How answer engines decide which brands to mention.

Every AI-generated answer is assembled from two kinds of knowledge. The first is what the underlying model learned during training: a broad, compressed picture of the web as it was at a point in time. The second is what the system looks up when the question is asked — pages fetched from a search index or the live web, then used to ground the response.

Different products lean on these two sources differently. Perplexity and Google AI Overviews retrieve sources for almost every answer and show them as citations. ChatGPT and Gemini retrieve when a question needs fresh or specific information, and otherwise answer from what the model already knows. The balance changes often as the products evolve, so it is safer to plan for both than to optimise for one.

When an engine retrieves, it usually rewrites the user's question into one or more search queries, pulls back a set of candidate pages and reads the most relevant passages. A brand that is never in that candidate set cannot be mentioned from retrieval, however good it is.

The basics of being found are familiar from SEO, with a few differences in emphasis:

• The page has to be crawlable and render its main content without depending on heavy client-side scripts.
• AI crawlers need to be allowed in robots.txt — some sites block them by accident through blanket rules.
• The answer to a likely question should exist as text on a page, not only inside an image, PDF or video.
• Pages need to be specific. A single generic services page is a weak match for a precise question like "emergency boiler repair open on Sundays in Leeds".

Once a page is retrieved, the model has to work out what it says and who it is about. This is where entities come in. An entity is a uniquely identifiable thing — a business, a person, a product, a place — that systems can attach facts to.

Brands that are easy to understand state their key facts plainly and consistently: what they are, what they do, who they serve, where they operate and what makes them different. Structured data (schema.org markup in JSON-LD) repeats those facts in a machine-readable form, which removes ambiguity.

Answer-first writing helps too. Passages that open with a direct, quotable answer to a specific question are easier for a model to lift into a response than paragraphs that circle the point.

Models are designed to avoid repeating claims they can't support, so they favour information that appears in more than one credible place. A brand's own website saying it is "the leading provider" carries little weight on its own. The same facts repeated on independent sources — industry directories, review platforms, local press, partner pages, community discussions — make the brand far safer to mention.

Consistency matters as much as volume. If a business's name, services, address or opening hours differ across the web, engines have a harder time deciding that all those references are the same entity, and confidence drops.

With a set of found, understood and trusted candidates, the engine writes its answer. It typically mentions a small number of brands, chosen for how well they match the specific intent of the question — location, budget, specialism, audience — and how clearly the sources support that match.

This is why the same brand can appear for one phrasing of a question and vanish for another. A clinic that is clearly described as "family dentistry" may appear for family-related prompts and be absent from prompts about cosmetic work, even if it offers both. Visibility follows the evidence, prompt by prompt.

No one can guarantee a mention in an AI answer, and you should be wary of anyone who says they can. What an agency can do is remove the reasons a brand is left out. In practice that work falls into five areas:

1. 01Retrievability — crawl access, rendering, site structure and page speed, so the content can be fetched and read.
2. 02Content engineering — specific, answer-first pages that match the questions buyers actually ask.
3. 03Entity clarity — consistent facts across the site and structured data that states them explicitly.
4. 04Corroboration — consistent, credible mentions on independent sources the engines already trust.
5. 05Measurement — a defined set of prompts tracked over time, with the cited sources recorded for each answer.

Most of this is good marketing practice that predates AI search. What is new is the measurement: tracking prompts instead of keywords, and paying attention to which sources the engines cite when a competitor is chosen instead.

Start with a baseline. Write down the questions a real buyer would ask an AI assistant at each stage of their decision, run them across the engines that matter for the market, and record whether the brand appears, how it is described and which sources are cited.

Repeat the same set on a regular schedule. Answers vary from run to run, so look at trends across the whole set rather than reacting to a single answer. Useful measures include prompt visibility (the share of prompts where the brand appears), share of answer against named competitors, and the sources most often cited in the category.