b2400291bf7b","added_by":"auto","created_at":"2025-09-26 06:09:06","extension":"png","order_by":9,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":20589,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-7660686/v1/623d5b17196e44b51eca9ae7.png"},{"id":92229417,"identity":"3178bc83-984d-4117-891c-fdaa6e64c43a","added_by":"auto","created_at":"2025-09-26 06:09:06","extension":"png","order_by":10,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":14460,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-7660686/v1/9aa16fd77de871260efb0c48.png"},{"id":92229421,"identity":"02e53040-caa9-431a-ad51-c1c893d2d265","added_by":"auto","created_at":"2025-09-26 06:09:06","extension":"xml","order_by":11,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":74846,"visible":true,"origin":"","legend":"","description":"","filename":"NCOMMS25745530structuring.xml","url":"https://assets-eu.researchsquare.com/files/rs-7660686/v1/ce01ac2bb1f5c225e7fa9fe9.xml"},{"id":92229420,"identity":"07f3e103-b67b-4ab9-ab32-4b50b26252a0","added_by":"auto","created_at":"2025-09-26 06:09:06","extension":"html","order_by":12,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":86142,"visible":true,"origin":"","legend":"","description":"","filename":"earlyproof.html","url":"https://assets-eu.researchsquare.com/files/rs-7660686/v1/450edcd1ba59e03a3140e8db.html"},{"id":92229408,"identity":"1914f08a-f5e4-40d8-912f-89a01c4000a9","added_by":"auto","created_at":"2025-09-26 06:09:06","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":86295,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eSample prompts for PLS and Discover RECOVER video script generation\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-7660686/v1/9741a67d6f9eef5359fff16c.png"},{"id":92229409,"identity":"a3e784bf-f5da-4960-a533-6cab46a56672","added_by":"auto","created_at":"2025-09-26 06:09:06","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":44739,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eFlesch-Kincaid Grade Level Range\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe Flesch-Kincaid Grade Level (0-18) assesses text readability, with lower scores indicating easier comprehension. This figure shows the reading level spectrum, from \"basic\" (0-5) to \"advanced\" (13-18), with the \"average\" range (6-12) considered ideal for Plain Language. RECOVER targets a 6th-grade reading level to maximize accessibility for diverse audiences, including patients, caregivers, and the public.\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-7660686/v1/18a30addfe3dfb3dce73aa93.png"},{"id":92684438,"identity":"7ad7ac12-9a21-4ca9-b88c-d202aea04512","added_by":"auto","created_at":"2025-10-03 02:28:21","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":946834,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7660686/v1/2ca3b72c-49d7-4af5-9cd7-b7b79f45d548.pdf"},{"id":92229407,"identity":"3bdc4ca2-7128-47f0-a8e9-fdb785581a29","added_by":"auto","created_at":"2025-09-26 06:09:06","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":28741,"visible":true,"origin":"","legend":"Supplementary Table 1","description":"","filename":"SupplementaryTable12.docx","url":"https://assets-eu.researchsquare.com/files/rs-7660686/v1/6832d826d5fc1310725f1baf.docx"}],"financialInterests":"\u003cb\u003eYes\u003c/b\u003e there is potential Competing Interest.\nNita Jain describes the relevant competing interest: advisory board member for the RECOVER Vital Clinical Trial.","formattedTitle":"Balancing Efficiency and Engagement: AI-Assisted Content for Research Communications in the RECOVER Initiative","fulltext":[{"header":"Introduction","content":"\u003cp\u003eCommunication is a cornerstone of any successful research initiative, and for health and science organizations, it is critical for establishing credibility and engaging the public. Effective scientific communication goes beyond simply reporting findings; it involves making complex information accessible, balanced, and respectful to a diverse audience.\u003csup\u003e1\u003c/sup\u003e Technical manuscripts are like a professional chef's intricate recipe\u0026mdash;full of jargon and complex techniques that are inaccessible to most. Effective communication is the skilled cookbook author who translates that recipe into simple, step-by-step instructions so anyone can prepare and enjoy the result. The rapid advancement of Artificial Intelligence (AI) has revolutionized this field, offering new ways to streamline and enhance communication and media promotion efforts.\u003csup\u003e2\u003c/sup\u003e\u003c/p\u003e\u003cp\u003eA recent survey by the Institute for Public Relations found that 45% of organizations now use AI for various tasks, including content creation, ideation, and market research.\u003csup\u003e3\u003c/sup\u003e For communication professionals, AI has become a powerful tool to synthesize vast amounts of information, identify key themes, and inform strategic messaging.\u003csup\u003e4\u003c/sup\u003e This paper explores how the Researching COVID to Enhance Recovery (RECOVER) program, funded by the National Institutes of Health (NIH), has strategically integrated AI into its communication workflows. We focus on its use in drafting plain-language summaries and video scripts, which are essential for broad media promotion and public engagement.\u003c/p\u003e\n\u003ch3\u003eRECOVER Communications Teams using AI\u003c/h3\u003e\n\u003cp\u003eNYU's Clinical Science Core (CSC) for NIH RECOVER serves as the scientific leadership and operational backbone of the RECOVER Initiative, set up to better understand post-acute sequelae of SARS-CoV-2 infection, known as Long COVID. CSC has led all aspects of developing and conducting studies, including the dissemination of results for the RECOVER cohort studies at \u0026gt;\u0026thinsp;200 enrolling sites located in all 50 states and Puerto Rico. Since their inception in 2021, the RECOVER cohort studies enrolled a diverse population of nearly 30,000 adult and pediatric participants, collected\u0026thinsp;\u0026gt;\u0026thinsp;50,000 biospecimens, and have over 140 reports published or under review.\u003c/p\u003e\u003cp\u003eTo ensure this research reaches the widest possible audience, the CSC employs three distinct communication teams: Community Engagement, External Affairs, and internal Communications. These teams work in tandem to transform complex scientific findings into accessible formats for various media channels. A core principle of RECOVER is to ensure patient and community perspectives are integrated throughout the research and reflected in all communications.\u003c/p\u003e\u003cp\u003eHistorically, scientific publications are written by and for the scientific community, making them difficult for non-specialists to understand. This creates a significant barrier to the public accessing potentially life-changing research.\u003csup\u003e5\u003c/sup\u003e To bridge this gap, RECOVER's communication teams have embraced Plain Language Summaries (PLS) and videos as crucial tools. A PLS translates complex research into easy-to-understand language, a process that is now being optimized with AI.\u003csup\u003e6\u003c/sup\u003e\u003c/p\u003e\u003cp\u003eThe team leverages \u003cb\u003eNYUChatGPT\u003c/b\u003e, an AI tool developed by New York University (NYU) that leverages OpenAI's GPT-3 technology to assist with a variety of tasks, including generating text-based content to help the NYU community.