AI in Medical Writing: Lessons from the Field

“Snake Oil or Silver Bullet? Finding AI Tools You Can Trust,” an expert panel at the 2025 AMWA annual conference, shared practical strategies and cautionary tales that gave medical writers actionable guidance on using AI effectively. This lively conversation was moderated by Harriett Judson, Chief Operating Officer at PerfectIt, and featured panelists representing a range of industry perspectives: Lisa Chamberlain James, Senior Partner at Trilogy Writing & Consulting; Angela Winnier, Executive Director of Medical Writing at Pfizer; Katy Alderson, Head of Marketing at PerfectIt; and Philip Burridge, Director of Operations & Strategy at Morula Health.

The session opened with a candid acknowledgment of the many misconceptions surrounding AI, as well as the frequent disconnect between institutional leadership’s expectations of AI and the day-to-day experience of those who use AI. Still, the panelists expressed optimism about AI’s value in medical writing. Rather than being something to fear, they proposed that AI offers medical writers an opportunity to reimagine the way they do their work.

Four major themes emerged from the discussion, which used real-world examples to explore principles for the thoughtful integration of AI tools into medical writing.

Keep Humans at the Center

One major throughline was that human oversight is essential. The panelists emphasized that AI cannot replace human judgment or accountability. To facilitate this, guardrails should be put in place so that oversight isn’t sacrificed for speed. In addition, AI outputs must be transparent, traceable, and auditable. 

Angela illustrated this principle with an example from Pfizer. Although an AI-drafted clinical study report (CSR) appeared well-written to a medical writer who was new to the field, a clinician quickly recognized that conclusions about the significance of lab data were incorrect and unsupported by medical opinion or protocol definitions. Lisa’s experience using a publicly available AI tool to draft a thought piece reinforced the same point. The tool produced a draft that sounded like her but lacked nuance, and none of the 32 references provided were real. Lisa acknowledged, however, that even after making corrections, the AI tool ultimately saved her time.

In short, expert review is indispensable, and AI tools and their outputs must be transparent to support that review.

Build Strong Foundations: Quality Inputs Drive Quality Outputs

As organizations begin incorporating AI into medical writing, the first step is to establish the processes that enable AI to generate high-quality outputs. For AI to perform well, it requires high-quality, structured inputs.

To drive home this idea, Angela described an instance in which every efficacy endpoint for clinical remission was wrong in an AI-drafted CSR. The problem, it turned out, was that the statistical analysis plan (SAP) contained multiple definitions of clinical remission. In another example, Katy referenced a recent Grammar Girl post highlighting the importance of quality prompts. Two studies on AI completing real-world tasks produced sharply contrasting success rates. The key difference was the prompts. The study that used prompts written by experts achieved a 48% success rate, whereas the one that relied on prompts drawn from real-world project briefs achieved only a 2.5% success rate.

Several panelists were also hopeful that AI could serve as a catalyst for improvements in medical writing workflows. By demanding clearer inputs, standardized templates, and harmonized processes, AI integration can encourage organizations to address long-standing inefficiencies.

Adopt AI Incrementally and Strategically

As the panel moved on to how to adopt AI, it became clear that there is no one-size-fits-all solution. Rather than pursuing a sweeping AI revolution, the panelists advocated for a strategic, stepwise approach. They suggested breaking work into smaller parts, identifying friction points where AI can clearly add value, and then expanding from there. Low-risk, high-impact areas are a great place to start.

The panelists described several use cases where AI proved helpful to them, noting that AI excels with repetitive, structured tasks but fails at nuance:

• Supporting language fluency for multilingual teams

• Generating transcripts or summaries of kick-off meetings and comment resolution meetings

• Mining information for rapid familiarization with a new topic or therapeutic landscape (with appropriate follow-up)

• Drafting summary tables (eg, demographics tables)

Offloading these tasks can free writers and editors to focus on higher-level work. 

Ultimately, each individual or group must determine which use cases best fit their needs. Phil expanded on this point, noting that the opportunities and challenges faced by large companies versus freelancers or small companies will differ. While large companies have the resources to be early adopters, they risk putting large investments into AI tools that may eventually fail. In contrast, individuals and small companies can pilot new AI tools and leverage free options to support current projects, building the expertise needed to stay agile and competitive as advanced AI tools become more accessible.

Upskill to Lead Change

At first, there seemed to be some differences of opinion between panelists about whether writers need to be prompt engineers. The final consensus was that writers do not need to be experts in prompt engineering, but they do need to understand how inputs shape outputs and how AI workflows affect privacy and security. Writers and editors also need to be part of the conversation as organizations decide how they integrate AI into their businesses. By learning to use AI effectively and critically, writers and editors can position themselves as the subject matter experts who shape how AI is implemented across their organizations.

A question raised during the Q&A highlighted a future challenge: maintaining the expertise required to review AI-generated content. Training for medical writers and editors will need to focus not only on AI but also on developing and maintaining domain expertise.

Conclusion

By the end of the session, the takeaway was clear: AI isn’t a silver bullet or snake oil—it’s a tool. AI cannot replace medical writers, but it can be useful. With transparent systems, strong inputs, and a phased approach that maintains expert review, medical writers can harness AI to improve efficiencies, allowing them to focus on the expert, human-driven contributions that elevate their work.

About the author: Florence Roan is a scientist transitioning into medical writing, with experience spanning clinical practice, research, and regulatory oversight (IACUC & IRB). She earned her MD/PhD from Emory University and completed clinical training at Washington University in St. Louis (internal medicine) and the University of Washington (infectious diseases). She then completed a postdoctoral fellowship in immunology at the Benaroya Research Institute.