Search Console Generative AI Report: AI Overviews & AI Mode

What is the Search Console generative AI report?

The Search Console generative AI report (Google calls it the Generative AI performance report) shows how often links to your site appeared in Google’s AI Overviews and AI Mode. It reports impressions only, broken down by page, country, device and date. There are no clicks, no CTR, no position and no queries. Google announced it on 3 June 2026 for a subset of sites and says it reached all websites worldwide on 31 August 2026.

There are actually two reports: one for Search, which covers AI Overviews and AI Mode together, and a separate one for generative AI features in Discover. This guide focuses on the Search report, because that is where most sites see AI visibility.

The numbers are not new traffic. According to Google’s launch announcement, the same impressions are still counted in the main Performance report. The new report simply isolates the AI slice, which is useful for the first time since AI Overviews launched: until now you could only guess which pages were being shown inside AI answers.

Where to find the generative AI report in Search Console

1. Open your property in Search Console.
2. In the left menu, open Performance and choose the generative AI report for Search. The Discover version sits alongside it if your site gets Discover traffic. Google’s help page links straight to it; the address ends in /performance/search-analytics/ai.
3. Set the date range and the search type: Web: text-based or Web: multimodal.

If it isn’t there, the generative AI performance report help page gives two reasons: the site hasn’t had enough impressions in AI features yet, or it has been excluded from them under Settings → Search generative AI. Check the parent property too, because child properties inherit that setting by default.

Rollout timeline

Google hasn’t documented how far back the report goes; check the first date in your own report.

What the generative AI report shows

Why the chart and the Pages table disagree

The chart is aggregated by property: if one AI Overview links to two of your pages, the chart counts one impression. The Pages table is aggregated by page, so the same AI Overview gives each page one impression, two in total. On sites with several pages cited for the same topic, the Pages table can add up to much more than the chart. Neither is wrong; they count different things.

What it doesn’t show: clicks, queries and the API

• No clicks, CTR or position. Clicks from AI Overviews and AI Mode are in the Performance report, but mixed into all your other Web clicks.
• No queries. You can see that a page was shown in AI features, not what was asked.
• No split between AI Overviews and AI Mode. Both are counted together.
• Only AI Overviews and AI Mode, for now. Google says the list may grow, and Search Labs experiments are excluded, so don’t read the report as a measure of the Gemini app.
• No API. The Search Analytics API reference lists no generative AI dimension, filter or search type, and the bulk data export schema has no AI field either. The report is UI and export only.

How AI Overviews and AI Mode are counted in the Performance report

Google’s impressions, position and clicks documentation sets out the rules. Both features are counted under the Web search type, so they are already inside your Performance totals. Never add the generative AI report’s impressions to your Performance impressions; you would be counting them twice. In Google Analytics 4, clicks from both land in Organic Search, not the AI Assistant channel; see Search Console vs Google Analytics 4.

The shared position is what confuses most reports. When an AI Overview sits at the top of the page, every site cited in it is credited with that top position, even if its organic listing is further down. A page can therefore show a better average position and a worse CTR at the same time. Our guide to Search Console metrics explains how each one is calculated.

An illustrative example: an informational guide averaged position 3.2 with a 6.1% CTR before an AI Overview started appearing for its main query. It is now cited in that Overview, its average position reads 1.6, and CTR has fallen to 2.3%. Nothing is broken on the page. The results page changed around it. If this pattern sounds familiar, the “rankings held but clicks fell” diagnosis walks through how to confirm it.

Industry data points the same way. Ahrefs’ February 2026 study of 300,000 keywords found that an AI Overview correlates with a 58% lower average CTR for the top-ranking page, based on desktop data. It is a correlation, not proof, but it is a useful benchmark for your own pages.

The follow-up rule matters for AI Mode. A three-turn conversation can produce impressions under three separate queries, often long and conversational ones, which inflates your long-tail query count without any change in demand.

How to analyse the generative AI report with your Performance data

The report becomes useful once you put it next to clicks. The method is a simple page-level join.

1. In the generative AI report, choose the last full 28 days, open the Pages tab and export it.
2. Pull Web clicks, impressions, CTR and position for the same pages and the same dates.
3. Work out each page’s AI share: AI impressions ÷ Web impressions. Both Pages tables are aggregated by page, so the ratio is a fair comparison.
4. Add the CTR from the previous 28 days so you can see the direction of travel.

Step 2 is where an assistant saves the time. Attach the export to Claude or ChatGPT with GSC MCP Server connected and ask:

Expect small mismatches. The UI splits Web into text-based and multimodal, the API documents a single web type, and the export stops at 1,000 rows. For a site-wide view those differences rarely change the conclusion.

A worked example

The figures below are illustrative, for a kitchen appliance site:

The two guides are shown mostly inside AI answers, hold strong positions and have lost more than half their CTR. The comparison and product pages barely appear in AI features and are unaffected. That is a far more useful story for a stakeholder than “informational traffic is down”.

Decision rules

Treat these thresholds as starting points and tune them to your site.

