How to Measure Visibility in AI Search

AI search visibility measurement is a practice that tracks how a brand appears in Google's AI Overviews and AI Mode, and in answers from assistants like ChatGPT, Gemini and Perplexity. It draws on five sources: Search Console reports, Google Analytics 4 (GA4) data, server logs, manual tracking and measurement tools.

The subject here is measurement, not the definition of generative engine optimisation (GEO). I summarised Google's position in the AI in Search section of my SEO guide.

Why measure visibility in AI search?

Measuring visibility in AI search matters because an AI answer can show a brand without a click. In an answer a brand is named in the text, linked as a source, or absent. Classic rank tracking does not separate those three states.

Google's documentation says AI Overviews and AI Mode may use "query fan-out", which issues multiple related searches across subtopics. Your page can be a source for a generated subquery, not for the query the user typed. Measurement therefore covers a set of questions, not a single keyword.

What can and cannot be measured in AI search visibility?

AI search visibility can be measured as impressions, clicks, referral traffic, bot requests and a sampled mention rate. Five data points are measurable.

• Impressions in Google's generative AI features, by page.
• Clicks from AI Overviews and AI Mode, inside Search Console's total Web data.
• Sessions from AI assistants, in GA4.
• Requests from AI bots, in server logs.
• The share of answers that mention the brand, in a fixed question set.

Three data points cannot be measured.

• The total number of answers that name the brand without a link: with no click, nothing reaches GA4 or the server log.
• A separate click total and a query breakdown for AI features in Google's reports: Google's documentation describes neither.
• The answer each user sees: manual tracking and tools return a sample, not a full count.

How do you measure Google's AI features in Search Console?

Search Console measures visibility in Google's AI features with two reports: the Performance report and the generative AI performance report.

How does the Performance report count AI features?

The Performance report counts sites that appear in AI Overviews and AI Mode within the Web search type. Google states this in its AI features and your website document. Search Console Help lists no Search appearance type for AI features and gives three counting rules.

• Click: a click on a link to an external page in an AI Overview or AI Mode is counted.
• Position: an AI Overview occupies a single position, and every link inside it gets that position.
• New query: each follow-up question in AI Mode counts as a new query.

I covered querying performance data with SQL in my post on the Search Console BigQuery export.

What does the generative AI performance report show?

The generative AI performance report shows how many times links to your site were shown in AI Overviews and AI Mode. In the help page for the report, Google says the data reached all websites worldwide as of 31 August 2026. The report groups data in four tabs: pages, countries, dates and devices. The report has three limits.

• The chart is aggregated by property: two links from your site in one answer count as a single impression.
• The 1,000 row limit of the Performance report applies to this report too.
• The report may be missing if the site has too few impressions or was excluded with the Search generative AI control under Settings.

How do you separate AI assistant traffic in GA4?

GA4 separates AI assistant traffic with the AI Assistant channel in the default channel group. Google's default channel group documentation defines the channel as users who arrive from sources like ChatGPT, Gemini, Deepseek, Copilot or Grok. When the referrer matches Google's list of AI assistants, GA4 sets the session medium to ai-assistant. The channel excludes AI Overviews and AI Mode.

The channel appears under Reports, Acquisition, in the Traffic acquisition report. Change the dimension to "Session source / medium" to see which assistant sent the visits. OpenAI says ChatGPT automatically adds the utm_source=chatgpt.com parameter to its referral URLs. Read the source values of other assistants from your own report.

The default channel group cannot be edited. To set the assistant list yourself, build a custom channel group in four steps.

1. In Admin, open Channel groups under Data display and click "Create new channel group".
2. Click "Add new channel" and name the channel "AI assistants".
3. Add a condition on the Source dimension, choose "matches regex" and enter the regular expression.
4. Move the channel above Referral in the list and save the group.

The regular expression below is an example based on the assistants named in Google's custom channel groups documentation.

.*(chatgpt|openai|perplexity|copilot|gemini|claude).*

Custom channel groups apply to reports retroactively. Google asks you to update the expression when the assistant list changes.

How do you tell AI bots apart in server logs?

AI bots in server logs fall into three classes by user agent name: training bots, search bots and user-triggered bots. The table below lists the bot names that OpenAI, Anthropic and Perplexity document.

A training bot request means the content may be collected for model training, not that it appeared in an answer. A search bot request means the page was crawled for the assistant's search results. A user-triggered bot may visit a page when a user asks the assistant a question, so it is the closest signal to an answer. OpenAI's bot documentation and Perplexity's crawler documentation publish user agent strings and IP address lists.

