Medical writers' attitudes to AI and reported use
Medical writers differ in how much they know about AI writing tools, how far they use them in their work, and whether they see them as a help, a risk or a threat to the profession.
Use of AI writing tools
As AI writing assistants grew more popular, one healthcare communications consultant held in 2023 that more medical writers would inevitably start using them, and two surveys of medical writers already showed the trend. In March 2022, 17 of 87 medical writers (19.5%) had tried an AI writing assistant for their healthcare write-ups. In May 2023, of 76 medical writers, 52% used one occasionally and 17% regularly, and only 17% had never tried one, so that 69% used one at some level.
In September 2023, one member of EMWA's AI working group judged that a majority of medical writers had some understanding of AI-powered tools, most notably ChatGPT, but that only a minority used ChatGPT and related tools in their work. Other tools EMWA members mentioned were TriloDocs, Grammarly, Quillbot, Writefull and PerfectIt.
Knowledge, attitudes and practice concerning ChatGPT
A survey of medical writers' knowledge, attitudes and practice (KAP) concerning ChatGPT ran online, in English, from 14 June to 24 September 2023, distributed through LinkedIn and personal contacts. When the survey was conceived in 2023, ChatGPT was the AI tool most familiar to medical writers. In 2025 the medical writing community seemed divided on the practical use of ChatGPT: some writers recognised its potential to enhance their work by offering a strong foundational framework, while others hesitated over the inaccurate information and errors it may produce.
The survey's 106 respondents came from 21 countries, 61.3% of them from India and the others from Europe, Australia, Canada, the USA and other Asian countries. Of the respondents, 69 (65.1%) were women, 36 (34.0%) men and one preferred not to say. By age, 8 were under 25, 38 were 25–34, 38 were 35–44, 18 were 45–54, 2 were 55–64 and 2 were over 65. Their highest education was a bachelor's degree for 18, a postgraduate degree for 40 and a doctorate for 48. By sector, 9 worked in the pharmaceutical industry, 59 in medical communications, 11 in academic institutions, 21 in healthcare organisations and 6 elsewhere. By experience in medical writing, 16 had under one year, 50 one to five years, 20 six to ten years, 10 eleven to fifteen years and 10 over fifteen years. Nineteen held a professional certification in medical writing or editing, 30 belonged to a professional organisation, 36 had formal training in medical writing and 48 (45.3%) had used AI-powered tools for medical communications. Their specialisms ranged across medical education writing (48%), regulatory writing (14%), health economics and outcomes research (about 6%), medico-marketing (35%), science journalism (17%) and blog writing (about 2%). That less than half of the respondents had used AI tools for their writing suggested to the authors that the tools had yet to become the norm among medical writers, perhaps because of an inhibition or dilemma over their apparent limitations.
Overall, 82.1% of respondents said they had a general understanding of how ChatGPT works. Seven single-choice questions tested that knowledge, scoring one point for each correct answer, and placed 26.4% of respondents at low knowledge (a score of 0–2), 44.3% at average knowledge (3–4) and 29.2% at high knowledge (5 or more). The share answering correctly was 50.9% for what ChatGPT is, 55.7% for a task it cannot perform, 61.3% for what hallucination is, 27.4% for how it is being further trained, 73.6% for whether its content is subject to bias, 57.5% for its knowledge cut-off date and 60.4% for its responses to the same prompt.
Of the respondents, 40.6% agreed that ChatGPT can improve the quality of medical writing and 57.5% believed that it can save time. Strong agreement came from 38.7% on concerns about the accuracy and privacy of the data generated, from 54.7% on the need for guidelines to regulate the use of ChatGPT and other AI technologies, and from 49.1% on the need for formal training to use it effectively. On whether ChatGPT could replace human writers, 40.6% disagreed. Responses on its suitability for particular kinds of material, such as slide decks, patient brochures, manuscripts or books, varied widely, which the authors read as writers using ChatGPT differently for different types of writing. Most respondents named plain language summaries and blogs as the material it suits best, and the abstract or the introduction as the parts of a manuscript it can be particularly useful for drafting.
