AI now explains your products.
Who in your company sees what it says?
Patients, pharmacists and physicians ask AI assistants about indications, dosing and contraindications – often without ever visiting your website. I measure verbatim what these systems state about your products, verify every claim against your approved information, and trace each error to the team that can fix it.
Widely used, unstable and rarely cited.
When AI assistants answer clinical or consumer health questions, they blend official regulatory texts with third-party forums, outdated PDFs and foreign market approvals.
-
1 in 3
US adults use AI chatbots for health information – matching the share who use social media.
Source: KFF, March 2026 (US data; comparable Swiss figures not yet published).
-
38 %
of ChatGPT responses to 50 clinical drug-information questions were incorrect; 26 % carried a high risk of patient harm.
Source: Morath et al., European Journal of Hospital Pharmacy, 2024.
-
0 Links
Most inaccurate or incomplete drug answers are delivered without source links for the patient or pharmacist to verify.
Source: Morath et al., EJHP, 2024 & Seravel baseline tests, 2026.
Responses contained a statement that puts consumers at risk (omitting mandatory medical prescription under age 2).
53 of 56 verifiable product statements matched the approved information; 3 directly contradicted it.
40 of 96 statements were not backed by approved sources (e.g., model-guessed standard dosing regimens).
“Preparations of this type are often suitable for children (usually from toddler age onwards), though the package leaflet should always be checked for safety.”
“In toddlers under 2 years of age, this medicinal product may only be used on medical prescription.”
Root cause traced: The mandatory age restriction is locked inside a PDF leaflet rather than indexable HTML on the product webpage. The model bridges the gap with a generic category assumption.
-
01 · Publish dosing & age warnings as structured HTML Web & Regulatory
Surface mandatory safety facts directly on the product page so search-augmented AI systems read the approved text rather than guessing.
-
02 · Explicitly state indications and exclusions Medical Affairs
Define clearly what the preparation is approved for and which patient or device groups (e.g., contact lens wearers) are excluded.
-
03 · Align multilingual product naming (DE / FR / IT) Digital Governance
Ensure consistent phrasing across national languages to prevent models from blending foreign regulatory data with Swiss facts.
Set up once. Measure on a fixed cycle.
I test 12 to 16 realistic questions per product across the agreed AI assistants, multiple times each, and verify every statement against facts your team has confirmed. For prescription medicines, corrections go strictly through approved information channels, never promotion.
Establish the verified baseline
- 1Add brand & marketsDefine products, languages and target geographies.
- 2Read in official sourcesIngest approved web pages, IFUs and package leaflets.
- 3Confirm reference factsA qualified person on your side confirms each reference statement.
- 4Select key messagesDefine up to 10 core safety or positioning facts to track.
- 5Fix the question set12–16 everyday patient, pharmacist or physician queries per product.
- 6Run baseline measurementAutomated multi-run capture across models with human sample review.
Track remediation impact
- 7Review findings & sampleHuman verification of flagged safety risks and contradictions.
- 8Assign remediation measuresPrioritised action list per topic, mapped to Regulatory, Medical or Web owners.
- 9Verify effect in next runRe-run the exact same prompts (logged by model version and date) to prove whether fixes changed the answers.
Seravel is my proprietary measurement instrument, currently in beta. For the pilot phase I am partnering with pharma, OTC and medtech organisations. Read about Seravel and the method →
Two ways in. One clear bridge.
Product pages, leaflets, directory entries and press releases are maintained by different teams and external agencies. When nobody owns the whole architecture, outdated versions stay online – and AI models surface them. Where cross-functional ownership is the real gap, I step in on the client side for a fixed term.
AI Answer Audit
For Medical, Regulatory and Brand leads who need immediate clarity on what AI assistants tell patients and professionals about their products.
Fixed-Term Digital Leadership
For executives facing a major platform decision, multiple agencies without unified delivery steering, or a vacant digital leadership role.
What executives ask first.
Clear boundaries on governance, compliance and how client-side advisory differs from agency work.
Why does this matter if the AI is not ours?
Because patients and healthcare professionals act on the answer directly in the chat, often without visiting your website. If an assistant states an outdated dose or omits a warning, the person asking may never see your approved information. You cannot control the model, but you can audit what it states, trace which sources it relies on, and remove outdated copies still online.
Is this pharmacovigilance or medical information?
No. The audit is not a pharmacovigilance or medical-information service, and it gives no medical or legal advice. It measures verbatim what AI assistants say and where errors likely originate. Your existing processes for drug safety, medical enquiries and regulatory approvals remain strictly in charge.
Can you make the AI answer correctly?
No external party controls model weights or provider algorithms. What you directly influence is your own digital evidence: structured HTML product pages, current leaflets, and clean directory data. The audit shows where to act, and the 14-to-28-day re-measurement verifies whether answers have improved.
How is this different from a consultancy or an agency?
I am not an implementation agency. My role is strictly client-side: measuring, framing decisions, evaluating vendors, setting acceptance criteria and steering delivery. I take no vendor commissions and no agency margins, which ensures recommendations hold even when they cost a supplier the contract.
Start with one product – or one conversation.
For the free Seravel Snapshot, send me your company, the product name, desired languages and a contact. You receive a one-page baseline excerpt within 48 hours. If your priority is an upcoming platform decision or a leadership gap, book a confidential call instead.
Please do not send confidential or personal data before we agree a secure channel and scope.