Insights · For the moments that define the job
How do patients use AI for health information, and what should communications teams do about it?
When do patients turn to AI?
Communications plans are built around the moments that matter to a business: a clinical readout, an approval, a launch. Patients keep a different calendar. A symptom they cannot explain. A diagnosis they do not understand. A treatment decision. A prescription they cannot afford. A side effect they are not sure they should report. Those are communications moments too.
And increasingly, people take them to AI. In a 2026 Pew Research Center survey, 25% of U.S. adults had used AI chatbots to investigate symptoms, 22% to understand a diagnosis and 22% to learn about treatments. People do not move through these moments in a neat sequence. They overlap, loop back and collide.
What question should communications leaders be asking?
Not only "Does AI mention our brand?" That is a marketing metric. The communications question is: when a patient or caregiver faces an inflection point, what is AI synthesizing, and is the clinical and practical reality intact?
What does each stage need from communications?
Each stage demands different source material, and each carries a different failure mode when AI summarizes it.
| Stage | What people ask AI | The communications job |
|---|---|---|
| Symptom discovery and triage | 15% of U.S. adults use chatbots to decide whether to see a doctor | Plain-language disease education, with clear red flags and limits on what an online explanation can establish |
| Diagnosis and results | 48% of Australian ChatGPT health users ask about a specific condition; 20% of U.S. adults use chatbots to decode lab results | Clear definitions, subtype explanations and biomarker context, including what a test result does not tell them |
| Treatment evaluation and shared decision-making | 46% of U.S. ChatGPT health users look for additional or alternative treatment options; 30% of Australian users ask about medicines, tests, side effects and interactions | Evidence and indicated populations clear enough to anchor a productive clinician conversation |
| Starting treatment, access and support | Prior authorization, copay assistance, out-of-pocket costs, administration | Someone on the team actively governs the accuracy and currency of public-facing access facts |
| Ongoing management and monitoring | 24% of Australian ChatGPT health users ask AI to help create a plan to improve their health | Point people toward support programs and clinical escalation routes |
At the treatment stage the stakes are highest. An efficacy benefit stripped of its trial population is incomplete; a list of adverse events without clinical incidence is paralyzing. When an answer engine drops those boundaries, it can create inadvertent off-label assumptions in public, a reputational and compliance risk. At the access stage, outdated details make even a fluent answer misleading, often defaulting to list price instead of what assistance makes real.
Can a correct AI answer still leave a gap?
Yes. In an October 10, 2026 audit for the first edition of Ally's Read, we tested how four AI systems (through developer APIs, with web search on) answered a question about ivonescimab ahead of its November 14, 2026 FDA goal action date for a specific lung cancer indication in combination with chemotherapy. All 75 responses stated its regulatory status accurately. By brand-level accuracy alone, the job was done. Then we added realistic stakeholder context:
- Caregiver asking about side effects. 11 of 13 answers detailed side effects; only 4 mentioned the HARMONi trial's progression-free survival benefit.
- Oncologist asking for clinical evidence. 11 of 12 answers cited the progression-free survival benefit; only 1 mentioned side effects.
- Across all four stakeholder contexts. Only 10 of 50 answers included both the benefit and the side effects.
- The bare question. Of 25 answers, 3 mentioned the clinical benefit and none mentioned side effects.
Because the caregiver prompt asked about side effects and the oncologist prompt asked for evidence, these descriptive tests do not prove that persona alone drove the difference, and they do not capture private patient or physician chats. The lesson holds anyway: testing a bare question in a vacuum hides how AI behaves when real people bring their priorities to it. Methodology and full results in Ally's Read.
How should a communications team act on this?
Do not make it a separate AI workstream that nobody owns. Bring the patient journey into your next planning meeting alongside your stakeholder and media strategy.
- Pick one consequential moment. A diagnosis inflection point, a treatment comparison or an access hurdle. Write down the questions a patient or caregiver reasonably asks, including the real-world constraints that change what they need to know.
- Audit the answers and the citations. Run those questions, with their context, across the major answer engines. What is missing? Is the clinical evidence current? Did the engine stretch a subgroup result across the broader population? Are eligibility and cost conditions explicit?
- Fix it where the signal breaks. The right fix is rarely optimizing for an algorithm. It might be a clearer explanation in a data release, an updated patient FAQ, a deeper briefing with a specialty reporter or a partnership with a patient advocacy group. Sometimes it reveals an evidence gap you cannot responsibly fill yet.
- Keep the meaning intact through review. Journalists, clinicians and advocacy organizations exercise independent judgment; their credibility is not a channel you control. The job is to bind the key message, citable source, indicated population and safety qualifications together so they are cited without distortion. Real-world test results give Communications an audit trail to support proactive FAQs and unbranded disease materials through MLR review.
- Re-test. Once new materials clear review, re-run comparable questions to see what shifted, without assuming one publication caused one AI answer.
How does CommsRx approach it?
CommsRx starts with the AI insight from GPS: what answer engines are telling people, and where the meaning breaks. The Meeting Kit, part of the CommsRx Green Thread, takes that insight into the room. It is built from the strategy the team has already confirmed and leads with the stakeholder's story before the logic. Medical, Corporate Affairs, Patient Advocacy and Market Access each see the problem through their own lens; the kit shows the whole path, and the team can put the hard question from each chair to a governed AI persona before anyone commits. Everything drafted in CommsRx goes through the team's own review.
Related questions
Sources
- Pew Research Center, August 25, 2026, "From diagnoses to treatments: Why Americans use AI chatbots for health" https://www.pewresearch.org/science/2026/08/25/from-diagnoses-to-treatments-why-americans-use-ai-chatbots-for-health/ (checked October 21, 2026)
- Ayre J, et al. "Use of ChatGPT to obtain health information in Australia, 2024: insights from a nationally representative survey." Medical Journal of Australia 2025; 222(4). https://www.mja.com.au/journal/2025/222/4/use-chatgpt-obtain-health-information-australia-2024-insights-nationally (checked October 21, 2026)
- Ayo-Ajibola O, et al. Journal of Medical Internet Research 2024; 26: e55138. https://www.jmir.org/2024/1/e55138 (checked October 21, 2026)
- CommsRx, Ally's Read, first edition, October 2026. https://commsrx.com/read/ (checked October 21, 2026). CommsRx is built by Signal Fidelity Group, founded by Abhi Basu.
About this page
Adapted from the October 21, 2026 edition of AI in PR, Abhi Basu's LinkedIn newsletter. General guidance, not medical, legal or regulatory advice. Drafted with AI assistance, then edited and checked by Abhi Basu, founder of CommsRx and Signal Fidelity Group. Last reviewed October 22, 2026. Corrections: help@signalfidelitygroup.com.