How to Read the Search Console Generative AI Report: AI Search Performance Metrics That Matter

Last updated: September 11, 2026

For the first time, Google is telling site owners how often their pages appear inside AI Overviews and AI Mode. The Generative AI performance report in Search Console, announced on 3 June 2026 and rolled out to every website worldwide on 31 August 2026, is the only official source of ai search performance metrics for Google’s AI features. It is also narrower than most people expect: it reports impressions, not clicks, and it cannot tell you which questions triggered those impressions. This guide explains exactly what the report measures, what it leaves out, which metrics you can build from it today, and how to turn a new column of numbers into an ai search visibility strategy you can act on.

According to the Search Console Help Center, the report exists to show how organic impressions from Search generative AI features change over time, which pages earn the highest or lowest impressions in those features, and where those impressions originate by device and country. Google is explicit that the data was already included in the overall performance report; the new view isolates it rather than adding new traffic.

If you want the on-page work that follows from this report handled automatically, sign up for free and let NytroSEO keep your metadata and structured data aligned while you focus on the numbers.

The timing matters. At Google I/O in May 2026, Google said AI Mode had passed one billion monthly users a year after launch, with queries more than doubling every quarter, and that Gemini 3.5 Flash had become the default model powering it. A surface that size, with no reporting behind it, was the largest blind spot in search analytics. The report closes part of that gap. The rest of this article is about using the part it closes and working around the part it does not.

Where the Search Console Generative AI report lives and what it counts

The report sits under Performance in the left-hand navigation of Search Console, alongside the familiar Search results report, and Google also gives it a direct address at search.google.com/search-console/performance/search-analytics/ai. A separate Generative AI report exists for Discover; this article covers the Search version only.

What it counts is deliberately simple. An impression is recorded every time a link to your site is shown to a user inside a generative AI feature on Google Search. Today that means two features: AI Overviews and AI Mode. Google says it expects to add to that list over time, and it excludes experiments running in Search Labs because those are still in development.

Three details from the Help Center change how you should read the numbers:

• The chart is aggregated by property. If two results from your site appear in the same AI answer, the chart counts one impression. If you add a URL filter, the chart switches to per-URL aggregation, so chart and table totals will not always match.
• Impressions are assigned to the canonical URL, not to a duplicate, exactly as they are in the standard performance report. If your canonicalisation is messy, your AI impressions will be attributed to pages you did not expect.
• The newest data is preliminary and is drawn as a dotted line. It can still change over the following hours, so do not annotate a weekly report on yesterday’s figure.

The report offers four dimensions: pages, countries, dates and devices. That is the full list. There is no query dimension, no search appearance breakdown by feature, no position, and no click column. The same 1,000-row table limit that applies to the standard performance report applies here, and there is an export button for both the chart and the table, with unavailable values written as zeros in the download.

One more thing to check before you draw any conclusion from an empty tab: Search Console has a control that lets a site opt out of Search generative AI features altogether. If someone on your team switched that on, the report will show nothing because your pages are not eligible to appear. The Help Center lists that exclusion as one of the two reasons a property may see no data, the other being that the site has not yet earned enough impressions in AI features.

Lee Agam, founder and CEO of NytroSEO, sees the report as a diagnostic tool rather than a scoreboard. In his view the mistake most teams will make in the first month is to treat AI impressions as a new traffic number and chase it. The more useful reading is structural: the report shows which pages Google’s AI features already trust enough to surface, and the gap between those pages and the rest of the site is the work. Because impressions attach to canonical URLs, he recommends fixing canonical consistency before comparing any two pages, otherwise the report is measuring your tagging rather than your content.

The five AI search performance metrics you can build from the report

The report gives you one raw number, sliced four ways. The metrics that matter are the ones you derive from it by combining that number with data you already have. These five are enough to run a monthly review.

The scorecard above summarises the five metrics; each is defined below with the calculation and the decision it supports.

1. AI impression share per page

Divide a page’s generative AI impressions by its total Search impressions for the same period. Both figures come from Search Console: the first from the Generative AI report, the second from the standard Search results report filtered to Web search type. A page with 12,000 classic impressions and 3,000 AI impressions has a 25 per cent AI impression share. The number itself is less important than the spread across your site. Pages far above the site median are your templates; pages at zero with healthy classic impressions are your candidates.

