AI in SEO

Google Search Console AI Visibility: What the Report Shows and How to Use It

Use Google Search Console’s AI visibility report without mistaking impressions for clicks. A five-step SEO/GEO workflow, one-page record, and the limits of the data.

Updated 27 October 2026. Google completed the worldwide rollout of its generative AI performance reports on 31 August. This guide now focuses on the reporting task and the limits of the data.

Short answer: In Search Console, open Performance → Generative AI on Search results to see how often a link to your site appeared in supported Google Search AI features. Use Pages, Countries, Devices, and Dates to see where those impressions came from. Then open the ordinary Search results Performance report and your site analytics for clicks and useful actions. The AI report measures visibility; it does not give you AI-specific clicks, prompts, or conversions.

Google says the dedicated Search report covers AI Overviews and AI Mode. Its separate Discover report covers supported generative features in Discover. The Search report's impressions are already included in overall Search Performance data, so do not add the two impression totals together. This distinction is the starting point for useful SEO and GEO reporting. Google's announcement and report documentation describe the scope.

What the report answers—and what it cannot

Did a link to our site appear in Google’s supported AI features?
Use Generative AI on Search results → Impressions. This measures visibility in AI Overviews or AI Mode for the selected dates and search type.

Which pages appeared?
Open Pages. These are the linked URLs Google credited, usually their canonicals.

Did visitors click from Google Search?
Use the ordinary Search results Performance report → Web, Page, Queries. It shows page clicks but does not isolate AI-feature clicks.

Did visitors subscribe, enquire, or buy?
Use site analytics and conversion records where those actions are measured.

Did another assistant show the page?
Use that platform’s own evidence or a separate monitoring method. Google’s report cannot measure non-Google systems.

The dedicated Search report also lets you inspect country, device, and date. As of Google's September 2026 update, its search-type filter can distinguish text-based Web searches from supported multimodal searches such as Lens. Select the scope that matches your question and keep it consistent when comparing periods. Google documents the current dimensions and search types.

It does not show exact queries or prompts, the passage used in an answer, why a page was selected, where its link appeared, separate AI Overview versus AI Mode totals, or AI-specific click-through rate. A rise in AI impressions is not evidence of a rise in visits, qualified leads, or revenue. Use the fields actually present in your property today.

A five-step reporting workflow

1. Choose a completed window and save the scope

Open your verified property and the generative AI Search report. Select a completed 28-day period or another window long enough for your site's volume. Record the dates, search type, and whether you are looking at Search or Discover. If the report is absent, Google says the site may not have enough AI-feature impressions to show it. Check the property's inclusion control only if the reporting question requires it; absence alone does not diagnose an SEO failure.

2. Start with pages, not a sitewide chart

Open Pages and note which URLs received impressions. A property total can combine several pages and search contexts. Pick an important URL, confirm that it is the intended indexed canonical page, and read it as a human: does it answer the likely task clearly, cite its sources, and offer a useful next step? The report does not reveal the exact prompt, so your explanation of why Google showed the link remains a hypothesis.

3. Put ordinary Search performance beside it

In Performance → Search results, use the same completed dates and Web search type, filter to that page, then inspect clicks, impressions, queries, average position, country, and device. This gives context on search demand and visits, but it does not let you calculate an AI-feature CTR by dividing all Web clicks by AI impressions. Web clicks can come from other Search placements. Query rows may be incomplete because of privacy and reporting limits. Google's Performance report help explains the measures.

4. Check the action after the visit

Use analytics to inspect engagement, subscriptions, leads, sales, or another action that matters for that page. Keep Google AI visibility, Google Web clicks, traffic attributed to other AI assistants, and on-site actions in separate columns. Our GA4 AI-assistant versus Search Console guide explains why those measurements answer different questions.

5. Decide on one page-level improvement and log it

Write a specific hypothesis before editing: for example, “This comparison page answers the definition but lacks a decision table that helps a buyer choose.” Improve the page for that reader need, record the date and exact change, then review comparable periods after Google has had time to recrawl it. Do not rewrite every page with low AI impressions. A zero may simply mean the feature did not show the URL in the available data. A page with impressions but few Web clicks may still be useful—or the query may have been satisfied without a visit. The report cannot distinguish those explanations by itself.

