Claims and instructions

The two kinds of product rule: claims say what a product may be said to do, be good for or suit, and instructions are followed whenever your AI talks about a product. Both reach products through exact keys, never by guesswork.

Claims and instructions

Product rules live on two tabs of Brand → (your brand) → Products: Claims and Instructions. The old Rules address forwards to Claims, so existing bookmarks keep working.

ClaimInstruction
What it isWhat a product may be said to do, be good for or suitSomething your AI must follow whenever it talks about the products it applies to
When your AI gets itWith the products it applies toAlways, with every product it applies to
Picks products for a visitor's goalYesNo
Can apply to every product of the brandNo — a claim always reaches products through keysYes
Example"Vitamin C contributes to the normal function of the immune system.""Not recommended for children, or for pregnant or breastfeeding women."

Every rule is two things: keys — the words a rule and a product have in common, usually an ingredient or a product name — and the text your AI reads. Instructions are handed to your AI as binding wording that overrides everything else it found, anything said earlier in the conversation and anything the visitor asks for.

Claims and instructions exist for products only. They are not a place for general instructions to your AI — the agent's own instructions field covers those. They are for the cases where the wording is not yours to choose: regulated health claims, dosage warnings, legal restrictions.
Every instruction that applies to a product is sent with it, every time. Claims are sent with their products too; only when a visitor asks for a goal does your AI narrow them down to the claims for that goal (see What your AI does in claims mode below).

How a claim or instruction reaches a product

A claim reaches a product when the two have at least one key in common. An instruction can reach products in one of two ways, and you pick which when you write it:

Products that share a key
default
The instruction and a product in the answer have at least one key in common. Use it for anything said about a particular substance or product.
Every product of this brand
The instruction reaches every product in the answer whatever keys it carries — including products with no keys at all — and carries no keys itself. Use it for wording that is true of your whole catalog. Claims never have this option.

There is no AI in this match, no similarity scoring and no ranking: a key match is an exact match on a tidied-up version of each key, and an instruction for every product simply always applies.

A product's keys include its versions' keys. Keys come from the rule_keys your shop sends and from the product fields you chose to turn into keys — on the product and on any of its versions. So a claim keyed on an ingredient that only one flavor contains still reaches the product, because that flavor is part of it — but your AI is told it holds only for the versions that carry the key. It says such a claim only about those versions and names them, never about the product as a whole, and follows such an instruction only for them. A key the product itself carries, or that every one of its versions carries, holds for the whole product.
An instruction for every product goes to your AI on every answer that mentions any product, so keep it short and keep it general — a word that must never be used, a disclaimer that belongs on everything. Anything true of only one substance or one product belongs in a rule keyed to it.

Both sides of a key match are tidied identically:

  1. everything is lowercased;
  2. accents are stripped — Vitamin C, vitamín c and VITAMIN C are one key;
  3. punctuation and symbols become spaces;
  4. repeated spaces collapse and the ends are trimmed.

So " C-Vitamin, (Ascorbinsav)! " and "c vitamin ascorbinsav" are the same key.

Synonyms and translations are not matched for you. A claim 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, which is exactly what the keys field is for.Word order matters too: vitamin c and c vitamin are two different keys.

Two smaller consequences: a key made only of punctuation (—, ()) saves happily but can never match anything, and a switched-on rule needs at least one key — if an instruction genuinely is meant for everything, set it to every product of the brand rather than leaving its keys empty.

Claims mode

Claims mode has no switch. It is on exactly while the brand has at least one switched-on claim, and off when it has none. One claim changes how your AI talks about every product of the brand, which is why the top of the Claims tab always says where you stand:

  • The card at the top — "Claims are on": "The AI describes what your products do only with these claims, and recommends products for a goal only when a claim covers it." Or "Claims are off": "The AI describes your products from their own data. Switch on a claim to limit it to your claims: recommended for food supplements, cosmetics, medicines and other regulated products."
  • A checklist under it — "N claims switched on", "N products have at least one claim" and, while claims are on, "N products have no claim. They still appear in answers, but the AI won't say what they do or recommend them for a goal."
  • Each row of the Claims list shows "Applies to N products" — the active products that carry one of its keys — or "No product has these keys" when that number is 0, which means the claim reaches nothing.

The confirmation

Before any change that would switch on the brand's first claim, you are asked "Switch claims on for this brand?":

"From now on the AI says what your products do only with your claims, and recommends products for a goal only when a claim covers it. Products without a claim are still shown, without saying what they do. Recommended for food supplements, cosmetics, medicines."

