The short version
- Patients ask AI first: a 2026 KFF poll (opens in a new tab) found 32% of US adults used AI chatbots for health information in the past year, and 41% of those users did so to look things up before deciding whether to see a provider.
- AI answers about drugs can be wrong in ways that matter: in a study in BMJ Quality & Safety summarized by AHRQ’s PSNet (opens in a new tab), Microsoft’s Copilot answered questions on 50 common drugs with 88.7% mean accuracy, and experts judged that 22% of a subset of answers could lead to death or severe harm if followed.
- Regulators are back: on September 9, 2025 the FDA announced (opens in a new tab) thousands of warning letters and about 100 cease-and-desist letters over drug ads, after warning letters fell to one in 2023 and zero in 2024.
- Marketing money is moving to digital: Fierce Pharma reports (opens in a new tab) eMarketer’s estimate of $24.8 billion in healthcare and pharma digital ad spending in 2025, against about $7.9 billion traditional.
- Doctors are harder to reach in person: Veeva’s 2024 field data, published via BioSpace (opens in a new tab), found 45% of health care professionals accessible to drug companies, down from 60% in 2022.
What does brand visibility mean for a prescription drug?
Being present and correctly described when people ask about a condition or a brand, not being recommended.
For most industries, AI visibility means being named when a buyer asks for options. A prescription drug is different. Patients cannot buy it on an assistant’s say-so; a prescriber decides, and a payer often decides whether it is covered. What a drugmaker needs from AI answers is narrower and stricter:
- Accuracy. The brand name, generic name, maker, approved indications and key safety information should match the approved labeling.
- Balance. The FDA’s Office of Prescription Drug Promotion (opens in a new tab) exists to ensure prescription drug promotion is “truthful, balanced, and accurately communicated.” An AI summary is not your promotion, but it shapes the same impression.
- Presence. When someone asks about a condition, the treatment options an assistant lists usually come from medical institutions and guidelines. Your brand appears if those sources discuss it.
- Practical facts. Patient support programs, savings information and how to talk to a doctor are questions people ask, and wrong answers cost prescriptions.
How far have patients and physicians moved to AI for drug information?
Far enough to matter: a third of adults use AI chatbots for health information, and most physicians use AI.
Patients and caregivers. KFF’s 2026 poll found 32% of adults turned to AI chatbots for health information in the past year, including 29% for physical health. About two-thirds of users wanted quick information, and 41% wanted to look something up before deciding whether to see a provider. The answer they get may also be the last word: 42% of those who asked about physical health did not follow up with a doctor or other health professional. And 77% of the public worries about the privacy of medical information given to AI tools.
Physicians. The American Medical Association’s 2026 survey (opens in a new tab) found 81% of physicians use AI professionally. Clinician tools are growing too. Research firm Sacra describes (opens in a new tab) OpenEvidence as searching licensed content including journals, specialty guidelines and drug labels, and reports that it earns its money from pharmaceutical and medical device advertising. Device makers face the clinician side of this too, covered in building device demand through AI answers.
Google. In our AI Overview study, all 20 question-form healthcare keywords we tested showed an AI Overview (100.0%), even though healthcare searches overall showed one only 43.0% of the time, mostly because local results crowd them out.
Meanwhile, the rep visit is less available. Veeva found half of accessible health care professionals meet with three or fewer companies, which leaves more of a prescriber’s learning to other channels, AI tools among them.
What do patients and prescribers ask AI about a brand?
Questions about what a drug is for, how it compares, what it costs and what to watch for. We wrote these prompts as illustrations; none was collected from real users.
| Asked by | What they want to know | Example question (written by us) |
|---|---|---|
| Patient | Indication | “What is [brand] approved to treat?” |
| Patient | Comparison | “How is [brand] different from [other brand] for the same condition?” |
| Patient | Cost and access | “Is there a savings program for [brand] if my insurance does not cover it?” |
| Patient | Generic | “Is there a generic version of [brand]?” |
| Caregiver | Safety | “What side effects should I ask the doctor about with [brand]?” |
| Prescriber | Label detail | “What does the prescribing information say about dosing in kidney impairment?” |
Each question is a chance for an assistant to get your facts right or wrong. Comparison and cost questions are especially risky, because the answer often blends sources of uneven quality, including telehealth and compounding sites. Latham & Watkins notes that the overwhelming majority of FDA’s September 2025 warning letters concerned online promotion of compounded GLP-1 products (opens in a new tab), a reminder of how crowded and contested some drug categories are online.
How does an AI answer affect prescriptions?
Indirectly: it shapes the conversation a patient brings to the prescriber and the confidence a prescriber has in the brand.