\u003csup\u003e7\u003c/sup\u003e A finalized, peer-reviewed manuscript serves as the input for NYUChatGPT and is guided by a prompt. This carefully crafted prompt, aligned with our established plain language standards guide, is provided to NYUChatGPT (See Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e). As a result, a plain language summary of the manuscript is returned. This use of AI allows the team to accelerate the creation of accessible content, thereby speeding up media promotion efforts and the overall dissemination of critical health information.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003eThis prompt instructs NYUChatGPT to generate a PLS that adheres to specific readability levels (6th grade), avoids jargon, and effectively communicates the key findings and implications of the research to a non-scientific audience. This NYUChatGPT-produced initial draft serves as a foundation for the team to further refine and polish, ensuring accuracy, clarity, and accessibility for the target audience. The final draft is sent to the manuscript\u0026rsquo;s authors to ensure accuracy prior to publishing on the RECOVER website.\u003c/p\u003e\u003cp\u003eDigital media allows for user-friendly presentations of research findings: it puts evidence into practice, allowing health professionals and the public to better understand key research findings.\u003csup\u003e8\u003c/sup\u003e Images and video can also be integral aspects of data collection, analysis, and reporting research studies.\u003csup\u003e9\u003c/sup\u003e Digital media content has become a powerful platform for public engagement with science. By offering PLS in video format and incorporating infographics, we can more widely disseminate our findings. The Discover RECOVER series, launched by the Community Engagement (CE) team in collaboration with researchers and Patient, Caregiver, and Community Representatives, communicates manuscript findings in brief videos. The CE team utilizes NYUChatGPT in creating plain-language scripts for Discover RECOVER. AI is prompted to synthesize and summarize the key points of an uploaded manuscript in simple language. The CE team used the key points to draft the script in English. The script is then reviewed by the authors and the Patient, Caregiver, and Community Representatives, ensuring that the AI-generated content accurately reflects the key findings of the manuscript.\u003c/p\u003e\u003cp\u003eIn alignment with RECOVER\u0026rsquo;s commitment to equal access in research, Discover RECOVER video scripts and infographics are translated into Spanish for improved access to findings. The CE team utilizes NYUChatGPT to draft the first version of Spanish subtitles and translated infographics for the videos. The CE team inputs the final English version of the Discover RECOVER script into NYUChatGPT and instructs the AI to generate a plain language script at a 6th grade reading level with active voice utilizing a Spanish Translations Plain Language Living Word Bank. This Spanish Translation Word Bank is a project managed by the CSC and Administrative Coordinating Center (ACC) Communications team that includes standard translation preferences for RECOVER deliverables. The AI-generated script is then shared with CSC Communications for quality assurance. The Communications team reviews the script to ensure the translation is in accordance with the preferred terms for RECOVER. This is an important step in the process as generative AI often makes translation errors, being unable to capture the complexity of healthcare information.\u003c/p\u003e"},{"header":"Methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e\u003cp\u003eTo evaluate the effect of generative-AI enhanced content on web traffic and engagement for the RECOVER website, recovercovid.org, we used Google Analytics 4 (GA4) to compare user interactions from two distinct time periods: December 1, 2023 to September 10, 2024, when all content was human-generated, and September 10, 2024 to January 20, 2025, when NYUChatGPT was deployed to assist with all research summaries, written and video, in English and Spanish. GA4 was configured to collect data on the following metrics:\u003c/p\u003e\u003cp\u003e\u003col\u003e\u003cspan\u003e\u003cli\u003e\u003cp\u003eViews. The number of mobile app screens or web pages that users saw. Repeated views of a single screen or page are counted.\u003c/p\u003e\u003c/li\u003e\u003c/span\u003e\u003cspan\u003e\u003cli\u003e\u003cp\u003eActive Users. The number of distinct users who visited the recovercovid.org website.\u003c/p\u003e\u003c/li\u003e\u003c/span\u003e\u003cspan\u003e\u003cli\u003e\u003cp\u003eAverage engagement time per active user. The average time that recovercovid.org was in focus in an active user\u0026rsquo;s browser or app, expressed in seconds.\u003c/p\u003e\u003c/li\u003e\u003c/span\u003e\u003c/ol\u003e\u003c/p\u003e\u003cp\u003eWe examined traffic and engagement metrics on:\u003c/p\u003e\u003cp\u003e\u003col\u003e\u003cspan\u003e\u003cli\u003e\u003cp\u003eThe \u0026ldquo;Research Summaries\u0026rdquo; page, which contains the PLS of various RECOVER studies and hyperlinks to the externally published full studies.\u003c/p\u003e\u003c/li\u003e\u003c/span\u003e\u003cspan\u003e\u003cli\u003e\u003cp\u003eThe \"Videos\u0026rdquo; page, which hosts \u0026ldquo;Discover RECOVER\u0026rdquo; and other video assets.\u003c/p\u003e\u003c/li\u003e\u003c/span\u003e\u003cspan\u003e\u003cli\u003e\u003cp\u003eThe \u0026ldquo;Publications\u0026rdquo; page, which hosts a separate page for every RECOVER publication, links to their plain language deliverables (summaries and/or videos), and to the externally published article.\u003c/p\u003e\u003c/li\u003e\u003c/span\u003e\u003c/ol\u003e\u003c/p\u003e\u003cp\u003ePaired t-tests were conducted to determine whether observed differences in website traffic and engagement between the two time periods were statistically significant.\u003c/p\u003e\u003cp\u003eTo assess the human oversight required to develop resources with the assistance of generative-AI, we performed a raw count of the number of changes made by human reviewers from the original AI-generated draft to the finalized resource. This assessment was performed on all AI-assisted video scripts and written PLS published between September 10, 2024 and January 20, 2025.\u003c/p\u003e\u003cp\u003eTo ensure readability of the AI-enhanced PLS and English-language video scripts, we employed the Flesch-Kincaid Readability Test. This established and widely used metric calculates a grade level equivalent for written text, effectively estimating the years of education required for comprehension.\u003csup\u003e10\u003c/sup\u003e Flesch-Kincaid Readability levels are a composite of the number of words, sentences, and syllables in a piece of text.\u003csup\u003e11\u003c/sup\u003e A lower Flesch-Kincaid grade level indicates greater readability and broader accessibility. Given the importance of clear communication in our PLS, we targeted a 6th-grade reading level. While NYUGPT was prompted to generate content at this target level, the initial AI-generated drafts exhibited a significantly higher average reading level of 12th grade. Therefore, the Flesch-Kincaid test served as an objective measure of the discrepancy between the intended and actual complexity of the AI-generated text. Specifically, we assessed the Flesch-Kincaid level for both the initial AI-generated draft and the final, human-revised version of each PLS and English-language video script. This two-stage assessment allowed us to quantify the impact of human editing on improving readability and achieving our target grade level, thereby ensuring the PLS and video scripts are accessible to our audience.