# When a shopping guide should ask again > A fictional shopping guide separates clarification, missing matches and health refusal, then makes a supported product choice inspectable. - Author: Ratik - Published: 2026-09-10 - Category: Concept Walkthroughs - Canonical: https://www.arclift.ai/blog/guided-product-discovery-concept “I’m not sure which format.” “Show me a protein bar.” “Will this help me sleep?” These three questions open different paths in PLAIN / LIVING’s Guided Product Discovery concept. One needs clarification. One has no match in the declared catalog. One asks for a health recommendation the guide cannot make. The experience uses fictional products, original AI-generated imagery and a local scripted conversation. There is no live AI model, health assessment or connected support service. Its purpose is to show what happens when a shopping guide reaches a limit, alongside the product choice it can support. Explore the Concept: [Guided Product Discovery](/concepts/guided-wellness#ask) ## Give different problems different next steps An unclear preference does not require a product recommendation. The guide asks whether you want capsules or powder and explains their physical formats. The customer supplies the missing detail before a selected product appears. A protein bar request has a different problem: neither product in this catalog is a protein bar. The guide names that absence and does not offer a canister of powder as a substitute. The no-match state has no pictured selection or add-to-basket action. A question about sleep reaches the health boundary. The guide declines to recommend a product or predict an effect. A visible human route opens a request summary identifying the limit and an appropriate professional route. It contains no recommendation or personal details. Preparing that request changes the local screen; nothing is sent and no person receives it. The optional text field recognizes authored examples and declared format aliases. Unrecognized input receives an unknown response. Health terms take precedence over a format mention, so adding “capsules” to a recognized health question does not produce a format match. Text is limited, masked from analytics and cleared after submission. These rules describe this authored demo, not the reliability of an unrestricted language model. ## Show what a supported choice rests on Choose Capsules and the guide shows Daily Essentials: 60 capsules, from the declared Catalog · C-02 record. Choose Powder and it shows Plain Blend: 300 g, Neutral flavor, from Catalog · P-01. The reason is explicit: the product has the format the customer chose. That match says nothing about suitability for a health concern. The product plan keeps the pictured item and pack together. Switch products and the explanation describes the current comparison. It does not claim that powder matches an earlier capsule preference. Source facts remain available beside the choice, with the record identifier visible for inspection. Correcting catalog fields is a separate task. The linked AI-Ready Product Catalog concept contains that workspace, while this experience keeps attention on the customer’s decision. ## Let the customer inspect before adding The plan includes a short reading item about understanding a pack label. Open it, save the sample article locally and return to the same plan. It explains format and pack size without making a product claim. Adding a product updates a local basket that preserves the selection and pack. Remove an item or reset the experience to clear the basket, preference, saved article, input and handover state. No checkout, payment, subscription or real order is created. ## What production would need A real build would need approved catalog information, supported questions and an owner for every boundary. The store would need to decide when the guide can clarify, when it must stop and how a real handover reaches a person. Availability, variants and the actual cart would require their own integrations and checks. If a language model were introduced, its answers would need separate evaluation against these rules and source boundaries. Acceptance should include missing facts, unsupported requests, health questions and catalog changes. Passing this scripted demonstration does not establish that another system would respond the same way. The supporting reading also needs an owner. A product change should reach the catalog, the guide’s explanation and any linked material that depends on it. ## What this demonstrates PLAIN / LIVING makes the guide’s limits inspectable, then connects a supported preference to source facts and a basket. The concept does not establish higher conversion, greater trust or any health outcome. Try the three questions before following a supported format into the basket. For your store, bring a decision customers struggle with and the facts available to support it. That gives us a concrete starting point for a focused prototype. ## Your brand’s next step [Shopping Assistants & Guided Selling](/services#shopping-guides) [Start a Conversation](/contact?service=shopping-guides&example=guided-wellness) ## Related reading [Product records for AI discovery: what to fix, what not to promise](/blog/product-record-readiness)