As we read the evidence, the paths run like this:
- Patient path. A patient or caregiver asks about symptoms, a condition or a brand they saw advertised. The answer names treatment options and, ideally, the right facts about yours. The patient raises it with a prescriber, who decides. How hospitals and practices get named in that search is covered in how providers win patients through AI.
- Prescriber path. A physician checks the label, the evidence or a comparison in an AI tool. A clear, current label and published evidence make an accurate answer more likely.
- Access path. After a prescription, patients ask about coverage, savings and pharmacy options. A wrong answer here can mean an abandoned prescription; we infer this, as no study measures it.
What makes pharma different from every other industry in this series is that the commercial value of AI visibility is bounded by regulation and medical judgment. The aim is not to persuade the assistant. It is to make the accurate version of your brand the easiest one for any tool to find.
The size of the stakes shows in ad budgets. Fierce Pharma, citing iSpot data, reports pharma and over-the-counter brands spent more than $7 billion on linear TV ads through early December 2025, up about 16%, and it reports a forecast that digital will make up 82% of healthcare and pharma ad spending by 2027. Every one of those ads now sends viewers to a search box or an assistant to check what they heard.
What decides how an assistant describes your drug?
The sources it finds: assistants lean on institutions, labels and independent coverage, and none publishes its selection rules.
Documented by the platforms. Google and OpenAI say their AI answers run web searches and show links to sources. Neither explains which drug pages they trust.
Observed in studies. An audit of 615 sources ChatGPT cited for consumer health questions, by Jacques and colleagues (opens in a new tab), found 75.7% came from institutional sources such as medical institutions, government sites and professional associations; commercial health platforms took 12.4%. Our summary of what cited commercial health sites have in common covers the details. Across buyer questions in other categories, our brand entity study found independent coverage was the strongest predictor of being named: each tenfold increase in independent sites naming a brand went with 4.7 times the odds of being recommended. That study did not test prescription drugs.
Observed accuracy problems. The BMJ Quality & Safety study used Drugs.com as its reference and found mean completeness of 76.7%. Inside Precision Medicine reports (opens in a new tab) that only 54% of the 20 answers the experts reviewed aligned with scientific consensus. The study tested Microsoft’s Bing chatbot, an earlier version of Copilot; newer assistants may do better or worse.
Trust factors specific to pharma.
- The label is the anchor. DailyMed, run by the National Library of Medicine, holds the labeling companies submit to the FDA (opens in a new tab), including prescribing information and patient labeling. Clinician tools describe answering from labels.
- Fair balance. The FDA notes that 100% of pharmaceutical social media posts in one 2024 review highlighted benefits, but only 33% mentioned potential harms. An assistant that learns from unbalanced material will repeat it.
- Institutional voices. Medical societies, patient advocacy groups and government health sites carry most of the weight in health answers.
Our inference. Assistants will describe your drug from whatever is most available and most consistent. If your label, patient information, brand site and third-party references agree, the accurate version has the best chance of being the one repeated. Earlier-stage assets face a related test with partners and investors, covered in how biotech companies get found for partnering.
What does a pharmaceutical brand risk by ignoring AI answers?
Inaccurate descriptions, missing safety context and lost conversations with prescribers; nobody has priced the loss yet.
- Errors you did not write. An assistant that misstates an indication or omits a boxed warning damages patient trust, and you may be asked about it. Our guide to fixing wrong brand information in AI answers covers what can be corrected.
- Absence from condition answers. If the institutions and guidelines that assistants cite do not discuss your brand, it may not appear when patients research their condition.
- Confusion with copies. In categories with compounded or look-alike products, assistants can blend your brand with others.
- Unstable answers. In our consistency study, only 25.2% of the brands ChatGPT named for a question came back in all five runs, so one check proves little.
- Regulatory attention. The FDA says it is already using AI and other tools to monitor drug ads. Your own channels need to be beyond reproach, because they feed what assistants say.
How does GEO work for a pharmaceutical company?
By making accurate, balanced, reviewed information about each brand easy to find and consistent everywhere. No mention is guaranteed.
- A clean entity record for each brand. State the brand name, generic name, maker and approved indications the same way on your brand site, corporate site, patient information and public reference pages, consistent with the label.
- Current, crawlable labeling. Keep prescribing information and patient labeling up to date in DailyMed and on your site as readable text, not only as image-based PDFs.
- Balanced brand pages. Pages that present benefits and risks together, in plain language, give assistants a balanced source to quote. Every page should go through your medical, legal and regulatory review.
- Authoritative third-party coverage. Support disease education through medical societies, patient advocacy groups and peer-reviewed publication, within the rules on industry funding and disclosure. How that third-party record turns into trust in AI answers is set out in how brands build authority for AI search.
- Access and support facts. Publish savings program terms, eligibility and how to reach support clearly, so cost questions get the right answer.
