Your package leaflet is approved. What AI tells consumers isn't.
It's late. Someone has a cold, a blocked nose, and a medicine for something else already on the bedside table. Instead of calling a pharmacy, they ask an AI: “Can I take these together?” The answer arrives in seconds and sounds certain. Whether it matches your approved package leaflet, nobody at your company knows.
This is no fringe habit. In the US, about one in three adults used an AI chatbot for health information in the past year1. We measure what those answers say about your products, question by question, against your own approved information. No one steers the AI itself. You can steer your content, your evidence and your visibility.
The answer reaches the consumer before anyone at your company sees it.
Confident, and sometimes unsafe
Experts reviewed a subset of 20 answers that an AI search chatbot gave to common patient questions about medicines. They judged 66% potentially harmful, and 22% could have led to severe harm or death if followed2. In a physician-led test of four chatbots on 222 patient questions, 5 to 13% of answers were unsafe, depending on the chatbot3.
Ask twice, get a different answer
When hospital pharmacists re-entered drug questions into ChatGPT at a later date, only 3 of 12 answers came back identical4. Across ChatGPT, Claude and Google's AI, repeated product-recommendation prompts returned the same list of brands less than once in a hundred times5. A single answer is an anecdote. Only repeated measurement shows the pattern.
Cited isn't the same as used
In the pharmacists' test, none of the 50 ChatGPT answers named a source4. And where AI search does cite, a 2026 preprint across ChatGPT, Google and Perplexity shows that being cited and actually shaping the answer are two different things6. Whether your approved wording is what shapes the answer can't be assumed. It can be measured.
You can't steer the model. You can steer these three things.
No one can tell ChatGPT, Claude or Gemini what to answer. So the audit focuses on where you can act, and measures whether it works.
1 · Your content
You see which questions trip the AI up, such as a dose, a German pack instead of the Swiss one, or a missing warning. That is exactly where you improve your product page, FAQ, structured data or the package leaflet as a web page. The next measurement shows whether the answers change.
2 · Your evidence
Every question, every verbatim answer, every review and every measure is recorded with date and person. So you can show Regulatory, Quality or management that you are watching, spotting deviations and acting on them.
3 · Your visibility
For general questions such as “What helps with …?” you see whether your product is named, which competitors the AI prefers and whether your key messages land. Re-measuring every 14 or 28 days shows whether that changes after your measures.
The audit does not stop AI errors. It shows where they arise, what you can do in your own content, and whether it works.
One question, five answers, the wrong country
Illustration — fictional product, no real dosing advice
- Question
- “How much Musterflu can I take per day?”
- Your reference
- The Swiss package leaflet: the strength of the Swiss pack and its maximum daily amount.
- Result (5 runs)
- 2 of 5 answers give the dosing of the version sold in a neighbouring country, which has a different strength than the Swiss pack.
- Where it comes from
- Both answers rely on a foreign pharmacy page. Your Swiss product information is cited in none of the five runs.
- What helps
- Mark your product information clearly as the Swiss version, with country and strength on a citable web page, not only in a PDF. Then measure again to see whether the answers move.
Illustration only. The audit measures deviation from your approved statements; it is not medical advice and does not assess clinical content.
Four steps from question to finding
-
Step 1
You define what's correct
For each question you provide a reference statement from your approved public sources: package leaflet, product page, approved claims. On request, the system proposes statements with the source passage, and you confirm or correct them. Only facts confirmed by a person count, and every answer is measured against them and nothing else.
-
Step 2
Questions the way consumers ask them
12 to 16 questions per product in plain, everyday language, two for each typical situation: what it's for, whether it can be combined, how to dose it, which product suits a need. Including the delicate ones about children, pregnancy and other medicines.
-
Step 3
Asked like a consumer would, more than once
The questions go to the AI engines you choose, without your references and without any hint that a test is running. Each question is asked more than once, with and without live web search, because answers vary. Every answer is stored word for word, with the date and the model.
-
Step 4
Checked against your reference
An AI model from a different provider than the one that answered breaks each answer into individual statements and checks them against your reference: confirmed, wrong, unsupported, or a key message missing. It rates the deviations by error type and severity. A person then reviews a sample of the ratings, and safety findings stay marked as provisional in the report until a person has confirmed them. You get the original answer, the deviation, and a suggested owner for the fix.
Automated runs use the engines' official programming interfaces (APIs). An API answer can differ from what the consumer app shows, so where it matters we add manual checks in the chat apps, assessed the same way, and every result says which route it came from. Automation runs and documents the checks; qualified people set the reference and review the findings.
At a global pharmaceutical group I was accountable for three internal audits in digital governance and closed the findings they raised; the result was a group-wide directive for digital compliance.
Signs it's worth a look
- Dosing & interactions: AI gives guidance that differs from your package leaflet.
- Intended use: AI describes uses beyond or different from the approved purpose.
