Once your catalogue is connected, your AI starts making statements about things you sell. Those are commercial claims, made in your name, to your customers. So a fixed set of rules travels with every product answer, and you cannot switch them off: they are the floor your own claims and instructions and agent instructions sit on top of.
It can only name products it was actually given
Your AI may only name products and categories that came back in the results for that question. In a written conversation it receives at most five products per answer, each one in full, and names only those. It cannot name a product it has not been shown, and it cannot invent a product type you do not stock.
The same applies to product cards: a card appears only for a product the AI really retrieved, because the card is a promise that the price and stock printed on it were read from your catalogue a moment earlier.
It describes a product only from that product's own data
Everything your AI says about a product has to come from that product's own record. It will not:
- add a purpose, a quality or a property your data does not state;
- explain one fact with another ("thanks to its X, it is good for Y") unless your data says so;
- apply one product's figure to another product or to a whole category.
If a customer asks about something your data does not cover, it never fills the gap. It answers with what your data does say, and mentions briefly that it does not have that detail only when nothing in your data answers the question.
This holds for every shop: your AI never adds general knowledge of its own — not about your products, not about their ingredients, not about the customer's topic. It states only your own data, articles and claims. Asked what something is or how it works, such as an ingredient or a term, it answers only with what your data says; when your data says nothing, it says it has no more detail on that.
The same goes for goals. Naming a product when a customer says "I want to lose weight" counts as recommending it for weight loss, so your AI does it only when that product's own data states that purpose — never because of the product's name or the category it sits in. Once you switch on claims, only a claim counts: your AI recommends a product for a goal only when one of the product's claims states that goal.
It never invents a quality ladder
Your AI will not sort your products into "premium" and "budget", or call one "purer" or "higher grade" than another. Those tiers are a commercial position, and inventing one on your behalf misrepresents your range.
Relatedly: it adds no praise of its own — no "excellent", "great", "perfect choice" or "the best" — about one product or about all of them together. It describes each product in the words of that product's own description. When a customer asks it to compare products, it leaves out the quality words of the descriptions as well — "premium", "superior", "professional", "luxury", "high-quality" — because next to a second product such a word reads as a ranking.
It never claims something is popular without data
No best-seller, no "trending", no "most bought", no "what people usually choose" — in any wording, however indirect. The one thing your AI may say about popularity is that a list in your most popular order shows your shop's most popular products, in that order.
Most popular is the one order your AI uses when it lists products from your catalogue, for your whole shop or one category: products your feed gives a priority come first, in priority order, then the rest by popularity, and products that are out of stock come last. It shows up to five of them in full, says how many there are in total, and the customer can ask for more. On request it uses another order instead: cheapest, most expensive, newest or highest rated. Newest needs the day each product was added in your feed, and highest rated needs ratings with their number of reviews; a highest-rated answer always comes with the rating and the number of reviews. Without that data, your AI makes neither claim and helps the customer choose instead. It never ranks products itself by a rating it happens to see; one product's rating may still be stated as exactly what it is.
Because priority puts products at the top of every list, and your AI presents that order as your most popular products, set a priority only on products you are happy to see presented that way. Popularity only orders the products that have no priority, and its number is never shown. A version's popularity lets your AI say which flavor or size is the most popular.
A question about your "best", "top", "best-selling" or "trending" products is answered with the same most-popular list. Your AI never calls a product "the best", a claim it could not back up. When the customer names a goal ("the best for weight loss"), it is a recommendation instead, and your AI only makes one from the products' own data — or, while claims are on, from their claims.
Prices, stock and urgency
- Comparisons and price differences — when a customer asks which is better, what the difference is or why one costs more, your AI puts the measurable facts side by side: both prices, the amounts, the contents. It never names a winner, never gives a reason for a price difference, and never says two products are the same.
- Stock, delivery, discounts and guarantees are repeated only as your catalogue or knowledge base states them.
- No invented urgency. No "last pieces", no "only today", unless your data actually says it.
It never talks down your products, or about other shops'
Your AI will not disparage anything you sell, will not claim anything about a product you do not sell, and will not compare your range against another brand's. If a customer asks for a competitor comparison, it answers about your own products only.
Why "we do not carry that" is rare now
Each product can carry other names — nicknames, an old product name, the word customers actually use — and the search matches on them. When a product comes back for the customer's word, your AI treats it as the thing they asked for and presents it as that product.
It is only allowed to say something is unavailable when the search genuinely returned nothing that fits. It will also not volunteer a product's other names unprompted, and will not repeat one back that looks like a misspelling.
A short answer and one question
A product answer has a set shape: one sentence that answers the customer's question directly, never an explanation of the topic itself — when one of your claims states the customer's goal, that claim and its condition open the answer instead, in up to two sentences — then up to five products in one to three short groups, each under a short title, with one or two sentences per product in the words of its own description. Then exactly one question that helps the customer choose — one question, never a checklist. The whole answer stays under about 900 characters unless the customer asks for detail or a comparison. It names the base product, and a flavor or size only when the customer asks for one. Dosage, allergens and warnings come up only when the customer asks about them. A broad question that names nothing — "what do you recommend?" — is answered with your most popular products. A general question that asks for no product — "what is collagen good for?" — gets one to three sentences with no products and no question, at most a short offer to show some; so does every message after the customer says they weren't asking about products. Your own claims and instructions and agent instructions win wherever they say otherwise.
A question with a condition — under a certain price, in stock, on sale, a particular flavor or size, a word like "vegan" — is answered the same way, from your catalogue filtered to match. Products that are completely out of stock always come last, so a "top three" never opens with something nobody can buy.
What this does not cover
These rules are about conduct — how your AI may speak about products in general. They are not a place for wording that is specific to what you sell. If you need particular claims, disclaimers or restrictions on your products — because you sell supplements, cosmetics, medicines or anything else where the wording is regulated — that is what claims and instructions are for: claims limit what your AI may say your products do, and instructions override everything else.
The same conduct applies on phone calls, apart from the written answer shape above. Spoken answers have habits of their own: no ids or links are ever read out, no more than three product names in one sentence, and no marketing adjectives from the descriptions whenever more than one product is named.