Supplements & wellness: AI as script operator, not health advisor
In regulated categories — supplements, wellness, cosmetics — AI Voice Agents have narrowly defined legal boundaries. We show what's allowed, what isn't, and how to configure the script so AI doesn't slip into 'medical advisor' mode.
AI Voice Agent in wellness/supplements is an operational tool, not advisory. It reads from a prepared script (store owns content), confirms orders, suggests catalog products — but doesn't diagnose, dose, or check ingredient interactions. Boundary is simple: AI = operator, store = expert.
- Allowed: read label, confirm order, suggest catalog product.
- Banned: diagnosis, dosing, ingredient interactions, 'health-stacking'.
- Legal basis: URPL (Polish health regulator), food safety act, GDPR Article 6.
- Store writes script → legal sign-off → AI reads. Not the other way around.
What AI may say to a wellness customer
In Poland, supplement and cosmetic categories fall under three overlapping regulatory regimes: URPL (registry and oversight of health products), food safety act (supplements = foodstuffs), GDPR for any conversation with sensitive data.
For an AI Voice Agent confirming orders this means a very narrow space of what's allowed. The boundary is binary: reading from a store-approved document = OK. Generating new health content during the conversation = NOT OK.
Consequences of crossing the line are not theoretical. URPL actively reacts to marketing communication implying medicinal action. Penalty: up to 5,000,000 PLN.
AI is not a naturopath or health consultant
Most common copy trap: writing the script in 'Your virtual wellness advisor' style. This framing creates legal exposure even if content is neutral. A customer hearing 'advisor' has the right to expect advice.
Safe framing: AI is an order operator, catalog assistant, product consultant (not health). 'Medical advisor', 'wellness specialist', 'virtual naturopath' — these are the red lines.
- 1GOOD: 'AI confirms your order and helps pick from our catalog'. Operator + catalog. Legal boundary clear.
- 2GOOD: 'AI reads label info if you'd like to check it'. Reading approved content. No generation.
- 3BAD: 'AI will advise what works best for joint pain'. Implies diagnosis + medicinal recommendation. URPL/PIP automatically.
- 4BAD: 'Your virtual health consultant'. Regulated professional title. Creates exposure even if content is neutral.
Script approved by the store, not by AI
Industry AI Scripts (Starter+): store inputs scripts as text. AI reads verbatim, with permitted variation only for variables: customer name, order ID, items, address.
Scale+ adds versioning + A/B test per script — enables legal sign-off per version. Before publishing, the field 'approved by: [lawyer name]' is required (frozen).
Cross-sell in wellness (Growth+) is allowed but with modified map: instead of 'ingredient synergy' (banned), use 'customers often order together' (commercial fact, not medical claim).
What to do when customer says 'I take medication'
During calls customers sometimes reveal sensitive data (Article 9 GDPR special category): mentioning meds, pregnancy, allergy, condition. AI Voice Agent should:
First, not respond substantively. The 'AI = operator' rule means: 'Thank you for the information. If you have product questions related to health, please contact our support — number in product description'.
Second, record the call with appropriate retention. Standard 90 days for confirmation calls, but conversations with sensitive data usually require longer — see GDPR & recordings article.
Four things to remember
AI = operator. Store = expert. Binary boundary. AI reads script, store owns content.
Framing matters: not 'advisor', but 'catalog assistant'. Regulated professional title creates exposure even with neutral content.
Lawyer signs off the script BEFORE publishing. Scale+ has 'approved by' field in panel — frozen, can't be skipped.
Sensitive data = don't respond substantively. Rule: 'thank you, please contact our support'. Recording with longer retention.
Wellness store?Test AI in operator mode.
1,000 trial credits (~20 confirmation calls) · full access to your chosen plan · cancel anytime. Card required · no charge during trial · auto-activates on day 8.