- Country versions: AI mixes up the Swiss product with versions sold abroad, for example in strength or pack size.
- Brand & recommendation: AI recommends a competitor, or confuses your brand, when consumers ask “what's best for X?”
- Availability & claims: AI gets format, availability or claims wrong.
What this is not
- No medical, regulatory or legal assessment: whether a deviation matters is for your own experts to judge.
- No ranking: position changes from one answer to the next5, so we measure how often and how accurately your product is described.
- No guarantee: no one can dictate what a model says. So the audit works on your content, your evidence and your visibility, and shows whether the gaps close.
Start free. Fixed prices after.
Snapshot
Free
Five questions, one engine, one page. Four about your product – what it is for, age and dosing, pregnancy and breastfeeding, side effects and interactions – and one general question about your product category. The safety statements in the answers are compared with your package leaflet: matches, inaccurate, contradicts, or not in the leaflet. Deviations are shown with both wordings side by side.
In return: 15 minutes to go through the result together.
AI Answer Audit
from CHF 15,000
Every question asked more than once, across the engines and languages you choose, with and without live web search. Every answer stored word for word and checked statement by statement against your approved reference — by a model from a different provider than the one that answered — with error type, severity and a suggested owner per fix. 12–16 questions per product, about 2 weeks.
Final scope and price confirmed on the scoping call.
Monitoring
Recurring
The same questions, re-measured on a fixed cadence. Because the same AI service can change within months: in one study, GPT-4 solved a test task correctly 84% of the time in March 2023 and 51% in June7. Each re-measurement also shows whether the answers have changed since your fixes.
For brands whose AI answers need watching.
All prices in CHF, excluding VAT. Final scope is confirmed during the initial scoping call.
Often asked before a test
What do you need from us for a test?
For the snapshot, only the product name; send the package leaflet if you like, otherwise I use the publicly available one. For the audit, your product or brand name, your domain, and for each question one reference statement with its public source — for example your package leaflet, product page or approved claims. We agree the questions together. Please share only information that is already public and approved for external use.
Who sees the results?
Only you. Results are not published and not compared with other companies.
Which AI engines are tested?
The engines you choose, confirmed before we start. Every result states the engine, the model and the date, and whether the answer came through the engine's API or its chat app.
Does the test tell us whether an answer is a compliance issue?
No. It shows where an AI answer deviates from your approved statements. Whether a deviation matters from a regulatory, legal or medical point of view is for your own teams to decide.
How reliable is the result?
AI models generate every answer anew, so the same question can get a different answer next time. That is why, in the audit, each question is asked more than once and the pattern across all questions counts, not a single answer. Every result states the date and the model.
Can you make the AI answer correctly?
No, nobody can. ChatGPT, Claude, Gemini and others generate their answers themselves. We show where they deviate from your approved information, what you can improve in your own content, and then measure whether the answers change.
Sources
- KFF (March 2026). Poll: 1 in 3 adults are turning to AI chatbots for health information. KFF Tracking Poll on Health Information and Trust, US adults, n = 1,343. (US data; comparable figures for Switzerland are not available to us.)
- Andrikyan, W. et al. (2025). Artificial intelligence-powered chatbots in search engines: a cross-sectional study on the quality and risks of drug information for patients. BMJ Quality & Safety, 34(2). (Bing Copilot; expert harm rating on a subset of 20 answers, hence wide confidence intervals.)
- Draelos, R. L. et al. (2025). Large language models provide unsafe answers to patient-posed medical questions. arXiv preprint 2507.18905. (222 questions, 888 answers from Claude, Gemini, GPT-4o and Llama 3; physician ratings.)
- Morath, B. et al. (2024). Performance and risks of ChatGPT used in drug information. European Journal of Hospital Pharmacy, 31(6), first published online 2023. (50 real-world drug-information questions collected by hospital pharmacists; ChatGPT version of 2023.)
- Fishkin, R. & O'Donnell, P. (2026). AIs are highly inconsistent when recommending brands or products. SparkToro; summary in Search Engine Journal. (2,961 runs of 12 prompts by 600 volunteers; not peer-reviewed; one author's company sells AI tracking.)
- Zhang, K., He, X. & Yao, J. (2026). From Citation Selection to Citation Absorption: A Measurement Framework for Generative Engine Optimization Across AI Search Platforms. arXiv preprint 2604.25707. (Descriptive study, 602 prompts; no causal claims.)
- Chen, L., Zaharia, M. & Zou, J. (2024). How Is ChatGPT's Behavior Changing over Time? Harvard Data Science Review. (GPT-3.5 and GPT-4, March vs. June 2023; general tasks, not health.)
Request a free snapshot
Tell me which product you have in mind. On five real questions, one engine, one page, I'll show you what AI assistants tell consumers about it today, and where that differs from your package leaflet.
Request a snapshot