Search Console Generative AI Report 2026

Administrator in Search Engine Optimization Sep 20 / 16 minutes ago

Open the generative AI performance report and count the metrics. There is one. Impressions. Nothing sits beside it. No clicks, no click-through rate, no average position, and no list of queries.

That single column decides what you can honestly say about your AI visibility. Google announced the report on 3 June 2026, then finished the worldwide rollout by 31 August 2026. Most teams have it now. Plenty have already read it wrong.

The common error is cheap to make and expensive to unwind. Someone opens the Search Console AI report, sees a healthy impression count, and adds it to the impressions in the main Performance report. The new total looks like growth. It is not. Those AI impressions were already sitting inside the Web number before this report existed.

So this is a piece about an instrument, not a strategy. I will cover what the report counts, what it refuses to tell you, which claims it can carry into a client deck, and which claims it cannot. Then I will point you at the places that hold the missing half.

What this guide settles

By the end of this page you will manage four things. Read the report without inflating your own numbers. Explain to a stakeholder why no click column exists. Distinguish a real absence of AI visibility from a reporting threshold. And build a weekly habit around the four dimensions you genuinely get.

Three of my arguments run against the usual coverage. First, the missing click data is not the worst problem with the Search Console generative AI report. The missing query dimension is, because without a query you cannot connect an impression to an intention. Second, an empty report says almost nothing on a small site, and most write-ups treat it as a verdict. Third, the most valuable number in the whole view is the country split, and hardly anybody discusses it.

I keep the strategy advice deliberately thin. If you want the optimization side, our comparison of how AEO, GEO and SEO actually differ covers it, and our practitioner guide explains how the main Performance report works. This page stays on the measurement question.

The report counts one thing, and counts it narrowly

An impression here means Google displayed a link to your site inside a generative AI feature on Search. That is the entire definition. Google words it plainly on the official help page: impressions are how many times links to your site were shown to a user in a generative AI feature on Google Search.

Three surfaces are covered, split across two reports. The Search version counts AI Overviews and AI Mode. The Discover version counts the generative AI features inside Google Discover. They are separate views with separate totals, so you read them separately and never add them together.

Now notice what an impression does not mean. It does not tell you whether a person saw the link, registered your brand name, or acted on either. It means Google placed a link to your page into an AI answer, and that answer appeared. If you already know what AI Mode is and how query fan-out works, you know a single answer can pull passages from a dozen sub-queries. So read the page rows as a ranking of which pages AI answers reach, not as an arithmetic breakdown of a site total.

The double-counting trap

The impressions in this report already sit inside your Web search totals. They are not new data added to your account. Google states it directly. The generative AI performance report includes data from the Web search type in the Performance report for Search results.

Read that twice, because the wording is easy to skim. The Search Console AI report is a filtered view of numbers you already owned. It is a slice, not a supplement. Before June 2026 those impressions were in your Web totals and you simply could not isolate them.

So three habits have to go. Adding AI impressions to Web impressions and calling the sum total visibility. Reporting an impression lift in the month you first opened the report. Building a year-over-year chart that swaps instruments halfway through.

The third one is the sneakiest. Your 2025 baseline never contained a separate AI figure, so adding one to 2026 manufactures growth out of a reporting change. It is the same trap as GA4’s AI Assistant channel, which started at zero and looked like a launch, and it belongs in the same bucket. When a number jumps, rule out the reporting artifact before you blame a ranking loss, and rule it out before you claim a win.

Four things the report refuses to tell you

No clicks, no click-through rate, no average position, and no query dimension. Those four absences shape everything the Search Console AI report can support.

Clicks matter most to a commercial team, and their absence is the headline complaint. I think the query gap damages practitioners more. An impression with no query attached tells you a link appeared. It cannot tell you what the person wanted. You can watch a product page draw AI impressions and still have no idea whether the prompt was a pricing question, a rival comparison, or a support problem.