Before any change that would switch off, move or delete its last one, you are asked "Switch claims off for this brand?":

"The AI will describe your products from their own data again, including what they do."

Press Switch on or Switch off to go ahead; Cancel changes nothing. The question comes wherever the change is made: a row's switch, a bulk action, the add and edit form, a move between the tabs, a delete, and approving drafts on the review screen.

What your AI does in claims mode

  • It says what a product does only with that product's claims, in the same meaning, never stronger, and with any condition the claim states. A claim about an ingredient is said about the ingredient, not about the product as a whole.
  • It never says what a product without a claim does, whatever its name, category or description suggests. The product still appears in answers, with its facts.
  • It recommends a product for a goal only when one of its claims covers that goal.
  • For a goal no claim covers, it shows no products and says briefly that it cannot recommend one for that. A product the visitor named stays in the answer.
  • Asked what something does when no claim covers it, it says it can't tell. It may still say what it is and what it contains.
  • It never mentions claims or rules to the visitor. Asked where a statement about what a product does comes from, it says it is the shop's approved product information.

With claims off, your AI describes products from their own data, and no claim is sent with them. More on how answers are put together: How Yaplet finds answers.

The claim check

In claims mode, a step called the claim check works out whether the visitor has a goal, and which of the brand's switched-on claims really fit it:

  • One or more claims fit — their products come into the answer, in your shop's "most popular" order, and products none of those claims apply to are left out.
  • A goal, but no claim fits — no product is shown for that goal (help articles stay), and your AI says so briefly.
  • No goal — a question about price, flavor, ingredients or delivery, say — nothing is added or left out.
A product the visitor names is never left out. Ask "Is Thermo Tea good for weight loss?" and Thermo Tea stays in the answer. When it has no claim for that goal, your AI gives only its facts. The products whose claims cover the goal follow.

A few limits worth knowing:

  • Only in claims mode. The claim check uses a little AI, paid for like the rest of the answer.
  • Every product that should come up for a goal must carry one of the claim's keys. When a claim fits, a suitable product without the key is left out.
  • Only keys on active products count. A claim whose keys no active product carries brings in nothing.
  • No import step, no button. A new or edited claim counts from the next product search.
  • Not on phone calls. A voice agent keeps to each product's claims, but on a call they do not bring products in or leave them out. The same goes for questions about your catalog as a whole, and for the answer written after a visitor submits a form.
Never write a claim about what a product contains. The form says it plainly: "Never write a claim about what a product contains ("Contains gluten"): someone asking for gluten-free food would be shown exactly the products with gluten. Put that in an instruction instead." Claims are for what a product is good for, fits or is suitable for.

Writing claims well

  • One claim per rule. Each statement gets its own row, with the keys of the products it is about.
  • The first sentence is the claim; every sentence after it is its condition, which your AI always says together with the claim. For example: "Creatine increases physical performance in successive bursts of short-term, high intensity exercise. The beneficial effect is obtained with a daily intake of 3 g of creatine."
  • A nutrition claim is a plain sentence, such as "The product is fat-free." Keep the legal threshold behind it ("no more than 0.5 g of fat per 100 g") in the rule's Note for your team, which your AI never sees.
  • Put every spelling of the key in one claim, so it matches however your shop writes it.

Adding a claim or an instruction by hand

Press Add claim on the Claims tab, or Add instruction on the Instructions tab. The tab decides which one you write, and the window is called Add a claim or Add an instruction (Edit claim or Edit instruction when you change one).

Which products it applies to
Instructions only: Products that share a key, or Every product of this brand. Choosing every product hides the Keys field — such an instruction carries none.
Keys
required
Press Enter after each. Up to 100 keys, each up to 200 characters. Not shown for an instruction set to every product.
The claim / The instruction
required
Up to 5,000 characters. Write an instruction exactly as it must be followed; write a claim exactly as it may be said, with any condition it needs.
Note for your team (optional)
Up to 2,000 characters. Where it comes from, who signed it off. Never sent to the AI.
Switched on
A switch. A rule you type yourself starts switched on — a person wrote it, so there is nobody left to review it. Switched off, it is saved but your AI never sees it.

Saving the brand's first switched-on claim, or switching off its last one, asks the claims-mode question first.

