How Yaplet finds answers

What actually happens between a visitor's question and Yaplet's reply — a search across everything the brand owns, in every language it is written in, with prices read live and links built fresh. Plus what to check when Yaplet ignores something you added.

You do not configure any of this. It is worth understanding anyway, because it explains why an answer came out the way it did, and what to change when it comes out wrong.

A search over everything the brand owns

When a question arrives, Yaplet runs a search across everything attached to that brand — its knowledge bases, its documentation, its uploaded files, the website pages we have read for you, your question-and-answer pairs and your product catalog. There is no per-source setting deciding what is searched; if it is on Brand → Knowledge and it is indexed, it is in scope.

In the chat, a brand that has both help content and products gets them searched separately, at the same time. That way a large catalog cannot crowd your help articles out of an answer, and the other way round.

The same content backs every place the brand's AI speaks:

  • the chat widget
  • phone calls taken by its voice agent
  • the suggested replies your team sees in the inbox
  • answers assembled from your API tools
  • the AI social posts it writes

That is why adding one article improves all of them at once, and why a gap shows up everywhere at once too.

Scope stops at the brand. One brand's AI cannot read another brand's content, even inside the same organization, and every result is checked against your own organization before it is used. Two brands that need the same article need it in both — see Copying and moving content between brands.

One search per topic, and nothing skipped

Every message that is not a plain greeting, thanks or goodbye is searched before Yaplet answers — a complaint, an offer or a question that is not phrased as a question all get looked up, so Yaplet never decides from the wording alone that it "does not need" your content.

When one message asks about several things, Yaplet looks each one up separately and merges what it finds. It answers the topics your content covers and says plainly which ones it has nothing on, instead of guessing at the rest. Only when it finds nothing at all does the "When Yaplet has no answer" setting in the Human hand-off section of AI agent settings apply.

At the Deep search depth, Yaplet also searches the main topic with two more wordings of the question, which catches content that is phrased differently. The level is chosen under Brand settings → Advanced; Brand settings describes the two levels.

Only your own content, never general knowledge

Yaplet states only what your own content says: your articles and documents, your product data and your claims. It never adds facts from the AI model's general knowledge — not about your company, not about your products or their ingredients, and not about the visitor's topic either. This is the same for every shop and on every channel: the chat, email, Facebook and Instagram, the inbox's suggested replies, form answers and phone calls.

When a visitor asks what something is or how it works — an ingredient, a term, a process — Yaplet answers only with what your own data says about it. When your data says nothing, it says it has no more detail on that.

If visitors often ask what something is and Yaplet has no answer, the fix is content: add a short article or a Q&A entry that explains it in your own words.

Several languages at once

Yaplet searches in every language the brand's sources are written in. A brand holding an English knowledge base and a Hungarian one keeps full accuracy in both at the same time: a Hungarian question still matches the Hungarian articles properly.

The brand's own language is the fallback for anything that carries no language of its own: uploaded files, website pages and Q&A entries.

Searching in several languages does not mean replying in several languages. Yaplet answers in the language the visitor wrote in, whichever language the source it found happens to be in.

When Yaplet points a visitor at one of your articles, it does not repeat a link that was stored when the article was indexed. It looks up the brand's web address at the moment it answers and builds the link then.

The practical consequence: changing your public web address never leaves a stale link buried inside an AI reply. Links people have already saved elsewhere still break — that part is unavoidable, and Your public web address explains it — but nothing Yaplet says from that point on is wrong.

Catalog questions do not go through search at all

Some questions are not about a product — they are about your catalog. "What do you sell?", "What kinds of protein do you have?", "What are your top three vitamins?" name nothing that a search could match.

For those, Yaplet asks your catalog directly for its shape instead of searching:

  • with no category, it gets your top-level categories with a product count for each;
  • with a category, a condition or a sort, it gets the matching products in one order called "most popular" — products with a priority first, in priority order, then the rest by popularity, out-of-stock products last. The first ones come with live prices and stock, at most 5 per answer. When the visitor asks for more, Yaplet gets the next 5.