Average position is a smaller loss than it looks. In the main Performance report, average position is impression-weighted and pooled across every query, country and device, so it moves when your query mix rotates even if your ranking holds. A position number inside an AI answer would be harder still to interpret, because no stable list of ten results exists to occupy.

The query gap has one partial workaround, and it is manual. You test prompts yourself and log what returns. Our GEO guide covers Share of Model and the wider metric set that accompanies that work, so I will not repeat the procedure here.

One sentence of regulatory context, and only one. The United Kingdom’s Competition and Markets Authority has set a timeline requiring Google to introduce page-level controls for generative AI features by March 2027, and the regulator also expects click-through data, although Google itself has promised neither a click column nor a date.

The four dimensions you actually get

Impressions is the only metric, but four cuts are available.

• Pages. Google groups the data by the final URL the AI feature linked. That is your canonical URL, so review your canonical choices before you treat a page row as a surprise.
• Countries. Where the search originated. This remains the most under-used dimension in the report.
• Devices. Desktop, tablet, or mobile. This cut belongs to the Search version.
• Dates. Day, week, or month. Search Console runs on Pacific Time, so a team in Europe or Korea reads a day that begins and ends in California.

URL filtering is available too, which is what makes the report usable on a large site. Pick a folder or a URL pattern, filter to it, then read the country split inside that filter. That combination answers the question the report was built for. Which parts of my site do AI answers reach, and where in the world.

An empty report is not proof of zero AI visibility

A blank view can mean three different things, and only one of them is genuinely bad news.

Google names the reasons. A site may not have received enough impressions in generative AI features on Google Search. A site may have been excluded from Search generative AI features. And during the phased launch some properties simply had no access yet, which stopped mattering once Search Engine Journal confirmed the worldwide rollout on 31 August 2026.

So the honest reading of an empty report on a small site is uncertainty, not absence. You sit below a threshold whose value Google has never published. Treat it exactly as you treat a page with four impressions in the main Performance report. The number is real and it carries almost no information.

Investigate the exclusion case before you panic. Search Console carries a property-level Search generative AI control in its settings, and Google finished rolling that control out worldwide on the same date it finished the report. Every property defaults to include, so somebody has to switch it deliberately. If a colleague did, the report goes quiet while your ordinary ranking stays healthy, because Google says the control is not used as a ranking signal for search results outside those generative AI features. Check that setting first. An older route reaches a similar place by accident, because the nosnippet directive also removes a page as an input to AI answers, and our piece on what blocking snippets costs you in AI Overviews and AI Mode walks through that trade.

How to read an impression with no click beside it

Treat the number as an exposure signal and nothing more. An impression count tells you Google is willing to place your pages inside its generative answers. That is genuinely useful information. It is not a traffic measurement, and no spreadsheet arithmetic will convert it into one.

A few readings hold up. Direction over time on a page set you fix in advance, remembering that a frozen list can only lose members, so pair it with one that recruits newcomers, as our rank tracking guide argues for keyword panels. The proportion of your AI impressions landing on pages you actually want cited. And the country split. Everything else demands a second instrument.

The brand mention world faced this problem first and solved it honestly. Mentions frequently produce no click at all, so practitioners track why brand mentions matter even when no click follows. Do the same here. Pair the impression trend with branded search volume, with direct traffic, and with your own conversion records. If the impression line climbs for six months and nothing else moves, you have visibility without value, which is the zero-click problem in a single chart.

A labeled worked example

The numbers below are invented for illustration. They describe no real site and they are not a benchmark.

Say a mid-size software company opens the report and finds 40,000 AI impressions for the month. The main Performance report shows 900,000 Web impressions and 30,000 clicks. A junior analyst records 940,000 total impressions. That line is wrong. The 40,000 already live inside the 900,000, so the total remains 900,000.

Now do something genuinely useful with identical figures. Filter the AI report to the documentation folder. Suppose 26,000 of the 40,000 sit there, against 9,000 on product pages. AI answers are mostly reaching support content. Examine the country dimension next. Suppose the United States supplies 19,000 and Germany supplies 4,000. You now hold a real question to chase. Why does the product side barely appear, and is the German gap a language problem or a content problem.