2. AI-to-classic impression ratio, site-wide

The same calculation at property level gives you a single trend line to report each month. Because Google aggregates the chart by property, use the exported table for the page-level numbers and the chart for the property-level trend, and keep the two apart in your notes. The ratio tells you whether AI features are becoming a larger or smaller share of your visibility, independent of seasonality in total impressions.

3. Page gap list

Export the pages table, join it against your full URL inventory, and list every indexable page that has classic impressions but no generative AI impressions. This is the most actionable output the report produces. It answers a question nobody could answer before: which of your pages are being ranked by Google but not selected by its AI features. Most of the remaining sections of this article are about what to do with that list.

4. Country and device skew

AI Overviews and AI Mode did not launch everywhere at once, and device behaviour differs. Compare your AI impression share by country and device against your classic distribution. A market where AI impressions are a much smaller share than classic impressions may simply lack the feature, or may be a market where your content is not being selected. Confirm feature availability before treating the gap as a content problem.

5. Week-over-week trend with annotations

Track total AI impressions weekly and annotate the series with the events that move it: your own publishes and refreshes, Google’s announced updates, and model changes. Google’s Search Status Dashboard records the update windows. In August 2026 alone there was a spam update and a switch to Gemini 3.7 Flash for some AI Mode traffic; without annotations, a movement caused by either would be misread as the result of your own work.

Once you have identified the pages Google’s AI features already surface, you need to keep them that way. Book a strategy meeting with the NytroSEO team if you want help turning the page gap list into a prioritised plan for a large site or an agency portfolio.

The AI search performance metrics the report still hides, and how to fill the gap

The report is honest about its scope, so you should be too. It does not show clicks, click-through rate, queries or average position for AI features. Clicks from AI Overviews and AI Mode are still counted somewhere: they sit inside the overall Search performance numbers, but they are not broken out, so you cannot calculate an AI-specific click-through rate from Search Console alone. That is the single biggest limitation for anyone who needs to prove revenue from AI visibility.

Two other gaps matter. There is no query dimension, so you cannot see the prompts or follow-up questions that produced an impression. And the report covers Google only; nothing in it tells you whether ChatGPT, Perplexity, Gemini’s standalone app or Copilot cite you. For the ai visibility picture across every engine, you need two supplements.

The diagram maps each hidden metric to the method that fills it: a fixed prompt panel for citations and mentions, and crawler-log checks for retrieval.

Supplement one: a fixed prompt panel

Write twenty buyer questions that matter to your business, run them monthly in ChatGPT with search enabled, Perplexity, Gemini and Copilot, and record for each whether you were cited with a link, mentioned without one, or absent. This is the method our own guide to measuring AI search visibility is built on, and it is still the only way to observe citations directly on non-Google engines. Keep the question set fixed so the month-to-month comparison is valid, and separate mentions from citations, because they drive different outcomes.

The prompt panel also catches shifts the Google report cannot. In August 2026, independent tracking reported that ChatGPT changed how it fans out searches, increasing its use of site-scoped queries, with forum citations falling sharply and official brand and documentation sources gaining share. A change like that moves your llm visibility overnight and leaves no trace in Search Console.

Supplement two: crawler-log checks

The prompt panel tells you whether you were cited. Your server or CDN logs tell you whether you could have been. Confirm that Googlebot, GPTBot, PerplexityBot and Google-Extended are fetching the pages on your gap list. A page that is not being fetched by an AI crawler will never appear in that engine’s answers, and an AI search indexing audit that starts with the logs will find blocked bots, CDN rules and robots.txt mistakes faster than any amount of content work. If the bots are being fetched but the page still earns no AI impressions, the problem is on the page.

Reading ai search performance metrics for agencies: a client-safe interpretation

Agencies face a specific risk with this report: a client will open it, see a number labelled impressions, and ask why it is smaller than the classic figure or why it is not growing as fast as the invoice. The interpretation you give them in the first month sets expectations for the year.