Because this is a lookup rather than a search, the answer is built from your real categories and your real order, not from whichever handful of products a search happened to surface. It works on every text channel and on phone calls, whenever the brand has an active product.

Filtered and sorted lists

The same lookup can narrow and sort, for one category or for the whole shop. A question with a condition or asking for a particular order goes this way too:

Conditions
A price limit (at least, at most, or a range — what the visitor pays, sale price included), in stock, on sale, a flavor, size or color, or a word in the product's data such as vegan or lactose-free. A product matches when one of its versions meets every condition together — a pistachio version that is in stock and under the price. If a flavor or size matches nothing, Yaplet tries again with the closest of your shop's own option values.
Sorts
Every list comes most popular first unless the visitor asks for another order. On request: highest rated, newest, cheapest and most expensive. Each is built from your own data — the ratings with review counts and listing dates your feed sends (see The Product API), and your prices. Where that data is missing, the list says so and Yaplet does not make the claim. If none of the matching products carries a priority or a popularity, the list is simply alphabetical and Yaplet calls none of it most popular.
Best or top products
"What are your best products?", "your top three vitamins?", your best-sellers or what is trending — with no goal named — is answered with the most popular order, and Yaplet may present those products as your most popular ones — never as "the best". A goal ("the best for weight loss") makes it a recommendation instead, which needs the product's own data — or, when claims are on, one of its claims (see below).

A product whose every version is out of stock always goes to the end of every list, so a "top three" never opens with something nobody can buy.

Yaplet uses this route only for questions that name no specific product. A question naming something it does not recognize — a nickname, a model name, an unfamiliar word — is searched instead, because the search knows your products' other names.

Prices and stock are read live

Product prices, sale prices and availability are not stored with what Yaplet has learned. They live on each version of a product and are read from your catalog at the moment the answer is written.

This is why the sentence and the product card can never disagree, and why changing a price costs nothing: the number was never part of what was indexed, so nothing has to be re-indexed when it changes.

How a product answer is shaped

Yaplet uses at most 5 products per answer, and it names or links only those. Product answers stay short unless the visitor asks for detail or a comparison, and they end by helping the visitor choose. Dosage, allergens and warnings come up only when the visitor asks about them. To see how this sounds with your own products, try it on the Playground.

This is the platform's default. Your own claims and product instructions and the agent's instructions win wherever they say otherwise.

What Yaplet is and is not allowed to say about a product

Whenever an answer contains products, Yaplet keeps to a fixed set of promises. You do not configure these and you cannot switch them off. Your own claims and product instructions and agent instructions sit on top of them.

  • It names only products it was actually shown, and never a product type you do not stock.
  • It describes a product only from that product's own sheet. If a visitor asks about something the sheet does not cover, Yaplet never guesses.
  • It adds no praise or ranking of its own.
  • It claims popularity only from your own order. A list in the "most popular" order may be presented as your shop's most popular products. It never calls a product the best. Highest rated, newest or cheapest only when your ratings, listing dates or prices say so.
  • It compares with measurable facts only: prices, amounts, contents. It never names a winner.
  • It says nothing about other shops' or brands' products, and never talks down a product you sell.
  • It promises nothing your data does not state, such as urgency, discounts or when something comes back in stock.
These exist because an AI that ranks your products, invents a quality ladder or claims something is popular is not just imprecise — it is making commercial claims on your behalf that you never approved and cannot stand behind.
On phone calls the same promises apply, and answers are shorter because they are spoken.
Your products' other names count. A product's sheet lists its other names — search terms, nicknames, old names — so a visitor who uses one of them still gets the product. Yaplet says something is unavailable only when the search returned nothing.

Claims and instructions override everything else

If a product in the answer has claims or product instructions, Yaplet treats them as binding wording, so an article can never squeeze one out.