\u003c/p\u003e\u003cp\u003eData collection was performed continuously throughout both time periods, and all data were anonymized and aggregated to maintain privacy and ensure compliance with data protection regulations.\u003c/p\u003e\u003c/div\u003e"},{"header":"Results","content":"\u003cp\u003eBetween the two time periods, the change in views and active users was not statistically significant. However, there was an average increase in the engagement time of 4.37 seconds (95% CI 0.076s \u0026ndash; 8.67s)(\u003cem\u003eP\u0026thinsp;=\u003c/em\u003e\u0026thinsp;.0461) per active user.\u003c/p\u003e\u003cp\u003eBetween the two time periods, there was a general decrease in views and active users on the RECOVER homepage, publications page, and the research summaries page (see Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e). However, views and active users on the videos page continued to increase between the two time periods, 25% increase in views and 15% increase in active users, suggesting the video content continued to drive interest even as general interest declined.\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eChange in views, active users, and average engagement time\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"10\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c8\" colnum=\"8\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c9\" colnum=\"9\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c10\" colnum=\"10\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/th\u003e\u003cth align=\"left\" colspan=\"3\" nameend=\"c4\" namest=\"c2\"\u003e\u003cp\u003eViews\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colspan=\"3\" nameend=\"c7\" namest=\"c5\"\u003e\u003cp\u003eActive Users\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colspan=\"3\" nameend=\"c10\" namest=\"c8\"\u003e\u003cp\u003eAverage engagement time per active user (seconds)\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003ePage\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eP1\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eP2\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003e% Change\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\u003e\u003cp\u003eP1\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c6\"\u003e\u003cp\u003eP2\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c7\"\u003e\u003cp\u003e% Change\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c8\"\u003e\u003cp\u003eP1\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c9\"\u003e\u003cp\u003eP2\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c10\"\u003e\u003cp\u003e% Change*\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eHomepage\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e87883\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e31855\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e-63.753\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e56104\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e20761\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e-62.996\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e\u003cp\u003e29.304\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e\u003cp\u003e22.991\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e\u003cp\u003e-21.543\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003ePublications\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e17939\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e8397\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e-53.191\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e5179\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e2604\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e-49.720\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e\u003cp\u003e78.215\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e\u003cp\u003e57.871\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e\u003cp\u003e-26.010\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eResearch Summaries\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e7891\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e3317\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e-57.965\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e3220\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e1210\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e-62.422\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e\u003cp\u003e74.428\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e\u003cp\u003e83.799\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e\u003cp\u003e12.591\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eVideos\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e536\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e673\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e25.560\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e358\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e410\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e14.525\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e\u003cp\u003e35.824\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e\u003cp\u003e42.510\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e\u003cp\u003e18.663\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003ctfoot\u003e\u003ctr\u003e\u003ctd colspan=\"10\"\u003eP1\u0026thinsp;=\u0026thinsp;The time period from December 1, 2023 to September 10, 2024\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd colspan=\"10\"\u003eP2\u0026thinsp;=\u0026thinsp;The time period from September 10, 2024 to January 20, 2025\u003c/td\u003e\u003c/tr\u003e\u003c/tfoot\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003e*Statistically significant, \u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.046.\u003c/p\u003e\u003cp\u003eAll AI-enhanced deliverables required extensive human revision to ensure that they were clear, concise, accurate, and in plain language. We performed a count of the number of material changes from the initial AI-generated draft to the final product. The mean number of material changes for all AI-enhanced deliverables (n\u0026thinsp;=\u0026thinsp;17) was 19.875 (SD\u0026thinsp;=\u0026thinsp;15.046, 95% CI [11.858, 27.892]). We analyzed the Flesch-Kincaid Readability (FKR) grade level, ranging from 0 to 18 (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e), for the initial AI-generated draft of the content and the final, human-reviewed product. For all AI-enhanced deliverables, the mean FKR level of the initial AI-draft was 12.276 (SD\u0026thinsp;=\u0026thinsp;1.810, 95% CI [11.346, 13.207]) while the mean FKR level for the final products was 11.56 (SD\u0026thinsp;=\u0026thinsp;2.129, 95% CI [10.464, 12.654]). While both the original and final products still carry a mean FKR level equivalent to a 12th grade reading level, we observed a 6.02% decrease in the FKR level after human review across all AI-enhanced products (Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e). This decrease in readability level represents a simplification of the text after human-review.