Asked how often they used ChatGPT, 21 respondents (19.8%) said never, 27 (25.5%) rarely, 42 (39.6%) sometimes and 16 (15.1%) always. They used it largely for writing summaries (24.4%), routine tasks such as composing emails or drafting letters (19.5%), organising scattered points into a coherent paragraph (18.3%) and understanding complex topics (15.9%), and also for social media posts and other uses. They did not seem to find it particularly easy or difficult to use: 35 (33.0%) were neutral on whether it was easy, against 3 who strongly disagreed, 16 who disagreed, 26 who agreed and 12 who strongly agreed, while 14 found the question not applicable. The quality of its content was rated poor by 6 (5.7%), fair by 45 (42.5%), good by 34 (32.1%) and excellent by 4 (3.8%), with 17 (16.0%) answering not applicable. Although the largest group used it only sometimes, 51 (48.1%) said they might recommend it to other medical writers, 40 (37.7%) that they would and 15 (14.2%) that they would not. Thirty (28.3%) thought it had improved their writing efficiency, 39 (36.8%) said maybe, 14 (13.2%) said no and 23 (21.7%) answered not applicable. Asked whether their organisation or institution allowed them to use ChatGPT, 33 (31.1%) said yes, 15 (14.2%) no, 36 (34.0%) that there were no rules or guidelines and 22 (20.8%) that they did not know. The challenges they met most were needing several prompts to get the desired response and inaccurate content, and they also reported server errors, busy servers, a lack of referencing and superficial responses.
Only statistically significant associations were reported. Respondents from medical communications were associated with high knowledge of ChatGPT, and those from healthcare organisations with lower knowledge than other sectors (p=0.001). Respondents with high or average knowledge tended to agree or strongly agree that ChatGPT saves time in writing (p<0.001). Those with high knowledge tended to agree strongly with concerns about accuracy and data privacy, while those with low knowledge were predominantly neutral (p=0.001), and those with high knowledge appeared to align more strongly with the need for guidelines on using ChatGPT and other AI tools (p<0.001). Respondents with a deeper understanding of the technology behind ChatGPT were more likely to use it (p<0.001) and to recommend it to others, while those with limited knowledge more often answered maybe or no on recommending it (p=0.016). The authors read these associations as knowledge of ChatGPT influencing its adoption and the attitudes and practices of medical writers.
The authors suggested that the constant demand on medical communications writers to stay up to date may explain that sector's high knowledge. Writers in healthcare, in their reading, may use AI tools less because of the challenges of legal ethics, patient privacy and the accuracy and reliability of information, and may prefer to be self-reliant to avoid errors and keep the integrity and trust of the healthcare profession. In the authors' reading, most respondents used ChatGPT largely for tasks that do not jeopardise data privacy and ethics, such as understanding complex topics, organising scattered points into a coherent paragraph, and drafting letters, emails and social media posts. The authors concluded that a correct perspective on ChatGPT and other recent AI tools rests on a good understanding of them, which is essential both to formulate and to follow guidelines on the use of large language models in medical writing. Such guidelines, they held, will help writers produce quality work and maintain publication ethics while minimising errors and overcoming limitations.
Judgements of ChatGPT for professional writing
For the uses that writers in continuing medical education were exploring in 2023, one writer judged that the time needed to find and corroborate the sources of what ChatGPT returns and to fact-check it will swallow any efficiency savings.
