Vertical guide — Pharma
AI visibility for pharma.
What the engines tell patients about your medicine.
Patients and prescribers ask assistants about medicines before they ask anyone else. We measured what four answer engines say about two prescription medicines across 288 answers and 2,321 citations. The manufacturer's own label was a minority voice in both.
By Sri · Last reviewed: July 2026
The core thesis
The answer is being written without you.
When someone asks an assistant about a prescription medicine, the engine runs a retrieval step before it writes anything. Whatever it pulls back becomes the answer. In our sweeps that pool was dominated by consumer drug references, and on one brand by telehealth and compounding services that monetise the molecule.
This is not a ranking problem and it is not solved by publishing more. It is a retrieval problem, and the first step is knowing which questions you lose, on which engine, to which source.
Why pharma is different
Four things we found that don't happen in other categories.
Your label is a minority voice in the answer
Across the two medicines we measured, manufacturer-owned domains accounted for 8% of cited sources on one brand and 20% on the other. On most individual questions no manufacturer or label source appeared at all — the answer a patient reads was assembled entirely from third parties. It isn't that the content doesn't exist. Consumer drug references are written as direct answers to the questions people actually ask, while product sites are organised around the product and wrapped in regulated language that reads poorly as an extractable passage.
Who fills the gap is category-specific
The two medicines had near-opposite risk profiles. On the GLP-1, telehealth prescribers and compounding services made up 17% of citations and outnumbered the manufacturer's own sources by more than two to one. On the immunology biologic that channel was marginal, but patient forums rose to 7% and Reddit was among the most-cited domains overall. One brand's problem is a commercial channel selling around it; the other's is unmoderated patient discussion being treated as evidence. A single visibility score tells you neither.
Litigation marketing surfaces on safety questions
Both sweeps returned plaintiff-firm and drug-injury sites as cited sources on questions about risks and side effects. A patient asking what the warnings are can receive an answer grounded partly in injury-claim advertising. That is a materially different escalation from an answer grounded in a clinical reference, and most brand teams have no visibility into whether it is happening.
The engines disagree with each other, sharply
On an identical question set, the rate at which each engine cited a manufacturer source ranged from 15% on ChatGPT to 64% on Perplexity for one brand, and 26% to 82% for the other. Gemini and Perplexity reach for label and manufacturer sources far more readily than ChatGPT and Claude. Checking one assistant and generalising will mislead you in whichever direction that engine happens to lean.
The full method and findings are written up in how visible are pharma brands in AI answers.
What we ask
A fixed question set, asked the way patients ask it.
The question set is fixed and reproducible, so month-over-month movement is comparable rather than an artefact of asking differently. It spans safety, efficacy, comparison, dosing, cost, alternatives, sourcing risk and long-term concern:
- → most common side effects of [molecule]
- → does [brand] have any serious warnings
- → how well does [brand] actually work
- → [brand] vs [competitor] — which is better
- → how is [brand] dosed and administered
- → how much does [brand] cost without insurance
- → what are the alternatives to [brand]
- → is there a cheaper generic or copy of [brand]
- → can I buy [molecule] online without a prescription
- → is it safe to take [brand] long term
- → what happens when you stop taking [brand]
The boundary
We describe what AI said. Nothing more.
RxPulse is an observational tool. It is not patient-facing, holds no patient data, and asserts nothing about any medicine's safety or efficacy. Every signal it raises is an observation about an AI answer, written for a qualified medical or regulatory reviewer to adjudicate.
That boundary is deliberate. It keeps the work outside GxP validation scope and outside promotional review, which is what makes it something a medical affairs team can actually run this quarter rather than next year.
Questions
Frequently asked.
Is this a regulated or GxP system?
No, and deliberately so. RxPulse observes what public AI systems say about a medicine. It is not patient-facing, it holds no patient data, it makes no claim about any product's safety or efficacy, and it produces no regulated output. Every flag is an observation for your medical and regulatory team to adjudicate. That is what keeps it outside validation scope while still being useful.
Do you make claims about our product?
Never. The report describes the behaviour of answer engines — which sources they cite, in what order, and whether yours is among them. It quotes AI answers as evidence of what is being said. It does not assert anything about the medicine itself, and the language is written so it cannot be mistaken for doing so.
How is this different from pharma SEO?
Traditional SEO asks whether your page ranks. This asks whether the engine cites you when it composes an answer, which is a different mechanism with different winners. A page can rank well and never be retrieved, and a source can be cited constantly without ranking for anything. We measure the citation layer directly, per engine, per question.
Is one measurement enough?
No. Answer engines are non-deterministic, so a single run is a draw rather than a measurement. We ask every question three times per engine and report a hit rate plus a stability score. In our own sweeps single-run measurement was wrong in both directions — one brand read low, the other read high — which is exactly the kind of error that gets a report challenged in the room.
What do we actually receive?
A per-brand report covering source mix by tier, manufacturer citation share and position, engine-by-engine hit rates, contested questions, and an answer-by-answer appendix with every cited source in the order the engine used it. Run on a schedule, it becomes a trend line rather than a snapshot.
See what the engines say about your brand.
We run the fixed question set against your medicine across ChatGPT, Claude, Gemini and Perplexity, three samples each, and give you the source mix, the contested questions and the answer-by-answer detail.
Book a call