If you sell something where the wording is regulated, you cannot leave it to the AI to phrase claims about your products. Product rules fix that: they are sentences your AI is required to follow whenever a particular product comes up in an answer.
A rule is keys plus text
- Keys — the words a rule and a product have in common. Usually an ingredient or a product name.
- The text — the sentence your AI reads word for word.
When a product appears in an answer, every active rule sharing a key with it is handed to your AI as binding wording that overrides everything else it found, overrides anything said earlier in the conversation, and overrides what the visitor asks for.
Matching is exact, not clever
A rule reaches your AI when it and the product share at least one key. There is no AI in the matching and no "close enough" — just an exact match on a tidied-up version of each key.
Tidying means: everything is lowercased, accents are stripped, punctuation becomes spaces, and repeated spaces collapse. So C-Vitamin, (Ascorbinsav)! and c vitamin ascorbinsav are the same key.
Synonyms and translations are not matched for you. A rule keyed only on "vitamin C" will never reach a product whose only key is "ascorbic acid". Put every spelling of the same thing — every language, every trade name, every abbreviation — into one rule.
Where a product's keys come from
A product gets keys from two places, and the product's page shows you which came from where:
- the keys your shop sends with the product;
- the values of any product fields you nominate in the source settings — name your Ingredients field, for instance, and every ingredient becomes a key automatically.
Changing which fields become keys takes effect immediately, with no refresh to wait for.
Always, or when relevant
Every rule is one or the other, and this is the only control over whether a rule can be left out:
- Always — passed to your AI every single time a matching product comes up. Never filtered. Use it for conditions, restrictions, mandatory disclaimers and anything that must never be said.
- When relevant — passed only when it relates to what the visitor actually asked. Use it for situational guidance that only matters for some questions — not for permitted-claims wording, which belongs in an Always rule.
If the relevance check ever fails, it fails safe: every "when relevant" rule goes through, exactly as if it were an Always rule. Nothing is quietly withheld.
Writing one
On Brand → Products → Rules, press Add rule. You fill in the keys, the text (up to 5,000 characters), an optional note for your team that the AI never sees, and whether it applies always or when relevant.
A rule you type yourself goes live immediately — a person wrote it, so there is nobody left to review it. Rules an AI wrote are a different story: see importing rules from a document.
Where rules apply
Every answer a visitor can see: the chat widget, email, Facebook, Instagram, phone calls, and the AI reply suggestions your team sees in the inbox.
They are deliberately not applied where nobody is being answered — the panel that shows an agent what your knowledge holds, and AI social posts.
Two safeguards worth knowing
- Draft rules never reach your AI. Not on any channel.
- A product is never described with only some of its rules. If there were ever too much rule text for one answer — or simply too much retrieved content overall — whole products leave the answer instead, passages, product card and rules together.
If a rule is not doing what you expect, see why a rule is not applying.