\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eHuman-reviewer changes and Flesch-Kincaid Grade Levels for all AI-enhanced deliverables\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"5\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003e\u003cem\u003e# of Changes\u003c/em\u003e\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003e\u003cem\u003eFinal product\u003c/em\u003e\u003c/p\u003e\u003cp\u003e\u003cem\u003eword count\u003c/em\u003e\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003e\u003cem\u003eAI-Draft\u003c/em\u003e\u003c/p\u003e\u003cp\u003e\u003cem\u003eGrade Level*\u003c/em\u003e\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\u003e\u003cp\u003e\u003cem\u003eFinal Draft\u003c/em\u003e\u003c/p\u003e\u003cp\u003e\u003cem\u003eGrade Level*\u003c/em\u003e\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e\u003cp\u003e\u003cb\u003eAll AI-enhanced deliverables (n\u0026thinsp;=\u0026thinsp;17)\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eMean\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e19.875\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e283.176\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e12.276\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e11.559\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eSD\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e15.046\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e164.101\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e1.810\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e2.129\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e95% CI\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e[11.858, 27.892]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e[198.803, 367.549]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e[11.346, 13.207]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e[10.464, 12.654]\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e\u003cp\u003e\u003cb\u003eVideos (n\u0026thinsp;=\u0026thinsp;5)\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eMean\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e39.600\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e502.600\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e14.040\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e13.740\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eSD\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e7.987\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e139.581\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.789\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e1.141\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e95% CI\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e[29.682, 49.518]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e[329.288, 675.912]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e[13.06, 15.02]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e[12.323, 15.157]\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e\u003cp\u003e\u003cb\u003ePLS (n\u0026thinsp;=\u0026thinsp;12)\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eMean\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e10.909\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e191.750\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e11.542\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e10.650\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eSD\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e5.576\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e32.886\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e1.592\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e1.749\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e95% CI\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e[7.163, 14.655]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e[170.855, 212.645]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e[10.53, 12.553]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e[9.539, 11.761]\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003e*Grade Level\u0026thinsp;=\u0026thinsp;Flesch-Kincaid Readability Grade Level.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003eFigure \u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e: The Flesch-Kincaid Grade Level (0\u0026ndash;18) assesses text readability, with lower scores indicating easier comprehension. This figure shows the reading level spectrum, from \"basic\" (0\u0026ndash;5) to \"advanced\" (13\u0026ndash;18), with the \"average\" range (\u003cspan additionalcitationids=\"CR7 CR8 CR9 CR10 CR11\" citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e) considered ideal for Plain Language. RECOVER targets a 6th-grade reading level to maximize accessibility for diverse audiences, including patients, caregivers, and the public.\u003c/p\u003e\u003cp\u003eEach Publications page is devoted to a different scientific publication, and thus different scientific topics. We used the increase in average user engagement time for each publications page as a surrogate for public interest in various research topics. To determine topics of interest to the end user, we calculated the percent change in average active user engagement time for each publication\u0026rsquo;s webpage between the two time periods and then evaluated the subject matter of those pages that showed an increase in average engagement time. Based on this analysis, we were able to identify 10 general research topics that continue to garner user engagement (see Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e). The complete list of topics that saw increased engagement time between the two time periods can be found in S_1: Research topics and subtopics that continue to see increased average user engagement time.\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab3\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eResearch topics that continue to see increased average user engagement time\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"2\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eTopic\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003e# of webpages dedicated to this topic with increased average user engagement time\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eCardiopulmonary issues\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e18\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eNeuropsychiatric issues\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e17\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003ePASC Characteristics\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e11\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eInfectious Disease issues\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e9\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eInvestigating techniques\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e8\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eMiscellaneous\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e7\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eSleep/Fatigue issues\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e7\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eGastrointestinal issues\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e5\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eInvestigating Lab measures\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e4\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eENT issues\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e3\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eOB/GYN issues\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e2\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eGenitourinary issues\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e1\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eIn RECOVER, AI is leveraged to hasten distillation of research for PLS and video presentations. By automating these initial drafts, our teams save nearly 4 hours per project, significantly reducing the typical 10\u0026thinsp;+\u0026thinsp;hour process. With an anticipated 25\u0026ndash;30 manuscripts annually, this translates to a 200\u0026ndash;240 hour reduction in workload for PLS development and video scripts. This time saving allows communication teams to dedicate more resources to other projects, e.g. media promotion. This faster turn-around improves web user engagement through timely dissemination of information.