How to measure AI visibility

Our proprietary measurement system helps you understand how a domain appears in sampled AI answers. Read the scores alongside the questions, cited sources, and measurement date to put each result in context.

Two scores, two perspectives

AI score shows where a domain stands among the sites mentioned in the same sampled answers to discovery questions that do not name the brand. It helps you explore whether the site appears when a question starts with a need, category, or use case. Each discovery question asks for options to choose from, such as the best tools for a need, the providers in a place, or what to use for a task, the way a customer of that kind of site would ask. The site the answers mention most scores 100, and every other site scores by how close it comes to that one. A measured site never scores 0.

Brand score summarizes visibility in answers to questions that explicitly name the brand. It provides a separate view of what happens when the question already identifies the company. It does not rate a company’s quality or customer satisfaction.

Both scores use a scale from 0 to 100, and they measure different things. For the AI score, a higher number means a smaller distance to the site the answers mention most. For the Brand score, a higher number means greater measured visibility when the question already names the company. The two scores describe different situations and should be read separately.

Reading the AI score

To say at a glance whether an AI score is good or bad, the console reads it in four bands. They are a reading aid on the 0 to 100 scale, not part of the measurement, and they describe the distance to the site the answers mention most.

• Low, 0–24: AI barely mentions you.
• Fair, 25–49: The site AI mentions most is far ahead of you.
• Good, 50–74: The site AI mentions most is ahead of you.
• Strong, 75–100: You are the site AI mentions most, or close behind it.

A score of 100 means your site ranks first in its field, alone or tied.

A site the sampled answers have not mentioned yet scores a few points above 0, within the Low band, and its score can move inside that band before the first mention. A score above the Low band means the answers mention your site.

The band is judged on the score shown, and a score near a boundary can sit in the neighboring band on another day.

Look at the evidence

The extension shows discovery questions and cited source domains from the measured answers. Use the questions to understand the context being examined, then explore the sources to inform your research. These are measurement questions; they are not a record of your customers’ conversations.

A citation shows that a source appeared in an answer. It does not establish why the engine selected that source or whether the answer led to a website visit.

The Actions tab lists next steps derived from the measurement and your site’s public files; see What raises AI visibility. Whenever your AI score is below 100, it lists at least one step.

Where you stand in the field

Next to the competitor list we show your rank and mention share. Mention share is your slice of all the mentions of competing sites in the same set of answers, and every site, yours included, is counted by the same rule. Reference sources such as encyclopedias do not compete for the answer and are left out. The set of answers is the last seven days of sampled answers; in the console it is the period you choose. We do not rank sites by where an answer places them, because that order changes from one answer to the next. Where two sites appear in the same answers, the console shows how often each comes first. Rank and share are estimates from sampled answers, not a ranking the AI provider publishes.

The AI score is read from the same sampled answers and the same ranking as the rank. The site ranked first scores 100. Sites tied for first all rank first and all score 100. Every measured site has a rank: a site the answers never mention shares last place, for example 14th of 14. Only when no site is mentioned in any of the answers is there nothing to rank against, and no rank is shown. The score also reflects more than the ranking does, so a site with a lower rank can show a higher score than a site above it.

The field is the set of sites the ranking counts. For a site in one of our markets, it is every site named in the answers collected in that market. Other sites are ranked among the competing sites named in the answers to their own questions.

For analysis, the console also shows how answers present a site: whether it is linked, named, or only read; which third-party page is cited beside it; and whether the answer recommends it, lists it, compares it, or speaks against it. These readings are made by a model and can be wrong. We measure how often they match people’s own labels and publish the result here once the labeled set is complete. The console also shows how often the engine’s search returned one of the site’s pages, how often the answer showed or recommended the site, and how often a later answer to the same question showed it again. None of this enters the AI score or the mention share.

Domains we have confirmed belong to your organisation count as your site: a citation of one of them is a citation of yours, and they never appear as your competitors.

The language of the questions

A site is measured in its own language. English sites are measured on the shared questions of their market. A site in another language is measured on questions written in that language for that site, including the questions that name its brand, and it is not placed in a market. The panel says so, for example “Measured with Japanese questions”. Scores mean the same thing in every language, a place against the site the sampled answers mention most, but the answers come from different conversations, so we never rank sites across languages. Every search states the United States as the place it is made from, and each measurement records that place, so results from different places are never mixed.

Know what is covered

Current measurement uses OpenAI API responses with web search. Results describe sampled answers when the engine searches the web. Daily tracking in the console can also sample Google AI Overviews and AI Mode, read through a SERP data provider without personalization, in the site’s language. These answers are shown per engine and do not enter the AI score. A question that shows no AI Overview is counted toward how often the overview appears, not as an answer that left the site out. They do not estimate visibility across all AI conversations, including answers produced without search.

Measurements stand for an anonymous person with no history: they do not use any chat history, account memory, custom instructions, or the location of the person who asked for them. Answers in a consumer app may differ because of personalization, wording, settings, or changes to the underlying service. Other AI providers are outside the current measurement scope.

Read each result in context

Check the measurement date shown with the score. A result reflects the available measurement, rather than a live view of every AI answer. The score is an estimate from a sample of answers, so small differences between scores may not be meaningful. The console marks a score up or down only when the change is larger than the usual swing between answers.

In October 2026 we changed how discovery questions are written, so that every one of them asks for options to choose from. A score measured with the new questions is not directly comparable with one measured before the change. When the questions change, the console connects a site’s history across the change where it can, and marks a break where it cannot. Competitor rankings in the console use the current questions only. The score in the extension always rests on one set of questions.

Sites in the same market are measured on questions drawn from one shared set of discovery questions. Markets follow Google’s content categories; sites outside the markets we cover are measured on questions written for that site.

Because the AI score is a place against the other sites in the field, your score can change when the field changes, even if your site does not. A new competitor, or a leader that AI starts mentioning more or less often, moves every site below it. Read a change in the score together with your rank and the measurement date.

On 26 October 2026 the AI score changed. Before, it described how often answers mention a site on its own. Now it is the site’s place against the site the answers mention most. Scores measured before that date were restated on the new scale from the answers we stored, read the way we read them today, so a site’s history is one line. A number from an earlier screenshot or export is not comparable with a new one.

On 1 November 2026 the AI score changed so that a measured site no longer scores 0, and a site the answers have not mentioned yet can still move within the Low band. The same day, the questions that name an English site’s brand began to name the site’s category beside it, so that a name several companies share is not answered with a question back. Earlier scores were restated the same way from the answers we stored, read against the latest check of each site.

AI answers change. A score cannot guarantee future citations, rankings, or traffic, and a low result does not prove that a site never appears.

We operate an independent measurement service. Scores are our estimates and are not official rankings or endorsements from an AI provider.

Questions about a result

If a result is unclear, contact our team with the domain and measurement date. Our privacy policy explains the information the extension sends and how it is handled.

Acknowledgements

Data sources include WildChat-1M from the Allen Institute for AI, under ODC-BY, and LMSYS-Chat-1M, under the LMSYS-Chat-1M Dataset License Agreement. Market names come from Google’s content categories, under CC BY 4.0.