The two lists

Both tabs show the same kind of list:

ColumnWhat it shows
KeysThe first two keys and how many more. An instruction for every product shows an Every product badge instead.
Claim / InstructionThe text. Hover it to read all of it; click it to edit. The note for your team, if there is one, shows in small print underneath.
ProductsClaims only: "Applies to N products", or "No product has these keys".
OnThe switch. Switching it changes the rule at once.
ByAn icon whose tooltip says who wrote the rule: Written by hand, From a document or By an assistant.

Search the text, or filter by status (On and off, On, Off) and by who wrote it (Any origin or one of the three above). Updated is hidden at first; turn it on from the table's Columns menu.

Each row's ⋯ menu (More actions) offers Edit, Move to Instructions (on Claims) or Move to Claims (on Instructions), and Delete. Select several rows to Switch on selected, Switch off selected, move them to the other tab, or Delete selected together.

An instruction set to every product cannot become a claim, because a claim always reaches products through keys. Its row has no Move, and a bulk move to Claims leaves it where it is — the message says how many stayed.

Each tab also has its own Import from document button and its own Yaplet assistant button in the page header: Check my claims on Claims, which looks for switched-on products without a claim, claims that reach no product and keys that look wrong; Draft instructions with Yaplet on Instructions. See Yaplet and MCP. Yaplet asks before it writes anything, and creates new claims and instructions switched off unless you have checked the exact wording and asked it to switch them on.

Importing from a document

If your claims or instructions already exist as a document — a regulator's list of permitted claims, an internal wording guide — press Import from document (next to the add button) and let the AI turn it into drafts. Any shape of document works: a claims table, a policy, a contract. Importing uses AI, so it runs only when something can pay for it — your free 14 days, bought credit or a paid plan; if nothing can, the payment window opens when you press Start import.

The tab you start on decides what the import writes:

  • On Claims (Import claims) — every statement about what a product does becomes a claim of its own. Warnings, restrictions and other rules that stand on their own become instructions for the same keys, and statements the document makes about every product become instructions set to every product of the brand.
  • On Instructions (Import instructions) — everything becomes an instruction.

Everything arrives switched off, as a draft.

Say what the document is

What is this document?
Let the AI decide (the default), A list of permitted claims — nothing outside it may be said, Restrictions, conditions and warnings, or Something else.

::

This is a hint, never a switch that changes how the file is read. Left on the default, the AI judges from the text itself — and the finished screen tells you what it decided.

Pick the model

Where your plan lets you choose the AI model, an AI model dropdown sits in the window; without that right it is hidden and the import uses the default model. The dropdown starts on your organization's everyday model. The longer or more intricate the document, the more it is worth choosing a stronger one: a dense claims table with footnotes and conditions deserves a top model, a short plain list is fine on a cheaper one.

The size limit follows your choice and is shown right under it — "Gemini 3.1 Flash Lite: up to about 957 pages (670,574 tokens) in one import." If your document is longer than that, split it yourself and import each piece.

Upload it

PDF, Word or plain text, up to 30 MB.

A scanned PDF holds pictures of text, not text. It cannot be read, and you will be told so before anything is charged.

Add instructions (optional)

The Instructions for this document (optional) field is for anything the AI should know about this particular document — how it is laid out, which parts to ignore, what the keys should be. For example: "The first column is the ingredient; ignore the preamble."

Let it run

Press Start import. The window reports each stage: Mapping the document, then Reading N sections — M done, then Merging and writing drafts, with a running count of the statements found and how many of them the check added.

You can close it (Leave it running) — the import keeps running, and coming back to the tab of its type reopens the progress window on the run.

Review the drafts

The finished screen says what it covered: "N drafts created", "N statements found in the document, M of them added by the check.", and how it read the file — "Read as a list of permitted claims · language HU". Press Review drafts to go through them. Nothing is live until you approve it. ::

How the document is read

The AI does not blindly chunk your file. It first maps the whole document and divides it into sections along its own content, reads every section separately, and then checks each section again for anything it missed, merged into one sentence, or worded wrongly — that second pass is where the "added by the check" number comes from.

The whole document still has to fit your model's window once, for that first mapping pass. That is what the size limit measures.