• AI share above about 40%, steady position, falling CTR. A new title won’t fix it. Add what a summary can’t deliver, such as tools, original data, templates or first-hand detail, and make the title promise it. Judge the page on conversions as well as clicks.
• High AI share, steady clicks. Being cited is working. Keep the page accurate and current, and don’t restructure it on a hunch.
• Plenty of Web impressions, almost no AI impressions on an informational page. Check it is indexed and allowed a snippet (no restrictive nosnippet or max-snippet), then ask honestly whether it answers the question better than the pages that are cited.
• AI impressions falling while Web impressions hold. You may have lost citations to a fresher or more specific page. Compare the content, not just the rankings.

To spot pages heading into the first group before you export anything, ask for impressions rising while CTR falls:

Finding the queries behind AI impressions

The report has no query dimension, but the Performance data does. Two query patterns are good proxies for AI-heavy searches: questions and long, conversational queries. Run them through Search Console’s regex filter or the API:

• Questions: ^(how|what|why|when|where|which|who|can|does|do|is|are|should)\b
• Eight words or more: ^(\S+\s+){7,}\S+$

Both are RE2 patterns, the syntax Search Console uses, and both are case-insensitive in the UI by default. The Search Console regex cheat sheet has more variations, and our guide to keyword research with Search Console turns these queries into content.

One caveat: long, rare queries are often anonymised, and anonymised queries are left out of query tables. Your question segment will always undercount the real total, so watch its trend rather than its size.

Two more checks round this out. get_search_opportunities with the low_ctr strategy lists high-impression queries with CTR under 3%, and many of them will be informational queries with an AI Overview; the low-CTR playbook covers which ones a better snippet can still win. And compare_search_periods with compareTo set to previous_year shows which queries lost the most clicks year on year, as long as you compare clicks only (see the next section).

Data pitfalls: logging errors and broken comparisons

Google’s data anomalies page lists several issues that affect AI analysis:

• 13–17 August 2026. A logging error lowered impressions in the generative AI report for Search. Google restored the missing data on 21 August, so re-export anything you downloaded before then. Discover’s generative AI impressions were also hit on 13 August and 24 June.
• 13 May 2025 to 27 April 2026. A logging error affected impression reporting across the Performance report, and many sites saw impressions drop once it was fixed. Google says only impressions, CTR and position were affected. Clicks were not.
• Mid-September 2025. Not on Google’s list, but widely reported: when Google stopped supporting the num=100 results parameter used by rank trackers, many sites saw desktop impressions fall and average position “improve”.

The practical rule: any year-on-year comparison that reaches back before 27 April 2026 should use clicks only. Comparing October 2026 with October 2025 on impressions or CTR measures logging changes, not AI. Period-on-period comparisons that start after 27 April 2026 aren’t affected by that error.

Should you use the Search generative AI control?

Alongside the report, Google added a control under Settings → Search generative AI. It has three options: Include (the default), Exclude, or Inherit control from parent, the default for any property that has a parent. Here is what Google’s Search generative AI control documentation says excluding does:

• Your links and content stop appearing in AI Overviews, AI Mode and generative AI features in Discover, and you get no traffic or impressions from them.
• Content crawled from your site isn’t used to ground AI responses in those features. Similar content from other sites can still appear.
• It isn’t a ranking or inclusion signal for the rest of Search.
• It doesn’t affect AI training. That is what Google-Extended is for.
• Changes generally take a few days. Content is excluded within 1–2 days of the control going live, and some takes longer because of caching.

For most businesses the answer is to stay included. AI features retrieve pages through the same index and ranking systems as normal results, so an excluded site is simply absent while competitors are cited. Exclusion makes more sense when your revenue depends on people reading the content on your page, your AI share is high and your CTR on those pages has collapsed.

If you want evidence first, use inheritance. Create a URL-prefix property for one section, set it to Exclude, and compare its clicks over 4–8 weeks against a similar section left included. It isn’t a clean experiment, since seasonality and updates still apply, but it beats switching off a whole site on instinct. Bear in mind you are judging by total Web clicks, because Search Console never shows AI clicks on their own.

What to do with what you find

Google’s guide to optimising for generative AI features (updated 10 July 2026) is blunt: these features are rooted in core Search ranking, so optimising for them is still SEO. It says you don’t need llms.txt files, special schema or “chunked” content for Google Search. Its emphasis is on non-commodity content, a clean technical setup, and eligibility: a page must be indexed and allowed to show a snippet.

• Check eligibility on your top AI pages. Run inspect_url on them to confirm they are indexed, that Google chose your canonical, and that robots.txt allows crawling.
• Fix pages with high AI share and falling clicks with depth a summary can’t replace, not with title tweaks alone.
• Report AI visibility as its own line. AI impressions, AI share and clicks on AI-heavy pages tell a clearer story than a single traffic chart.
• Look beyond Google. Bing’s AI Performance report counts citations in Copilot and Bing’s AI answers. See why Bing matters for AI search, and our framework for measuring AI search visibility across Google and Bing.

It also helps to remember what this report is: Google’s first, deliberately narrow view of AI visibility, and its launch post says more metrics may follow. Until clicks arrive, the most honest picture comes from combining its impressions with the click data you already have. If you’re new to that kind of workflow, here’s how a Search Console MCP server works.

Frequently asked questions