They override the content it found, override anything said earlier in the conversation, and override what the visitor asks for. Every switched-on instruction and claim that reaches a product travels with it, and switched-off ones never reach Yaplet.

  • An instruction is followed every time Yaplet talks about the products it applies to.
  • A claim is what a product may be said to do, be good for or be suitable for — and it also picks the products for a visitor's goal.

When claims are on (claims mode)

A brand is in claims mode while it has at least one switched-on claim; the Claims tab then shows Claims are on. There is no separate switch: switching on the first claim starts it, switching off or deleting the last one ends it, and the dashboard asks you to confirm both. In claims mode Yaplet:

  • says what a product does, is good for or is suitable for only with that product's claims — in the same meaning, never stronger, and with any condition the claim states. A claim that names an ingredient is said about the ingredient;
  • still shows products that have no claim, with their facts, but without saying what they do;
  • recommends a product for a goal only when one of its claims covers that goal;
  • for a goal that no claim covers, shows no products — a product the visitor named stays — and replies briefly that it can't recommend a product for that, offering help with something else;
  • asked what an ingredient or a product does when no claim covers it, says it can't tell — it may still say what it is and what it contains;
  • never mentions claims, rules or approvals to the visitor. Asked where a statement about what a product does comes from, it says it is the shop's approved product information.

One exception: the inbox's suggested replies do not follow all of this, because a person reviews each one before it is sent. They still get each product's claims.

In claims mode, the claim check also decides which products come up for a visitor's goal: the claims that really fit it bring in the products carrying their keys, and other products are left out. A brand with no switched-on claim skips it, so it costs nothing extra. It does not run on phone calls, when a visitor asks about your catalog as a whole, or in the answer written after a visitor submits a form. The claims and product instructions page explains what belongs in a claim and what belongs in an instruction.

If there is more than one answer can carry, whole products leave the answer, together with their claims and instructions, rather than being described with only some of them.

Swapping content leaves no ghosts

If you take a knowledge base down, swap another one in, or move a source to a different brand, the old content stops answering immediately. It is not remembered, and it does not keep surfacing in replies after it has gone.

Why isn't Yaplet using this?

When something you added is not showing up in answers, work down this list. The sources listed on Brand → Knowledge answer most cases on their own: whether each one is published, and whether it is indexed.

Is it published?

An unpublished article or page is a draft: not shown to anyone, and not indexed. Publishing it indexes it straight away. Knowledge base categories have no Published switch — only the articles inside them do.

Does the source show pieces indexed?

Each source says how many searchable pieces have been stored from it, for example "12 pieces indexed". A source with nothing stored says Not indexed yet (or Not indexed), so there is nothing for Yaplet to find. The Indexed figure at the top of Brand → Knowledge adds up all your sources. A large article normally becomes several pieces.

Is the source on the right brand?

Check you are looking at the brand whose AI you are testing. Content added to one brand is invisible to the others.

Is learning paused, or at its limit?

Yaplet learns new content only while someone can pay for it: a paid plan, credit you bought, or the free 14 days. If none of these applies, nothing new is learned or refreshed until you add credit or choose a plan; what Yaplet has already learned stays. Older free workspaces that still have monthly free answers share 250 searchable pieces across the whole workspace, and the count is shown in the Indexed figure at the top of Brand → Knowledge, as "X of Y on the free plan". If a source says You've reached your knowledge limit, write to us.

Is the question actually about that text?

Open Brand → AI agent → Reports, find the answer in the Answers list and open its Details. Under Steps, the knowledge-search step shows the exact query Yaplet ran and what it matched, with a relevance score. That is the fastest way to tell "nothing matched" apart from "the wrong thing matched" — and the two need different fixes.

Reports is the screen to live in while you are tuning this. Filter the Answers list by the No answer and Agent handoff outcomes — the rows badged Failed and Handoff (no context) are precisely the questions your content does not cover yet.

© 2026 Yaplet. All rights reserved.