\u003c/p\u003e\u003cp\u003eOur analysis revealed the crucial role of human review and quality control in developing effective scientific and health communications. By prioritizing readability, human review can enhance accessibility, facilitating a broader understanding of scientific and health-related information. This improved comprehension is essential for informed decision-making regarding personal health and engagement with scientific advancements. A focus on readability is not merely a stylistic preference, but a critical component of responsible scientific communication and health education.\u003csup\u003e12\u003c/sup\u003e Our results indicate that AI-generated output often requires substantial editing to meet desired readability targets. Nine out of the twelve AI-generated PLS had a noticeably higher Flesch-Kincaid grade level than the final, human-revised versions. This discrepancy underscores the current limitations of AI in independently producing summaries that consistently adhere to plain language principles. However, with appropriate human oversight, AI tools enable speedier dissemination, improved readability, and greater team efficiency.\u003c/p\u003e\u003cp\u003eBy implementing a strategic approach to AI utilization, we aimed to enhance the quality and accessibility of our research summaries and video scripts. This refined approach allowed us to produce summaries and video scripts that were not only accurate but also more engaging and understandable to a broader audience. Subsequently increasing average engagement time, validating the effectiveness of these AI-assisted strategies. Optimizing for readability and relevance, we have successfully captured and maintained public interest in key research topics. This data-driven approach empowers our teams to make informed decisions about future content development, ensuring that our deliverables continue to meet the evolving needs and interests of our audience.\u003c/p\u003e\u003cp\u003eLooking to the future, we would further streamline the communication pipeline between investigators and the public. Information directly from investigators can be challenging to translate, leading to delays and potential misunderstandings. AI has the potential to not only reduce wait times in this process but also expedite the creation of understandable content. Our goal is to achieve consistently simple PLS and video scripts that are highly accessible in multiple languages, ensuring a wider reach.\u003c/p\u003e\n\u003ch3\u003eLimitations\u003c/h3\u003e\n\u003cp\u003eThis study, while demonstrating the potential of AI in generating PLS for scientific research, also highlights important limitations that warrant consideration. A primary limitation is the reliance on a relatively small sample size. Only twelve PLS and five videos (English and Spanish) were generated with AI assistance, limiting the generalizability of our findings. Future research with a larger and more diverse dataset of scientific articles, engagement data, and target reading levels is necessary to validate these initial observations.\u003c/p\u003e\u003cp\u003eWhile NYUChatGPT shows promise for content generation and translation of technical documents into accessible language, its capabilities may not fully capture the nuance required to communicate complex chronic diseases, particularly as it relates to patients\u0026rsquo; lived experiences. Furthermore, it may generate responses with hallucinated research material.\u003csup\u003e13\u003c/sup\u003e The risk of misrepresenting research findings or presenting them in a distorted fashion trivializing the experiences of marginalized groups remains a significant concern. AI systems may perpetuate erroneous information\u0026mdash; amplifying demographic health disparities or supporting since-discredited psychosomatic models of disease, thus preventing patients from getting the help they need.\u003csup\u003e14\u003c/sup\u003e These concerns are especially relevant for communication aimed at informing decision-making, particularly regarding health outcomes.\u003c/p\u003e\u003cp\u003eSimilarly, our framework also has methodological limitations in measuring impact. While viewership metrics can capture surface-level engagements, tracking changes at the level of policy decisions or clinical practice remains significantly more challenging. Our current system effectively measures engagement metrics, such as views, active users, and engagements over time. However, it doesn't directly measure whether those activities lead to real-world impact, such as improved health or effective policy decisions. As a result, we may overemphasize quantitative metrics at the expense of qualitative real-world impact.\u003c/p\u003e\u003cp\u003eEngagement and web traffic metrics alone cannot provide a complete picture of the effect of AI-generated content on communicating findings. These metrics cannot account for confounding factors such as the general interest in science and Long COVID, the limitations in the granularity of the available data, and the inherent limitations in using Google Analytics data. During our study period, views on the RECOVER homepage dropped 64%, which could indicate a general decreased interest in Long COVID. However, these metrics cannot explain why this decrease occurred. They cannot account for the changing socio-political climate that may impact user engagement. Similarly, we are limited by the granularity of the data available. Our data does not indicate whether a Publication\u0026rsquo;s summary was interacted with, only that the page was accessed. It is possible that not all users were exposed to AI-generated content when they viewed each page. Our analysis is also subject to the limitations inherent in using Google Analytics data. For example, a user who visits the RECOVER website, clears their browser cookies, and returns to the site, is considered a new user, potentially overestimating the number of unique users. Conversely, if two users visit the RECOVER website on the same device (e.g. a shared computer), they are recorded as a single user, potentially underestimating the number of unique users.\u003c/p\u003e\u003cp\u003eOur work provides empirical evidence of the current capabilities and limitations of AI in a specific context \u0026ndash; generating PLS, video scripts, and synthesizing evidence for communications. While AI holds promise, it is not yet a perfect solution and requires careful oversight. The potential for biases in AI algorithms calls for greater transparency in AI-generated content requiring further investigation.\u003csup\u003e15\u003c/sup\u003e Our findings demonstrate that human input remains essential to refine AI-generated text, ensuring clarity, conciseness, and appropriate readability levels for target audiences. This necessity for human intervention has implications for the efficiency gains promised by AI tools and highlights the importance of incorporating human review into any AI-assisted workflow.