- Keep ads separate. Advertising on AI and clinician platforms is a different channel from being cited, with its own rules; see the risks of native ads in AI answers.
- Monitor for accuracy and safety. Track patient and prescriber questions across ChatGPT, Google AI Overviews and AI Mode, Gemini, Perplexity, Copilot and Claude, and route problems to medical affairs and pharmacovigilance as your procedures require. See how many prompts to track.
Do not try to game it. The FDA has named undisclosed paid influencer promotion as a concern, and planted reviews carry the same risk; see legitimate GEO versus manipulation.
Where are the gaps in what we know about pharma and AI answers?
Large: nobody has linked AI answers about a drug to prescriptions, and accuracy studies age quickly.
- No prescription data. We found no study connecting what assistants say about a brand to prescribing or adherence.
- Accuracy studies are snapshots. The Copilot study, published in 2025, tested an earlier chatbot, and assistants change often.
- Rules are in motion. The FDA plans rulemaking on broadcast ads, and it is unclear how promotion rules will treat sponsored content inside AI tools.
- Industry figures have sponsors. Ad spending estimates come from eMarketer and iSpot, and OpenEvidence figures from Sacra’s profile, not audited filings.
- Revenue links are unproven everywhere. See does AI visibility drive business results.
Where should a pharmaceutical company start?
With an accuracy audit of what assistants say about each brand today, reviewed by medical, legal and regulatory teams.
List the questions patients, caregivers and prescribers ask about each brand and its condition. Ask each one in more than one assistant, and repeat it on a few different days. Check every answer against the label: indications, safety information, generic status, access programs. Note which sources the assistants cite and where the errors come from.
We can help with that work: talk to us about an AI answer accuracy review for your brands. It shows where assistants misstate your label, leave out safety context or confuse your brand with others, and which reviewed, public sources would most improve the accuracy of the conversations patients bring to their prescribers. The generative engine optimization service page describes how the ongoing label, entity and monitoring work is organized to fit your medical, legal and regulatory review.
Frequently asked questions
Can we pay an AI assistant to recommend our drug?
We know of no assistant that documents selling placement inside its answers. Ads on some platforms are shown separately and remain subject to FDA promotion rules.
Is content written for AI search subject to FDA promotion rules?
Content a company publishes about its products is promotional material and needs the usual review. Ask your regulatory counsel how rules apply to new formats.
Why does an assistant describe our drug wrongly?
Usually because the sources it finds disagree or are out of date. Check the label, your pages and third-party references first.
Should we worry about compounded or copycat products in AI answers?
Yes, where they exist. Most of the FDA’s September 2025 warning letters concerned online promotion of compounded GLP-1 products.
Do physicians’ AI tools use our website?
Clinician tools such as OpenEvidence describe answering from journals, guidelines and drug labels, so the label and published evidence matter most.
Sources
- KFF (2026), Poll: 1 in 3 Adults Are Turning to AI Chatbots for Health Information (opens in a new tab)
- AHRQ PSNet (2025), Artificial intelligence-powered chatbots in search engines: a cross-sectional study on the quality and risks of drug information for patients (opens in a new tab), summarizing Andrikyan et al., BMJ Quality & Safety 2025
- Inside Precision Medicine (2024-10), AI-Powered Chatbots Unreliable for Patient Drug Information (opens in a new tab)
- U.S. Food and Drug Administration (2025-09-09), FDA Launches Crackdown on Deceptive Drug Advertising (opens in a new tab)
- U.S. Food and Drug Administration (n.d.), The Office of Prescription Drug Promotion (OPDP) (opens in a new tab)
- U.S. National Library of Medicine (n.d.), About DailyMed (opens in a new tab)
- Latham & Watkins (2025-09), FDA Begins Crackdown on Direct-to-Consumer Pharmaceutical Advertising (opens in a new tab)
- Fierce Pharma (2025-12), 2026 forecast: Pharma ad dollars will continue shifting away from traditional TV (opens in a new tab)
- Veeva Systems via BioSpace (2024-05-23), New Veeva Pulse Findings Show Connected Engagement Creates an Advantage as HCP Access Drops (opens in a new tab)
- Fierce Healthcare (2026), AMA: Physicians’ use of AI doubled from 2023 to 2026 (opens in a new tab)
- Sacra (2026), OpenEvidence revenue, valuation and funding (opens in a new tab)
- Jacques et al. (2026), Authority Signals in AI Cited Health Sources: A Framework for Evaluating Source Credibility in ChatGPT Responses (opens in a new tab), arXiv:2601.17109.
- Underneath (2026), When does Google show an AI Overview? 1,248 US searches
- Underneath (2026), Do Wikipedia and schema make AI assistants recommend a brand?
- Underneath (2026), Ask an AI the same question 5 times: do the brands change?