Another writer saw no benefit from the medical writer's perspective, because the tool's role is to do the writer's job of researching, reading and synthesising, and distilling what is learned into clear, concise and accurate language, so that it has the potential to turn writers into reviewers, which may not be to many writers' liking. Although its access to information is vastly deeper, broader and faster than a writer's, it could not filter information from misinformation or take responsibility for its content, and a reviewer who wished to question a sentence it wrote had nobody to ask. On this view, those who could benefit from it were, to a large degree, content mills, which already do not care about accuracy or accountability, and predatory publishers, which could benefit immediately from its speed and power. It could have a legitimate side as a medical writing tool when what it can source and do is limited: confined to the information in a specific set of resources and given a prescribed template in which to apply it, it was potentially well suited, though its output must still be reviewed, scrutinised and, where necessary, questioned, which may create more opportunities for writers and editors who love to fact-check.
A writer who had not yet started using ChatGPT on their own computer had an IT friend put more than 10 questions to it on medical and scientific, political, philosophical and religious, and practical topics. The writer counted among its benefits that it helps to amalgamate information on research questions, but in that test its answers held many factual errors and errors of narrowness, it was clear that it may give biased answers, it gave references only when specifically asked for them and some of those were incorrect, and it carried a risk or temptation to plagiarise a tool that has itself plagiarised others without attribution. The writer did get a helpful answer on growing a potted rosemary plant indoors, but held that serious questions should not be put to ChatGPT, at least at that time, that it should not then be used for serious professional medical writing, and that little can be done to mitigate the problems other than doing one's homework, maintaining one's integrity and ethics, and not letting it do the work that human critical thinking should do.
Another writer held that AI is here to stay despite the absence of regulation, and that ChatGPT can be used as an advanced search engine provided that everything it generates is fact-checked and its references critically evaluated. The arguments against ChatGPT, on this view, recall those once made against the reliability of Wikipedia, which persevered and flourished despite the initial scepticism, and ChatGPT will likewise continue to spread and evolve whether or not writers use it. AI may not yet replace medical writers, but writers who use it effectively could outperform their peers, and it should be seen as a tool that complements and enhances their capabilities rather than threatening the profession; adapting to it will let writers take the opportunities it provides while keeping their relevance.
Seasoned freelance medical writers at a 2023 conference session were still at odds with AI, though, in the session's account, it is beyond doubt here to stay and all medical communicators, not only freelances, must make peace with it. AI was being used to conduct literature searches, generate images and draft initial content to guide further literature research and writing, and it may be most useful for developing content on disease states. Whether freelance or employed, medical writers and editors need education and guidance on what to do and not to do when working with AI content generators. Among the content and image generators these freelances named, some free and some fee-based, were the free and subscription versions of ChatGPT, OpenEvidence for searching the medical literature, Claude for generating text and DALL·E for generating images. Their bottom line was that freelances must embrace AI rather than fear it and find its best use for their own and their clients' needs. The value proposition for freelances lies in the quality of the AI–human pairing.
Fears and expectations for the profession
In November 2018, at a forum of executives who recruit, train and manage medical communicators, one attendee noted that AI in medical writing was not yet upon them but predicted that it "will explode upon us unexpectedly". The challenge would be to incorporate AI tools into work processes and to prepare writing teams to use them, and communicating the tools' benefits would be essential. The security of AI was also a concern. Managers generally viewed AI positively, as able to remove the drudgery of creating certain components of regulatory documents while letting medical communicators focus on more engaging aspects of document creation, and AI was not intended to replace medical writers. Attendees cautioned that technology cannot be the driver, and that medical communicators should be empowered to guide the development of AI and automated systems.
In September 2023, one member of EMWA's AI working group named one of the medical writing community's major fears as whether AI tools would replace what medical writers were doing, and added that writers were sceptical about the evolving skills that would be required to stay employable and that certain routine tasks they had recently been doing might become redundant. In May 2024, one speaker reported having heard that much of the conversation about AI, perhaps at EMWA's conference the year before and in general, had started from fear and from not knowing how the tools would affect writers' jobs. In the speaker's view it had since evolved somewhat, as writers came to understand that anyone who had actually used the tools knew many of their downsides: they do not necessarily do what was wanted at the first call, and they create new issues, perhaps not met before, that need addressing.