\u003c/p\u003e\u003cp\u003eJust as translating science is crucial for informing research, it's equally relevant to ensure the research reaches the people it affects. Communications teams can leverage AI to synthesize information from diverse sources to develop informed and accurate messaging for various audiences.\u003csup\u003e16\u003c/sup\u003e Members of these teams are expected to make highly technical research accessible. The use of AI posits itself as a useful tool for making research digestible, however, the quality and reliability of AI-generated summaries depend heavily on the data it's trained on and the algorithms used.\u003csup\u003e17\u003c/sup\u003e Once again, human oversight remains critical to ensure accuracy, avoid misinterpretations, and address potential biases that could inadvertently shape communication. The current discourse surrounding AI development reflects a tension between rapid advancement and cautious implementation.\u003csup\u003e18\u003c/sup\u003e\u003c/p\u003e\u003cp\u003eAs institutions seek to integrate AI into established practices like scientific and strategic communications, future research should explore the ethical dimensions to ensure responsible and beneficial use of AI in disseminating scientific knowledge and shaping public discourse.\u003csup\u003e19\u003c/sup\u003e Finally, this study used the Flesch-Kincaid Grade Level as a metric of readability.\u003csup\u003e20\u003c/sup\u003e Future work should explore other readability metrics and qualitative assessments of PLS and video scripts to provide a more comprehensive evaluation of AI-generated content.\u003csup\u003e21\u003c/sup\u003e\u003c/p\u003e"},{"header":"Conclusion","content":"\u003cp\u003eIntegrating AI into scientific communication for the NIH RECOVER program offers opportunities for efficiency in content creation, as demonstrated by NYUChatGPT's effectiveness in generating PLS without reducing engagement. However, our findings underscore the necessity of human oversight to ensure accuracy, clarity, and accessibility, as AI-generated content often requires revision. While AI modestly increased user engagement, it struggles with the nuances of scientific research and patient experiences, highlighting the risk of misrepresentation in sensitive areas like health disparities. Despite AI's potential to synthesize and translate information, it carries risks of bias and misinterpretation. Responsible AI implementation in scientific communication demands careful human review, ongoing refinement of AI tools, and a focus on both accuracy and audience sensitivity.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eSources of Funding:\u0026nbsp;\u003c/strong\u003eNational Institutes of Health (NIH) Other Transactional Authority Agreements OT2HL161847. https://www.nih.gov/\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eDisclaimer:\u0026nbsp;\u003c/strong\u003eThe content is solely the responsibility of the authors and does not necessarily represent the official views of the RECOVER Program, the NIH or other funders\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConflict of Interest:\u003c/strong\u003e NJ reports participation on the Advisory Board for the RECOVER Vital Clinical Trial. All other authors report no conflicts of interest.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eReporting Guideline:\u003c/strong\u003e SQUIRE\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eResearch Question:\u003c/strong\u003e What is the effect of deploying AI-enhanced summaries of research findings on user engagement with the NIH-funded Researching COVID to Enhance Recovery (RECOVER) program\u0026rsquo;s website?\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis research was funded by National Institutes of Health (NIH) Agreement OTA OT2HL161847 as part of the Researching COVID to Enhance Recovery (RECOVER) research program.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eDisclosures\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAuthorship has been determined according to ICMJE recommendations.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData Availability Statement\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe data that supports the findings of this study are available in the supplementary material of this article.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAcknowled\u003c/strong\u003e\u003cstrong\u003egements\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWe wish to thank the National Community Engagement Group (NCEG), all Patient, Caregiver and Community Representatives, and all the participants enrolled in the RECOVER Initiative, with special thanks to Juan Lewis for their support.\u0026nbsp;\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eJucan M, Jucan C (2014) The power of science communication. Procedia - Soc Behav Sci 149:288\u0026ndash;292. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.sbspro.2014.08.288\u003c/span\u003e\u003cspan address=\"10.1016/j.sbspro.2014.08.288\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003ePanda G, Upadhyay AK, Khandelwal K (2019) Artificial intelligence: A strategic disruption in public relations. J Creat Commun 14(3):196\u0026ndash;213. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1177/0973258619866585\u003c/span\u003e\u003cspan address=\"10.1177/0973258619866585\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eDavis-Gilbert OK, Fajardo O (2024) The impact of AI on the bottom line in PR. \u003cem\u003eInstitute for Public Relations\u003c/em\u003e. January 29, Available from: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.instituteforpr.org/the-impact-of-ai-on-the-bottom-line-in-pr/\u003c/span\u003e\u003cspan address=\"https://www.instituteforpr.org/the-impact-of-ai-on-the-bottom-line-in-pr/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBiswal SK (2020) The space of artificial intelligence in public relations: The way forward. In: Kulkarni A, Satapathy S (eds) Optimization in machine learning and applications: Algorithms for intelligent systems. Springer, Singapore, pp 169\u0026ndash;176. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/978-981-15-0994-0_11\u003c/span\u003e\u003cspan address=\"10.1007/978-981-15-0994-0_11\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBarreto JOM, de Melo RC, da Silva LALB, de Ara\u0026uacute;jo BC, de Freitas Oliveira C, Toma TS, de Bortoli MC, Demaio PN, Kuchenm\u0026uuml;ller T (2024) Research evidence communication for policymakers: A rapid scoping review on frameworks, guidance and tools, and barriers and facilitators. Health Res Policy Syst 22(1):99. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1186/s12961-024-01169-9\u003c/span\u003e\u003cspan address=\"10.1186/s12961-024-01169-9\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003ePMID: 39118156; PMCID: PMC11312384\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eDormer L, Schindler T, Williams LA et al (2022) A practical \u0026lsquo;how-to\u0026rsquo; guide to plain language summaries (PLS) of peer-reviewed scientific publications: Results of a multi-stakeholder initiative utilizing co-creation methodology. Res Involv Engagem 8:23. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1186/s40900-022-00358-6\u003c/span\u003e\u003cspan address=\"10.1186/s40900-022-00358-6\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eNew York University IT, Generative AI services [Internet]. New York: New York University; [cited 2024 Jan 4]. Available from: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.nyu.edu/life/information-technology/artificial-intelligence-at-nyu/generative-ai-services.html#private\u003c/span\u003e\u003cspan address=\"https://www.nyu.edu/life/information-technology/artificial-intelligence-at-nyu/generative-ai-services.html#private\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBuljan I, Malički M, Wager E, Puljak L, Hren D, Kellie F, West H, Alfirević Ž, Marušić A (2018) No difference in knowledge obtained from infographic or plain language summary of a Cochrane systematic review: Three randomized controlled trials. \u003cem\u003eJ Clin Epidemiol\u003c/em\u003e. ;97:86\u0026ndash;94. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.jclinepi.2017.12.003\u003c/span\u003e\u003cspan address=\"10.1016/j.jclinepi.2017.12.003\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. PMID: 29269021\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWalker EB, Boyer DM (2018) Research as storytelling: the use of video for mixed methods research. Video J Educ Pedag 3:8. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1186/s40990-018-0020-4\u003c/span\u003e\u003cspan address=\"10.1186/s40990-018-0020-4\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eFlesch R (1948) A new readability yardstick. \u003cem\u003eJ Appl Psychol.\u003c/em\u003e ;32(3):221\u0026thinsp;\u0026ndash;\u0026thinsp;33. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1037/h0057532\u003c/span\u003e\u003cspan address=\"10.1037/h0057532\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. PMID: 18867058\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eFlesch R (2016) How to write plain English [Internet]. University of Canterbury; Jul 12 [cited 2025 Feb 12]. Available from: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://web.archive.org/web/20160712094308/http://www.mang.canterbury.ac.nz/writing_guide/writing/flesch.shtml\u003c/span\u003e\u003cspan address=\"https://web.archive.org/web/20160712094308/http://www.mang.canterbury.ac.nz/writing_guide/writing/flesch.shtml\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eRedish J (2000) Readability formulas have even more limitations than Klare discusses. ACM J Comput Doc 24(3):132\u0026ndash;137. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1145/344599.344637\u003c/span\u003e\u003cspan address=\"10.1145/344599.344637\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"},{"header":"Footnotes","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003e Jucan M, Jucan C. The power of science communication. \u003cem\u003eProcedia - Soc Behav Sci\u003c/em\u003e. 2014;149:288\u0026thinsp;\u0026minus;\u0026thinsp;92. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.sbspro.2014.08.288\u003c/span\u003e\u003cspan address=\"10.1016/j.sbspro.2014.08.288\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003e Panda G, Upadhyay AK, Khandelwal K. Artificial intelligence: A strategic disruption in public relations. \u003cem\u003eJ Creat Commun\u003c/em\u003e. 2019;14(3):196\u0026ndash;199. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1177/0973258619866585\u003c/span\u003e\u003cspan address=\"10.1177/0973258619866585\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003e Davis-Gilbert OK, Fajardo O. The impact of AI on the bottom line in PR. \u003cem\u003eInstitute for Public Relations\u003c/em\u003e. January 29, 2024. Available from: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.instituteforpr.org/the-impact-of-ai-on-the-bottom-line-in-pr/\u003c/span\u003e\u003cspan address=\"https://www.instituteforpr.org/the-impact-of-ai-on-the-bottom-line-in-pr/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003e Biswal SK. The space of artificial intelligence in public relations: The way forward. In: Kulkarni A, Satapathy S, editors. \u003cem\u003eOptimization in machine learning and applications: Algorithms for intelligent systems\u003c/em\u003e. Singapore: Springer; 2020. p. 169\u0026ndash;172. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/978-981-15-0994-0_11\u003c/span\u003e\u003cspan address=\"10.1007/978-981-15-0994-0_11\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003e Dormer L, Schindler T, Williams LA, et al. A practical \u0026lsquo;How-To\u0026rsquo; Guide to plain language summaries (PLS) of peer-reviewed scientific publications: results of a multi-stakeholder initiative utilizing co-creation methodology. Res Involv Engagem. 2022;8:23. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1186/s40900-022-00358-6\u003c/span\u003e\u003cspan address=\"10.1186/s40900-022-00358-6\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003e Barreto JOM, de Melo RC, da Silva LALB, de Ara\u0026uacute;jo BC, de Freitas Oliveira C, Toma TS, de Bortoli MC, Demaio PN, Kuchenm\u0026uuml;ller T. Research evidence communication for policymakers: A rapid scoping review on frameworks, guidance and tools, and barriers and facilitators. \u003cem\u003eHealth Res Policy Syst\u003c/em\u003e. 2024 Aug 8;22(1):99. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1186/s12961-024-01169-9\u003c/span\u003e\u003cspan address=\"10.1186/s12961-024-01169-9\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. PMID: 39118156; PMCID: PMC11312384.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003e New York University IT. Generative AI services [Internet]. New York: New York University; [cited 2024 Jan 4]. Available from: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.nyu.edu/life/information-technology/artificial-intelligence-at-nyu/generative-ai-services.html#private\u003c/span\u003e\u003cspan address=\"https://www.nyu.edu/life/information-technology/artificial-intelligence-at-nyu/generative-ai-services.html#private\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003e Buljan I, Malički M, Wager E, Puljak L, Hren D, Kellie F, West H, Alfirević Ž, Marušić A. No difference in knowledge obtained from infographic or plain language summary of a Cochrane systematic review: Three randomized controlled trials. J Clin Epidemiol. 2018 May;97:86\u0026ndash;94. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.jclinepi.2017.12.003\u003c/span\u003e\u003cspan address=\"10.1016/j.jclinepi.2017.12.003\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. PMID: 29269021.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003e Walker EB, Boyer DM. Research as storytelling: the use of video for mixed methods research. Video J Educ Pedag. 2018;3:8. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1186/s40990-018-0020-4\u003c/span\u003e\u003cspan address=\"10.1186/s40990-018-0020-4\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003e Flesch R. A new readability yardstick. \u003cem\u003eJ Appl Psychol.\u003c/em\u003e 1948 Jun;32(3):221\u0026thinsp;\u0026minus;\u0026thinsp;33. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1037/h0057532\u003c/span\u003e\u003cspan address=\"10.1037/h0057532\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. PMID: 18867058.