Copy this one-page AI visibility record

Use one record per important canonical page. Leave unknown fields blank rather than filling them with estimates.

Canonical page and reader task: URL; what a visitor should accomplish.
Completed dates and search type: start/end; text-based or multimodal Web; Search or Discover.
Google AI impressions: page-level report value; country/device context if useful.
Ordinary Google Web results: same page and dates; clicks, impressions, relevant queries, position context.
On-site action: observed subscriptions, enquiries, sales, or “not measured”.
Page observation: what the current page offers; source or content gap.
One proposed change: exact edit, owner, date, and why it helps a reader.
Review result: comparable later window; continue, change, defer, or insufficient data.

For a small site, a monthly snapshot may be more useful than reacting to a handful of daily impressions. Compare like-for-like windows and check country or device changes before interpreting a trend. Chart totals and page rows can differ because they use different aggregation; the report's help page explains this. Search Console dates use Pacific Time, another reason to record the range rather than relying on memory.

How to interpret three common patterns

The AI report shows an important page, but clicks are low

First ask whether the page gets relevant ordinary Web impressions and what the current title and snippet promise. Check the live result and the page itself. If the page only repeats the short answer, add a genuinely useful example, comparison, worksheet, source, or tool that is worth a visit. Do not assume the AI feature caused the low clicks; position, intent, competing results, and the period can all matter. The metadata workflow helps when search appearance is the concrete issue.

Ordinary Search is visible, but the AI report shows little or nothing

Do not treat that as a penalty or a request to manufacture “AI-ready” pages. Check whether the page is indexed and eligible for snippets, then assess whether its answer is clear and supported by useful information. Google's AI features guidance says established Search foundations still apply. The AI report cannot tell you which prompts could have shown your page or why a competitor appeared.

AI impressions change after an edit

Record the edit and inspect the same page over comparable completed windows. Check regular Web performance and actual on-site actions separately. Search appearance can vary for reasons outside your edit; a before-and-after chart alone does not establish causation. If counts are tiny, report “too little evidence” and extend the observation period.

What GEO changes for the page itself

GEO—generative engine optimization—is often used as a label for improving visibility in AI-generated answers. For Google Search, begin with pages that people can use: crawlable and indexed where intended, eligible for snippets, clearly written for a real task, and supported by first-hand examples or reliable sources. Use descriptive headings and contextual links so readers can move from an overview to the detailed answer. Those are sound editorial and SEO practices; they are not guaranteed AI citation tactics.

Google describes “query fan-out” as part of how AI Mode can retrieve information for related aspects of a question. That does not mean you should publish a separate page for every possible sub-question. Use the idea to check whether an existing guide answers natural follow-up tasks, then add missing substance where it belongs. Our search-intent mapping guide helps decide when two reader tasks actually need separate pages.

Google does not require an llms.txt file, special AI schema, artificial “chunks,” or a Markdown duplicate for eligibility in its Search AI features. Structured data should match visible content and a supported feature when you use it. The more defensible improvement is useful, verifiable material that a reader cannot get from a generic summary. See Google's AI features documentation and people-first content guidance.

Should you exclude your site from Google's AI features?

Search Console includes a Search generative AI control under Settings. Google's default is inclusion. Excluding the site prevents its links and content from appearing in supported Google Search generative AI features, so it also removes the chance of traffic and impressions from those features. That is a business and publishing decision, not a routine SEO optimization. Evaluate the likely effect on discovery, rights, and revenue for your site before changing it. This guide does not change that setting.

What to do next

Pick one important page from the report, fill in the record above, and choose a reader-facing improvement that you can verify. If you are new to Search Console, begin with the five reports to check first. If the page is missing from Google's index, use the page-indexing decision tree before drawing any GEO conclusion. Keep AI visibility, ordinary Search visits, and on-site outcomes distinct in the next review.