The example command below counts requests per bot in a combined-format access log.

grep -oE "GPTBot|OAI-SearchBot|ChatGPT-User|ClaudeBot|Claude-SearchBot|Claude-User|PerplexityBot|Perplexity-User" access.log | sort | uniq -c | sort -rn

Bot counts have two limits. A user agent name can be spoofed, so compare the request's IP address with the list the company publishes. Google-Extended is a token used only in robots.txt and has no separate user agent string in logs. I covered reading an access log in my post on log file analysis for SEO.

How do you track brand visibility in AI answers by hand?

Tracking brand visibility in AI answers by hand means asking assistants a fixed question set at regular intervals and recording every answer in the same fields. It takes six steps.

1. Build the question set: group your customers' decision questions by informational, comparison and purchase intent. I covered ways to find intent in my post on user intent in SEO.
2. Fix the surfaces: for example AI Overviews, AI Mode, ChatGPT, Gemini and Perplexity.
3. Fix the conditions: use the same language, country and device every round, and ask each question in a new chat.
4. Ask each question with the same wording every round, and store the full answer text with its date.
5. Fill in three fields per answer: is the brand mentioned, is your site cited, which page is cited.
6. Repeat the round at a fixed interval and calculate the mention rate: answers that mention the brand divided by total answers.

Example column headers and one row for the record table are below.

date,surface,question,brand_mentioned,site_cited,cited_url
2026-10-05,ChatGPT,"how to choose running shoes",yes,yes,https://example.com/running-shoe-guide/

Manual tracking has three limits. The same question can return a different answer each time. The answer can change with the user's location, language and account. Google says AI Overviews appear only when its systems determine that they add to classic Search. The resulting rate is a sample: it shows direction, not an exact count.

How do the sources for measuring AI search visibility compare?

The sources for measuring AI search visibility differ in the stage each one shows: impression, click, visit, crawl or mention. The table below compares six sources by data shown and by limit.

What do AI visibility measurement tools do?

AI visibility measurement tools automate the manual tracking method: they ask assistants a question set at regular intervals and record the answers that mention the brand.

rankzup.ai, a tool I founded, measures how visible a brand is in ChatGPT, Gemini and AI Overviews answers. Ask three questions when you choose a tool.

• Which assistants does the tool query, and in which country and language?
• Do you define the question set yourself?
• Does the tool store the full text of the answer?

Google's generative AI guide says to be wary of third-party tools that promise ranking success or claim to use "internal" Google metrics. The guide adds that no such tool has access to Google's internal ranking or AI systems.

What does Google say about visibility in AI search?

Google says visibility in AI search needs no additional requirements and no special optimisation. Google's generative AI optimisation guide treats the work as SEO and lists five tactics you can ignore for Google Search.

• llms.txt files and other "special" markup.
• "Chunking" content into small pieces.
• Rewriting content just for AI systems.
• Seeking inauthentic "mentions".
• Overfocusing on structured data.

The consequence for measurement: indicators tied to those tactics do not measure visibility on Google. Requests to an llms.txt file, for example, say nothing about visibility on Google, because Google Search does not use the file. The guide points to the generative AI performance report for measurement.

Which mistakes are common when measuring AI search visibility?

The common mistakes in measuring AI search visibility are trusting a single answer, chasing a rank and looking only at traffic. Four mistakes follow.

• Drawing a conclusion from one query: the answer can change on the next attempt.
• Rank obsession: Search Console assigns one position to every link in an AI Overview, so track the mention rate instead.
• Looking only at traffic: an answer can name the brand without a link, and GA4 then shows nothing.
• Changing the question set every round: the rounds can no longer be compared.

How do you build a measurement routine for AI search visibility?

A measurement routine for AI search visibility combines a short weekly check with a monthly report. The weekly check has two tasks.

• Ask the question set, and record the mention rate and the citation rate.
• Check sessions and key events for the AI Assistant channel in GA4.

The monthly report has four tasks.

• Export the generative AI performance report from the pages tab, and compare impressions with the previous month.
• Count bot requests in the server logs by the three classes.
• Update the custom channel group's regular expression for new source values in the report.
• Add questions about new products to the question set, and keep the wording of existing questions.

Join the data from the four sources in one table keyed by page URL. I covered the joining method in my post on data-driven SEO.