Notice what you still cannot claim. You cannot say those 40,000 impressions cost you clicks, and you cannot say they earned any. No click data exists in the report, so both statements are guesses.

Search and Discover behave like two different instruments

A Search version of this report and a Discover version both exist, and blending them will mislead you. Discover traffic behaves nothing like search traffic. It spikes, it vanishes, and no query sits behind it, because nobody typed anything.

Keep two habits apart. Read the Search AI view against your Web search work. Read the Discover AI view against your Discover work, which for most publishers is a bonus channel rather than a baseline. Pooling both into one AI impressions number produces a figure no decision can use.

One further difference matters. The device cut belongs to the Search version only. That is a smaller loss than it first looks, because the feed lives in the Google app and the Chrome mobile new tab, so a Discover-heavy publisher has almost no desktop audience to separate out. What it does cost you is any device comparison across the two reports.

The country dimension is the quiet win for the US, Europe and Korea

The country cut is the only way to see where your AI impressions originated, and it is the part of this report I would build a standing chart around.

If you sell into the United States, Germany, France, the Netherlands and South Korea, a single global impression total conceals everything you need. A 20 percent rise that is entirely American tells a European team nothing whatsoever. Split the number by country and the Search Console generative AI report starts answering commercial questions. Which markets do AI answers already reach, and which ones have never heard of us.

Korea needs a caveat. This report covers Google surfaces only. Naver remains a major search destination in Korea, Kakao runs the Daum portal alongside it, and nothing in this view can observe either. Our guide to GEO beyond Google in the US, Europe and South Korea covers that engine mix. Pair the country split with your market plan, and remember Korea runs sixteen hours ahead of Pacific Time, so a Korean working day lands across two of the report dates.

Where this report sits next to GA4, rank tracking and your logs

This report measures exposure inside Google's AI features. It is one layer in a stack, and the remaining layers answer the questions it drops.

• Referral traffic. Visits from AI assistants land in analytics, never here. Our piece on GA4's AI Assistant channel and the referrers it never sees covers that side, including sessions arriving with no referrer that sit in Direct forever.
• Position and page shape. A rank tracker still shows what occupies the space above your listing. Impressions alone cannot.
• Your own data. Server logs, CRM records and conversion events survive every reporting change Google makes. Build the habit of layering metrics by how much you can trust them.

I rank them in that order deliberately. Search Console impressions are cheap, clean and narrow. Analytics is broad and lossy. First-party data is expensive and true. Use all three, and stop demanding that one report becomes the whole answer.

What belongs in a client report, and what I would cut

Include the AI impression trend, label it unmistakably as impressions, and never let it touch a traffic total.

Not much earns its space. A month-over-month AI impression trend on a page set you fixed in advance, the country split for your target markets, and a short note on which page types the AI features reach. That is genuinely all.

Cut everything pretending to be a click. Cut any AI share of voice figure built from impressions, because you have no denominator. Cut the total visibility metric that sums both reports. And cut the sentence claiming AI Overviews took your traffic, unless you can demonstrate the click loss on identical queries in the main report.

Here is the practical danger. A one-page dashboard in Data Studio, which Google renamed back from Looker Studio in April 2026, or a BigQuery export feeding it, will happily display an impression count beside a revenue figure, and that adjacency is how a misleading chart gets born. A new dimension is not automatically a new KPI. Our guide to which KPIs actually prove ROI argues it better than I can here. Impressions from AI features are a leading indicator at best, and a vanity line at worst.

A weekly routine that does not overclaim

Fifteen minutes weekly is plenty, once you stop demanding things the data cannot deliver.

• Open the Search version and read the last 28 days. Record the total. Direction matters and the absolute value does not.
• Read the page rows. Ask whether the pages Google surfaces are the pages you wanted surfaced.
• Read the country rows for your priority markets.
• Check the Discover version if you publish news or feature content.
• Write one sentence in your notes. Not a chart. A sentence.

