AnswersAI visibility, defined

What AI visibility is, what it is made of, and how it is measured

AI visibility is where a brand stands when a buyer asks an AI assistant what to buy or who to use. A chat assistant answers with a handful of names, so what matters is whether you are among them, whether you are the one recommended, whether you are described accurately, and whether the buyer is told where to get you.

What is AI visibility?

AI visibility, also called AI search visibility, is where a brand stands in the answers AI assistants give when a buyer asks what to buy or who to use. It covers whether the brand is named, whether it is the one recommended, whether it is described accurately, and whether the buyer is told where to get it.

It is a position in answers rather than a volume of traffic. The unit is the answer itself: which brands it names, in what order, what it says about each, and where it sends the buyer. A buyer steered to a rival inside an assistant never reaches your site, so nothing in the reports you already run shows it. What changes from SEO reporting sets out the rates to report in place of a rank.

Why does AI visibility matter commercially?

AI visibility matters commercially because a chat assistant answers a buying question with a handful of names and the comparing already done, so a brand left off that list is not among the options the buyer is given. A brand named and passed over is in the answer while a rival takes the recommendation.

A page of search results leaves the comparing to the buyer. An answer from a chat assistant arrives with it done: a few names, in an order the assistant chose.

Being named and passed over is easy to mistake for success, because the brand is visibly there, which is why being mentioned and being recommended are measured as separate things.

Which assistant a buyer asks changes who they are told to buy. In Delphi's study of 3,116 answers, the most recent collected on 24 August 2026, there were 166 questions where at least three assistants named a company, and on 11 of them every one of those assistants named the same company most often. A brand recommended in one assistant can be passed over in the next.

What is AI visibility made of?

AI visibility is made of five things, in the order a buyer meets them: whether an assistant can read the brand at all, whether it names the brand when asked about the category, whether it describes the brand accurately, whether it recommends the brand rather than a rival, and whether the answer ends somewhere the buyer can act.

Readable. Whether an assistant can reach your own pages and read the facts on them. Delphi checks which AI crawlers your robots.txt allows, and whether key facts are in the page as delivered rather than appearing only after a script runs. Which AI crawlers to allow covers the settings.
Named. Whether the brand turns up when a buyer asks about the category, and on how many of the questions asked. A common figure for this part is share of voice: the share of all the brand mentions in those answers that are yours.
Accurate. Whether what the answer says matches what the brand publishes, and whether it speaks of the brand favourably. If an answer tells a buyer a product is discontinued, that sale does not happen, whichever side is out of date. When an answer contradicts your own page covers how that is read, and when AI gets your company wrong covers the other ways an answer can describe a company inaccurately.
Recommended. Whether the brand is the one the answer puts forward, or one name among several while a rival is chosen.
Somewhere to act. Whether the answer tells the buyer where to get it, gives a concrete offer rather than a bare name, and raises nothing that stops the purchase, such as a claim that it is unavailable.

Delphi's plan starts at the weakest of the five, unless an assistant's search crawler is refused: opening it comes first, because until then it cannot read what you change on your own pages. What to fix first sets out why, and why a competitor shows up first applies the same five to a rival recommended ahead of you.

How is AI visibility measured?

AI visibility is measured by putting the questions a brand's buyers ask to several AI assistants, more than once each, and reading every answer for what it says about the brand. Answers vary from one asking to the next and from one assistant to another, so an honest result is a range rather than a single number, and it names the assistants it covers.

Delphi puts each question to ChatGPT, Gemini, Claude, Perplexity and Grok more than once, through each provider's own API, as a fresh question with no earlier conversation, so no one person's history shapes the answer. Every answer is stored word for word beside the model that produced it.

In Delphi's own study, whose most recent answers were collected on 24 August 2026, an assistant asked one question more than once inside a single measurement returned the same set of names 21.5% of the time.

The Delphi Index is published with a confidence band drawn from how far the assistants disagreed with each other. How many questions a measurement needs covers how that band is made and when a change counts as a result.

Delphi is not a daily tracker. Each brand can be measured once a month or twice a month, depending on the plan, and an extra measurement costs $49. With answers varying this much, a daily reading taken from a few answers would mostly record that variation rather than a change in position.

How do you improve AI visibility?

Three kinds of work can improve AI visibility: letting the search crawlers that feed assistants' answers read your site, putting the facts buyers ask about on your own pages, and being described accurately on the other sites assistants draw on. The first is a setting in your robots.txt file; the third depends on sites you do not control.

For its own AI features, Google says ordinary SEO is enough: no special optimisation is needed to appear in AI Overviews or AI Mode, as long as the site has not been excluded from Search's generative AI features in Search Console, which includes every site by default. To be shown as a supporting link, a page has to be indexed and eligible to appear in Google Search with a snippet. The work across assistants is set out under answer engine optimisation, also called generative engine optimisation, and what is specific to each assistant, from its crawler to its ads, in the answers for ChatGPT, Gemini, Claude, Perplexity and Grok.

Where should a team start with AI visibility?

A team should start by asking the assistants its buyers use the questions those buyers ask, in a fresh chat, and writing down who gets named: it costs nothing and shows whether there is a problem. Measure repeatedly once the team needs to know whether something changed, and let the weakest part of the result decide where the work starts, once any assistant's search crawler the site refuses has been let in.

A check by hand shows whether you appear at all. It cannot show whether anything moved: an assistant asked the same question twice usually returns a different list, so two checks by hand can differ with nothing having changed. How to track it over time covers what to record so that a change can be read. How to find out what ChatGPT says about you sets out the free version, and what to fix first sets out how a measurement orders the work.