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003e Flesch R. How to Write Plain English [Internet]. University of Canterbury; 2016 Jul 12 [cited 2025 Feb 12]. Available from: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://web.archive.org/web/20160712094308/http://www.mang.canterbury.ac.nz/writing_guide/writing/flesch.shtml\u003c/span\u003e\u003cspan address=\"https://web.archive.org/web/20160712094308/http://www.mang.canterbury.ac.nz/writing_guide/writing/flesch.shtml\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003e Jucan M, Jucan C. The Power of Science Communication. Procedia - Soc Behav Sci. 2014;149:288\u0026ndash;292. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.sbspro.2014.08.288\u003c/span\u003e\u003cspan address=\"10.1016/j.sbspro.2014.08.288\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003e New York University IT. Generative AI services [Internet]. New York: New York University; [cited 2024 Jan 4]. Available from: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.nyu.edu/life/information-technology/artificial-intelligence-at-nyu/generative-ai-services.html#private\u003c/span\u003e\u003cspan address=\"https://www.nyu.edu/life/information-technology/artificial-intelligence-at-nyu/generative-ai-services.html#private\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003e Panda G, Upadhyay AK, Khandelwal K. Artificial intelligence: A strategic disruption in public relations. \u003cem\u003eJ Creat Commun\u003c/em\u003e. 2019;14(3):208\u0026ndash;210. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1177/0973258619866585\u003c/span\u003e\u003cspan address=\"10.1177/0973258619866585\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003e Panda G, Upadhyay AK, Khandelwal K. Artificial intelligence: A strategic disruption in public relations. \u003cem\u003eJ Creat Commun\u003c/em\u003e. 2019;14(3):208\u0026ndash;210. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1177/0973258619866585\u003c/span\u003e\u003cspan address=\"10.1177/0973258619866585\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003e Panda G, Upadhyay AK, Khandelwal K. Artificial intelligence: A strategic disruption in public relations. \u003cem\u003eJ Creat Commun\u003c/em\u003e. 2019;14(3):208\u0026ndash;210. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1177/0973258619866585\u003c/span\u003e\u003cspan address=\"10.1177/0973258619866585\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003e Dormer L, Schindler T, Williams LA, et al. A practical \u0026lsquo;how-to\u0026rsquo; guide to plain language summaries (PLS) of peer-reviewed scientific publications: Results of a multi-stakeholder initiative utilizing co-creation methodology. \u003cem\u003eRes Involv Engagem\u003c/em\u003e. 2022;8:23. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1186/s40900-022-00358-6\u003c/span\u003e\u003cspan address=\"10.1186/s40900-022-00358-6\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003e Davis-Gilbert OK, Fajardo O. The impact of AI on the bottom line in PR. \u003cem\u003eInstitute for Public Relations\u003c/em\u003e. January 29, 2024. Available from: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.instituteforpr.org/the-impact-of-ai-on-the-bottom-line-in-pr/\u003c/span\u003e\u003cspan address=\"https://www.instituteforpr.org/the-impact-of-ai-on-the-bottom-line-in-pr/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003e Dormer L, Schindler T, Williams LA, et al. A practical \u0026lsquo;How-To\u0026rsquo; Guide to plain language summaries (PLS) of peer-reviewed scientific publications: results of a multi-stakeholder initiative utilizing co-creation methodology. Res Involv Engagem. 2022;8:23. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1186/s40900-022-00358-6\u003c/span\u003e\u003cspan address=\"10.1186/s40900-022-00358-6\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003e Flesch R. How to write plain English [Internet]. University of Canterbury; 2016 Jul 12 [cited 2025 Feb 12]. Available from:\u003c/span\u003e\u003cdiv id=\"Par89\" class=\"Para\"\u003e\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://web.archive.org/web/20160712094308/http://www.mang.canterbury.ac.nz/writing_guide/writing/flesch.shtml\u003c/span\u003e\u003cspan address=\"https://web.archive.org/web/20160712094308/http://www.mang.canterbury.ac.nz/writing_guide/writing/flesch.shtml\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/div\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003e Redish J. Readability formulas have even more limitations than Klare discusses. \u003cem\u003eACM J Comput Doc\u003c/em\u003e. 2000;24(3):132-5. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1145/344599.344637\u003c/span\u003e\u003cspan address=\"10.1145/344599.344637\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Artificial Intelligence, Plain Language Summaries, Lay Language Research, Health Communications, Science Communications, Community Engagement, Media Promotion, Research Promotion","lastPublishedDoi":"10.21203/rs.3.rs-7660686/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7660686/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003e\u003cstrong\u003eIntroduction\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe growing availability of AI tools is transforming health and science communication by streamlining content creation and promotion. This study investigates the impact of AI-assisted research summaries on user engagement with the NIH-funded RECOVER program's website and evaluates the efficiency and readability of the content.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eMethods\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWe analyzed Google Analytics 4 data from two distinct periods: one with entirely human-generated content and a second with AI-assisted content. We measured changes in page views, active users, and average engagement time, and assessed the review time and readability of the AI-enhanced summaries.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eResults\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThere was no significant change in page views or active users between the two periods. However, average engagement time increased by 4.37 seconds (P = .0461), suggesting AI-assisted content may be more compelling. Human review of AI-drafts averaged 19.88 changes, and readability improved, with the mean Flesch-Kincaid grade level decreasing from 12.28 to 11.56.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConclusion\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis study demonstrates that AI can be a valuable tool for accelerating the creation of accessible and engaging content. Our findings highlight a crucial balance: while AI can save effort and reduce cost in public engagement efforts, human oversight remains essential to ensure the accuracy, clarity, and accessibility of vital health communications.\u003c/p\u003e","manuscriptTitle":"Balancing Efficiency and Engagement: AI-Assisted Content for Research Communications in the RECOVER Initiative","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-09-26 06:09:00","doi":"10.21203/rs.3.rs-7660686/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"378a6a6c-b7f0-4c6f-a6d2-0f8ccb0964cd","owner":[],"postedDate":"September 26th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":55033075,"name":"Health sciences/Medical research/Translational research"},{"id":55033076,"name":"Scientific community and society/Scientific community/Publishing/Media"}],"tags":[],"updatedAt":"2025-10-03T02:20:14+00:00","versionOfRecord":[],"versionCreatedAt":"2025-09-26 06:09:00","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-7660686","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7660686","identity":"rs-7660686","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}