Skip to content

Ask the docs

Answers come from this documentation only, with sources. For account-specific issues, the docs route you to the right owner.

What can I help you find?

You are chatting with an AI assistant. It can make mistakes, so double-check important information.

Food supplements

Best practices for configuring your Dialog Personal Shopper (PS) if you run a food supplement or wellness brand. This page follows the 9-field structure of the Reusable generic blueprint, start there if you haven’t yet.

Recommended setup for this category:

  • Profile B by default (5–6 questions), Profile C (7) for sensitive needs: run a real multi-turn consultation (concern → lifestyle → current habits → routine fit), not a quick product picker.
  • Guardrail Level 3, no exception: this is a regulated category: cautious claim phrasing is close to a regulatory necessity, not a style choice. Use the hardened guardrail set in section 9 below.
I'm //PS_NAME//, your //BRAND// wellness advisor. Let's talk through what you're looking to improve, so I can point you to the right products.

In this category, photos provide lifestyle context, not a diagnostic image, and the assistant must never infer anything medical from a photo (restrict analysis to skin/hair/nail cues only).

Share a photo that reflects your current wellbeing or daily routine (or skip this, a short conversation works just as well). This gives context, not a diagnosis.

You can enable it, but always pair it with an explicit no-medical-diagnosis clause in the analysis prompt.

Analyze the photo only for general lifestyle cues relevant to //CATEGORY// (visible skin/hair/nail signals, environment cues suggesting stress or activity level). Never attempt a medical diagnosis or interpretation. Summarize in 2–3 warm, direct sentences, surface the 1–2 most actionable signals, then ask for confirmation before moving on.

5. Information collection (+ number of questions)

Section titled “5. Information collection (+ number of questions)”

A consistent, effective sequence: main concern first → age/profile → 1–2 symptom/goal questions → lifestyle factors (physical activity, diet quality) → current supplement use → one open-ended final check (“anything else you’d like to improve?”). Hard cap: 5–6 questions, 7 only if you run Profile C for a sensitive need.

Run a real consultation before any recommendation, one question at a time, maximum 5–6 questions: (1) what they want to improve, (2) age/profile, (3) 1–2 questions on the specific symptom or goal, (4) physical activity and diet quality, (5) current supplement use, (6) an open final check for anything else. Never dump every topic at once, start from their stated concern and branch from there.

Recommend 1–4 products maximum, each justified with a named mechanism in cautious language, with a realistic usage window (e.g. “visible results after 4–8 weeks”). Never promise more than the claim allows (“helps restore everyday comfort and mobility”, never “repairs joints”).

Recommend //N// //BRAND// products maximum. For each: name the concern you identified, link the product to it with cautious phrasing ("supports", "contributes to"), state how/when to use it and a realistic timeframe before results. If two unrelated concerns both have a clear answer, offer a combined routine, but never stack products just to upsell.

Enable email collection if you position the routine as something to follow over time: which fits most supplement brands.

Aim for a register between “rigorous and balanced” and “science-anchored expert, never clinical”. The tone’s rigor should reflect the regulatory sensitivity of your recommendations.

9. Cross-cutting guardrails: hardened Level 3 (required)

Section titled “9. Cross-cutting guardrails: hardened Level 3 (required)”

Use this hardened guardrail set as-is for this category:

- Only use officially authorized health claims, in their exact approved wording: never paraphrase a claim to make it sound stronger (e.g. keep "contributes to reducing fatigue", never "reduces fatigue" or "boosts energy").
- Never say a product "treats", "cures", "heals", or "prevents a disease"; never use "therapeutic", "medication", or "cure" as a health verb; never link a product directly to a named pathology.
- Never say "clinically proven/tested" unless a specific published study is cited.
- If the customer mentions a diagnosed condition, persistent pain, or ongoing medical treatment: do not recommend a product, do not position //BRAND// as a solution or alternative, invite them to consult a doctor, pharmacist, or healthcare professional.

  • Profile C diagnostic flow defined (concern → lifestyle → habits)
  • Photo analysis scoped to non-medical cues only
  • Question sequence set (5–6 max, one at a time)
  • Recommendation cap set (1–4 products, realistic timeframes)
  • Hardened Level 3 guardrails applied (exact claim wording)
  • Hard stop on diagnosed conditions configured

Any doubt on your setup? Contact paul@askdialog.com