Why is my claim or instruction not applying

You wrote a claim or an instruction, the product is clearly in the answer, and your AI carried on as if it did not exist. Work down this list — the answer is nearly always near the top.

1. Is it switched on?

Switched-off claims and instructions are never sent to your AI, on any channel. On the Claims and Instructions tabs, the switch in the On column shows which rows are on. For claims, the top of the Claims tab also tells you whether Claims are on at all.

Anything an AI wrote — a document import, or Copilot — arrives switched off on purpose and stays that way until a person approves it.

2. Do the product and the claim really share a key?

This is the most common cause by far. Matching is exact: the claim or instruction and the product must share a key after tidying (lowercased, accents stripped, punctuation turned into spaces). A product's keys come from the keys your shop sends in its rule_keys field, or from the fields the source turns into keys.

On the Claims tab, the claim's row tells you straight away: Applies to N products means it reaches that many active products, and No product has these keys means it reaches none.

To check one product, open Brand → Products → Products, click the product, open its Claims & instructions tab, and look at:

• Keys — every key it carries, grouped by where each came from.
• Claims and Instructions — listed separately: every switched-on one that reaches it today. The switched-off ones are behind a button underneath.

If yours is not in those lists, the keys do not match. Watch for:

• A synonym or another language. "vitamin C" does not match "ascorbic acid". Add both to the same claim or instruction.
• Word order. "vitamin c" and "c vitamin" are two different keys. Add both.
• A key made only of punctuation. Something like — saves happily but can never match anything.

3. Does the product have any keys at all?

If the product's Claims & instructions tab shows "This product has no keys, so no claim can apply to it — only instructions set to every product of the brand", no claim and no keyed instruction will ever reach it. Two possible reasons:

• your shop is not sending keys with that product, in its rule_keys field; or
• the source is not set to turn the right field into keys — check Fields that become keys in the source's settings.

Changing that setting re-derives every product's keys immediately, so you can check again straight away.

4. Is the product coming from a CSV feed?

CSV feeds carry no keys. Products read from a CSV file get no keys at all, so no claim can ever apply to them — only instructions set to every product of the brand. The Sources tab says this under the table.

The fix is a Product API source. You cannot convert a CSV source — you add a new one of the right kind.

5. Does a claim cover the visitor's goal?

While claims are on, your AI recommends a product for a goal only when one of that product's claims states the goal. For a goal no claim covers, it shows no products — a product the visitor named stays — and says briefly that it cannot recommend a product for that. A product without a claim still appears in answers, but your AI won't say what it does. That is claims working as intended, not a fault: if the goal should be covered, add a claim for it, keyed to the products it is true for.

6. Was the claim chosen for this question?

When a visitor asks about a goal, a quick AI check picks the claims that fit what they are after, and only those travel with the products. For an unrelated question your claim is not chosen — that is correct, not a fault. When the check chose no claim, each product's claims all travel as usual. Instructions are never filtered: an instruction that reaches a product always travels with it.

Still stuck?

Two rarer possibilities:

• The product is disabled or gone. A product that has been switched off, or has disappeared from its source, is dropped from the answer entirely rather than described without its claims and instructions.
• It is switched on, but your AI ignored it. This is unusual — claims and instructions are handed over as binding wording that overrides everything else. If you can reproduce it, contact support with the conversation.

See also: claims and instructions explained.