What an import produces

  • Import claims: one claim per statement. A list of permitted claims becomes one claim for each statement, word for word, so your AI can use exactly the one that fits. A condition, warning or consumer information the document attaches to one statement stays inside that claim's text. A restriction, condition or prohibition that stands on its own becomes an instruction for the same keys, so it is always followed — not only when a claim is used.
  • Import instructions: one rule per thing. Every statement the document makes about one substance, product or topic is gathered into a single rule. For a list of permitted claims it opens with the exclusivity sentence — "only the following statements may be made about X, and nothing beyond them" — then lists each statement verbatim and numbered, and states a condition they all share once at the end. Restrictions, conditions and prohibitions about the same thing are folded into its rule. A lone statement in any other kind of document becomes a rule on its own.
  • What the document says about everything is written once, as an instruction set to every product of the brand — so it also reaches products the document never names, and products carrying no keys. If the document really governs only part of your catalog, edit that draft to "Products that share a key" and give it the keys it should have.
  • Keys carry synonyms automatically. The import adds the other names for the same thing — common names, abbreviations, chemical forms and salts, spelling variants, and the English name when the document is not in English. A Hungarian rule keyed on Folát also carries Folsav, B9-vitamin and Metilfolát. Rules you type yourself are still yours to spell out, as the warning above says.
  • Footnotes that limit a claim — licensed to one company, valid only until a date — go into the note for your team, never into the wording your AI reads.

When an import does not finish cleanly

Two things can happen, and neither loses what was already written:

  • Some sections could not be read. The import still finishes, and an amber warning names the parts that were missed. Everything else is kept; import the file again to get the rest.
  • The run was interrupted — something stopped it halfway. The window reads "The import did not finish" and offers Resume (or Start over), which continues from the sections that never finished and never redoes what was already written. The tab reopens that window on its own when you come back to it.

The approval gate

Nothing an AI produced is ever live until a person switches it on. Claims and instructions written by a document import, and those written by the Yaplet assistant or another assistant, all arrive switched off. A switched-off rule is stored and visible in the dashboard, but it is never read when an answer is written — on any channel.

When a tab has switched-off rows, an amber banner above the list says so — "N switched-off claims waiting for review" — with a Review button. The review screen is where you go through them:

  • On a wide screen, the Imported documents list on the left lets you pick a document to review just its drafts, or click it again to review every draft at once. Each document is marked Claims or Instructions, and a row whose type differs from the title carries its own Claim or Instruction badge.
  • Each row shows its keys, its text and its team note.
  • Approve a row with its button, or use its ⋯ menu for Edit and Discard. Select several rows to use Approve selected, Approve all in this import or the Discard menu in the footer. Approving switches the rule on.
  • Discarding shows an Undo in the toast for 15 seconds. Undo puts the drafts back with their type — a claim comes back a claim.

A switched-on rule matched by keys needs at least one key; approving a batch that contains rules without one fails and names them.

Re-importing the same document

Import an updated version of a file you imported before on the same tab, and the review screen compares the two automatically. Claims are compared only with an earlier claims import of that file, instructions only with an earlier instructions import:

  • New — a rule with keys that had none before.
  • Changed — same keys, different wording, shown with the old text struck through. Approving it deletes the wording it replaces, so you can never end up with both live.
  • Rules that were identical are dropped silently — there is nothing for a person to look at.
A re-import never deletes anything. It only ever offers you drafts to approve, so your existing rules are safe no matter how the comparison turns out. Two rules are matched by their keys, and the AI writes its list of alternative names slightly differently each time it reads a document — so a rule you already approved can easily show up as New rather than Changed, even when the document has not changed at all. That is harmless when the worst case is an extra draft to approve; it would not be harmless if the same comparison were allowed to delete your work.If a document genuinely drops a claim and you want the old rule gone, delete it yourself from the list — search for it, tick it, and use the bulk delete.

Where claims and instructions are applied

Every answer a visitor can see:

  • the chat widget
  • email
  • Facebook and Instagram
  • AI reply suggestions your agents see in the inbox
  • phone calls

They are not applied to things that never answer a visitor: the inbox panel that shows an agent what the knowledge holds, agent-monitoring checks, and AI social-post generation.

What happens if there are too many

The whole block of claims and instructions for one answer is capped at 100,000 characters. At the cap, nothing is ever trimmed — instead the lowest-ranked products leave the answer whole, taking their passages, their product cards and their claims and instructions with them, until the rest fit in full.

The same applies if the rules cannot be read at all: the products leave the answer. Your AI is never allowed to describe a product while missing some of its claims or instructions.

You will not meet this ceiling in normal use — a product's own usage is shown on its page in the catalog.

© 2026 Yaplet. All rights reserved.