Once a quarter, go deeper. Compare the page set drawing AI impressions with the pages you intended to be cited. A mismatch usually indicates structure rather than authority, and clear entity and Knowledge Graph work on the page tends to repair more of it than another link campaign would. If the gap is that AI answers reach you but rarely link you, the remedy lives in how to earn citations in ChatGPT, Perplexity and Gemini, not in this report.

Frequently asked questions

What does the Search Console generative AI report measure?

Impressions, and only impressions. An impression counts each occasion Google displayed a link to your site inside a generative AI feature on Search, including AI Overviews and AI Mode. A matching Discover version exists. No clicks, no click-through rate, no average position and no queries appear anywhere in the view.

Does the report show clicks or click-through rate?

No. Clicks and click-through rate are both absent, and so is average position. Google has not committed to adding them and has published no date. The UK regulator expects click-through data as part of its wider requirements, but a regulator expectation is not a shipped feature.

Should I add AI impressions to my Performance report impressions?

Never. Google states the generative AI report includes data from the Web search type in the Performance report, so those impressions already sit inside your Web totals. Adding them manufactures a fake increase. The AI view is a filter on data you already owned, not an extra pot of visibility.

Why can I not see the prompts behind my AI impressions?

No query dimension exists in the report, so no prompt data lives there. That is the gap I find hardest to work around. The only practical substitute is manual prompt testing against a fixed question list, repeated over several weeks, which is slow and small but honest.

My report is empty. Does that mean I have no AI visibility?

Not necessarily. Google lists three explanations, and two of them still apply now that the rollout is complete. Your site may not have received enough impressions in generative AI features to be reported, or your site may have been excluded from those features. The third, no access during the phased rollout, stopped applying on 31 August 2026. An empty report on a small site indicates uncertainty, never a confirmed zero.

When did the report become available?

Google announced it on 3 June 2026 and released it in phases. By 31 August 2026 it had reached all sites worldwide. So a missing report is no longer a rollout question. If you see no data now, investigate the impression threshold or an exclusion setting instead.

What is the difference between the Search and Discover versions?

They cover different surfaces and deserve separate readings. The Search version covers generative AI features on Google Search, including AI Overviews and AI Mode, and it carries the device cut. The Discover version covers generative AI features in Discover. Pooling both produces a number no decision can use.

Can I see which countries my AI impressions came from?

Yes, and it is the most valuable cut available. The country dimension groups impressions by where the search originated, which makes a United States, Europe and Korea split possible for the first time. Combine it with URL filtering to read a single folder or URL pattern by market.

Does the report cover AI Mode as well as AI Overviews?

Yes, but you cannot separate them. Google names both AI Overviews and AI Mode among the generative features the Search version covers, and the dimensions on offer are pages, countries, devices and dates, so no AI Mode only figure exists inside the report.

Does this report say anything about ChatGPT, Perplexity or Naver?

No. It covers Google surfaces exclusively. Visibility inside ChatGPT from OpenAI, Perplexity, Claude, DeepSeek, Grok, Microsoft Copilot, Bing or Naver sits entirely outside it, and no Google report covers them today. For those engines you need referral data in your analytics, manual prompt checks, and a dedicated AI visibility tool.

The bottom line

Return to the opening. One metric, one column, and a long inventory of things it will not say. That is not a failure of the report. That is the report's actual shape, and reading it well means reading it narrowly.

If you do one thing this week, do this. Open the Search Console generative AI report, record your impression total, then delete every slide where somebody added that number to your Web impressions. The second step matters considerably more than the first.

My prediction. Click data arrives eventually, but not on Google's initiative and not soon. Regulators are pushing, publishers are pushing, and Google has every commercial reason to move slowly. Plan your measurement as though impressions are all you will get for the next year, because that is the likely case.

One question, and I genuinely mean it. Has your AI impression trend ever changed a decision you made, or has it only ever changed a slide? The answer separates a useful instrument from a decorative one.