Use the report to compare the client’s own pages against each other, never to compare the client’s total AI impressions against a benchmark. There are no reliable benchmarks yet, the rollout was staged, and data only accumulates from the point Google began collecting it for that property, so early charts look sparse. Search Engine Land noted at the global rollout that the report covers AI responses, AI Mode and AI Overviews with impressions, pages, countries, devices and dates, but no click data. Put that sentence in the report template so the limitation is stated before the number.

Two further cautions belong in every client note this quarter. First, Google logged an impressions data-logging bug affecting the generative AI report in August 2026, so month-on-month comparisons that span that period carry a caveat. Second, do not touch the Search generative AI control. Opting out removes the client’s pages from AI Overviews and AI Mode, which together reach billions of users each month, and the report will go blank. The control exists for sites with licensing reasons to exclude themselves; for almost every commercial site, the visibility loss outweighs any benefit.

Finally, explain why impressions without clicks still carry value. Pew Research data cited by Cloudflare in July 2026 found that when Google shows an AI summary, users click a traditional result link only about eight per cent of the time, roughly half the rate seen without a summary. In that environment, being the source named inside the answer is the top of the funnel. Report AI impressions alongside branded search growth and assisted conversions, and the client sees the shape of the value rather than a missing click.

Turning ai search performance metrics into an ai search visibility strategy

The report becomes strategy at the moment you stop reading it and start copying from it. Your pages that already earn generative AI impressions are, by definition, pages Google’s systems judged retrievable and useful for AI answers. Study their structure and apply it to the gap list.

In our experience across the pages we manage, the pages that earn AI impressions share five traits, and each one is checkable without special tooling:

1. A clear definition or direct answer in the first paragraph, so the page can be summarised without reading to the end.
2. Headings that match how people phrase questions, with each section leading on its answer rather than building to it.
3. Cited sources and dated figures close to the claims they support.
4. A named author with relevant credentials and consistent Organization and Person structured data.
5. Titles, descriptions and canonical tags that match the page content and do not drift after theme or plugin updates.

Google’s own position, set out in its guide to optimizing for generative AI features on Search published in May 2026, is that no special markup, file or AI-specific format is required, because AI Overviews and AI Mode are rooted in the same core ranking and quality systems as Search. That is good news for the gap list: the fixes are ordinary. Our guide to getting cited in Google AI Overviews rather than just ranked covers the content side in detail, and our overview of AI brand visibility and AI Overviews explains the entity side.

The fifth trait is where automation earns its place. On a site with a few dozen pages, keeping titles, descriptions and canonicals aligned is a Friday afternoon task. On a site with thousands of pages, or an agency portfolio of client sites, it is a recurring failure: builders regenerate metadata, plugins overwrite descriptions, and a theme update quietly changes canonicals. Because the report attributes impressions to canonical URLs, that last failure directly corrupts your measurement. NytroSEO applies approved metadata and structured data through a JavaScript header snippet, scoped by sitemap, and monitors for drift, which keeps the page you are measuring the same page from month to month. It does not fix rendering problems or write body content; those remain people’s work.

If you already connect Search Console to NytroSEO, the Google Search Console integration is the place to pull classic impressions for the ratio calculations in this guide, with the generative AI export layered on top.

A 28-day working loop

• Week 1: export the Generative AI pages table and the classic Search results table for the same 28 days; calculate AI impression share per page; build the gap list.
• Week 2: run the twenty-prompt panel across ChatGPT, Perplexity, Gemini and Copilot; check crawler logs for the gap-list pages.
• Week 3: fix access and canonical issues first, then apply the five traits to the ten gap-list pages with the highest classic impressions.
• Week 4: annotate the week-over-week series with what you changed and with any update on the Search Status Dashboard.
• Day 29: re-export and compare. Movement in AI impression share on the ten pages, not the site total, is the signal.

Frequently Asked Questions

Ready to act on your AI search performance metrics?

The report tells you which pages Google’s AI features already surface. Keeping every other page consistent enough to join them is the part that does not scale by hand. Sign up for free to apply consistent metadata and structured data across your site automatically, or, if you manage a large site or a client portfolio, book a strategy meeting with the NytroSEO team and we will walk through your gap list together.