September 16, 2026

How Should Biotechs Use AI for Investor Relations?

Artificial intelligence is quickly becoming the newest tool in the investor relations toolkit – but like any tool, its usefulness (or otherwise) depends entirely on how thoughtfully it is utilized.

AI should not replace professional judgment, disclosure discipline, or strategic messaging. Instead, it should help IR teams operate more efficiently while maintaining credibility with investors, analysts, and regulators.

The golden rule of using AI for IR

Unless you have paid a large sum of money for a private AI network (hint: you probably haven’t ) and are running AI inside your company in a secure environment, you should assume that all information you put into a public tool like ChatGPT immediately enters the public domain.

As such, you should not use AI for the generation of any materials (press releases, Q&A documents, corporate presentation content) that contain material nonpublic information (MNPI), as entering the data into the AI interface could amount to selective disclosure to a single party.

Some may push back and argue that it’s extremely unlikely that ChatGPT would use such information to trade a company’s stock, or to reshare it elsewhere. But the reality is, we have little to no idea about how AI providers use the information we input (as well as the answers they supply us with), and as such, prudence necessitates that we tread especially carefully.

Examples of IR functions AI could be used for:

• Perfunctory press releases: Such as those announcing a company’s participation in an investor conference or announcing an earnings date (containing no MNPI).
• Q&A practice: So long as you type in your prompt using generalities, AI can be great at helping draw up “potential questions,” since AI is very good at generating lists.
  • e.g. Prompt: “Come up with the 20 most likely questions investors will ask when a biotech announces negative Phase 2 data.”
  • e.g. Prompt: “Come up with the 20 most likely questions investors will ask when a biotech announces the departure of its CMO/CFO/CEO.”
• Competitive intelligence / peer news flow: Newswires, like GlobeNewswire and Business Wire, now actually tout the interconnectedness of their news distribution platforms with AI in the same way they used to boast about their relationships with search engines. Because of this, AI can be a highly efficient way to look for and collect data on your peers, competing drug development programs, etc.
• Investor meeting summaries: Zoom already has an embedded AI function that can provide either an entire transcript of a call, or else an AI-generated summary of what was discussed. This is a huge step forward productivity-wise, because previously someone had to manually take notes and then type them up afterwards. AI-generated meeting minutes still need to be vetted before they are saved and counted as a record of what was said, however.
• Summarize lengthy analyst reports (or compare the key differences between two reports): Since equity research is already in the public domain, it’s possible to upload PDFs of sequentially issued research reports and ask AI to ascertain the key differences. This is obviously no substitute for thoroughly reading them yourself but could be timesaving when looking at the research being published on, say, your peers.
• Summarize earnings calls: Whether your own or someone else’s, by uploading a public domain transcript of an earnings call into, say, ChatGPT, you can generate a summary of the call or ask the AI questions based on its content, e.g. “How many analysts asked questions?” or “What was the leading concern expressed?”

Beware hallucinations

It’s important to remember that every time you request data from an AI provider, you need to be wary of hallucinations.

AI models don’t “know” facts the way humans do. They predict likely words based on patterns in training data. Because of that, AI may:

• Invent sources, statistics, or quotes.
• Misremember or blend real facts together.
• Give outdated or wrong answers.
• Produce logical-sounding explanations that aren’t true.

As such, all work products generated by AI need to be vetted by a subject matter expert before they are shared.