Google AI Mode: what it is and what it leaves in your reports

Five pages on the first results page for this query, opened on 6 September 2026, explain what the product does for the person typing into it. None of the five names the Search Console report that Google's own optimization guide tells site owners to open. That report exists, it is three months old, and it is missing four of the five columns anybody would go looking for.

What is Google AI Mode?

Google AI Mode is a separate conversational surface in Google Search that answers in prose instead of a list of links. Google documents no extra requirement to appear beyond being indexed and snippet-eligible. A new site is not yet a candidate; a mid-authority site earns impressions it cannot convert; a consolidated site loses sessions it cannot trace.

Quick answer

• It is a surface, not the block above the results. A person opens AI Mode on purpose. An AI Overview arrives without being asked for, on top of a results list that stays underneath.
• Eligibility is the ordinary eligibility. Google documents that a page must be indexed and eligible to be shown in Google Search with a snippet, and that nothing else is required.
• Two Search Console reports touch this surface and neither one is whole. The performance report has clicks and no way to filter the surface. The generative AI report filters the surface and has no clicks.
• 93 % of AI Mode queries end without an external click, in two independent measurements, so a session count answers a question this surface never asks.

Read down the third column and the shape of the subject appears. Every row has an answer and every answer is partial, and the partial half is different each time. That is not a temporary state of the tooling: it is what reporting looks like when the thing being reported was composed rather than ranked.

What kind of problem this is

The model most people arrive with is that AI Mode is a channel. Channels have a report, the report has a traffic line, and the work is to make the line go up.

That model is wrong twice over, and the second error is the expensive one. The first is the easy confusion with what an AI Overview is: a summary that sits above a results page, in a place the user did not choose to go, with the ten links still below it. AI Mode is entered deliberately and replaces the list rather than crowning it. Google's own page on AI features notes that the two may use different models and techniques, so the set of responses and links they show will vary. They are not one feature with two intensities.

The second error survives after the first is corrected. A person who now understands that AI Mode is a separate surface still expects a report for it, and there is one, and it publishes a single metric. Impressions. No clicks, no click-through rate, no position, no queries. The useful question is therefore not how much traffic this takes. It is what can be known at all about a surface whose only published number is that you appeared on it.

The two Search Console instruments, and the column each one drops

Two documented reports touch AI Mode, and reading them side by side takes ten minutes and settles most of the arguments in this subject.

The performance report reference is precise about the counting. "Clicking a link to an external page in AI Mode counts as a click." "Standard impression rules apply." And a line that separates AI Mode from its sibling: "Position in AI Mode follows the same methodology as a Google Search results page." Compare that with what the same page says about the other surface, where "an AI Overview occupies a single position in search results, and all links in the AI Overview are assigned that same position". One surface reports positions the ordinary way. The other collapses every citation into one shared number. Almost nobody writes that difference down, and it changes how an average position column should be read. Getting the property that holds either is an afternoon of work, set out in how to set up Search Console.

What that report will not do is separate those rows from the rest. There is no AI Mode filter, so the clicks arrive inside ordinary organic data with nothing marking where they came from.

The second instrument is newer. Google announced generative AI performance reports on 3 June 2026, and its own optimization guide, last updated on 10 July 2026, points site owners there in one sentence: "To measure how your content is performing in generative AI features on Google Search and Discover, use the Generative AI performance report in Search Console." The report covers AI Overviews and AI Mode. It offers four ways to group the data: pages, countries, devices and dates. And it carries exactly one metric, defined on the page like this: "Impressions are how many times links to your site were shown to a user in a generative AI feature on Google Search."

The arithmetic that stops the obvious trick

The obvious move, once both reports are open, is to subtract. Take the impressions the generative AI report attributes to these features, take the total from the performance report, and read the difference as ordinary search. It does not work, and the reason is a single line of documentation that almost nobody has read: "If two results from the same site appeared in a generative AI search results feature, they count as a single impression in the chart total."

The two reports count the same event with different arithmetic. One de-duplicates per site inside a feature; the other applies standard impression rules. Their totals are therefore not on the same scale, and a subtraction between them produces a number that describes nothing. This is a consequence of the two documented rules read together rather than something Google states, and it is worth checking against both pages before anyone builds a dashboard on it.

What a two-month-old property actually sees

This blog is a useful instrument here because it has no authority to protect.

Between 1 August and 3 September 2026 its Search Console recorded 159 impressions across eight named queries in the ai overview family, sitting between positions 87.8 and 100. Over the same window, and over the property's entire life, not one named query containing ai mode appears at all. Search Console anonymizes low-volume queries, so that is zero named queries rather than proof of zero impressions, and the distinction matters more than it sounds: absence in this report is the normal state of a new site, not a diagnosis.

The demand is real and it is elsewhere. The head term google ai mode measures 29,000 monthly US searches at a keyword difficulty of 76, which is why this page targets the question form instead and expects to be read rather than to displace anybody.

Query fan-out, and why the query column will not arrive

Google describes the technique its AI features may use as issuing "multiple related searches across subtopics and data sources", and says AI Mode "is particularly helpful for queries where further exploration, reasoning, or complex comparisons are needed". Both quotes are ordinary product description. Together they carry a consequence that the product description does not draw.

The page that ends up cited is not necessarily the page that ranks for the sentence somebody typed. It is the page that answered one of the subqueries the system wrote for itself. So the list of keywords a site tracks and the list that decides whether it appears have stopped being the same list, which changes what a keyword file is for and is worth reading next to free keyword research.

Follow that one step further and the missing column in the newer report stops looking like an omission. A query dimension needs a query, and on this surface the string that fetched your page was composed by a machine mid-answer. There is no user query to attribute the appearance to in the sense the older report means. That is reasoning from two documented facts rather than a statement by Google, and it is the difference between waiting for a column and planning around its absence.

What is left to do, once the requirement is one sentence

Google's page on AI features is unusually short with the answer: "To be eligible to be shown as a supporting link in AI Overviews or AI Mode, a page must be indexed and eligible to be shown in Google Search with a snippet, fulfilling the Search technical requirements." It adds that "there are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary". Indexing is therefore not a step in this work. It is the gate in front of it, and when a page is missing from the index the order of diagnosis is a separate job, laid out in why a page is not indexed.

The same page closes two doors the checklists keep selling. "You don't need to create new machine readable files, AI text files, or markup to appear in these features." "There's also no special schema.org structured data that you need to add." Google's optimization guide goes further and names one file outright, which is a rare thing for a documentation page to do: "Prioritize effective SEO strategies over 'AEO/GEO hacks': For Google Search, you can ignore tactics like 'chunking' content, creating unnecessary AI text files (like llms.txt), or pursuing inauthentic mentions." What that file was proposed for, and who reads it, is a separate question answered in what llms.txt is.

That quotation deserves an uncomfortable paragraph rather than a quiet one, because this blog writes a self-contained passage of forty to sixty words under a question heading in every article, and Google has just described that as a tactic its Search can ignore. The passage stays and its justification changes. The GEO paper that named the discipline measured gains across several generated-answer engines rather than inside Google alone, and a block that survives being cut out of the page and still makes sense is good writing whoever reads it. What it can no longer be sold as is something Google rewards. The wider version of that job, for systems that compose an answer out of several sources at once, is generative engine optimization.

The middle row is where the editing pays and the bottom row is where the measurement does, and teams reliably spend both budgets on the top row. The mechanics of the requirement itself, sorted by what backs each instruction, are worked through in how to rank in AI Overviews.

One more thing is worth being honest about. On a surface where most queries end with no external click, what a citation buys is closer to recall than to a session, priced the way getting cited by ChatGPT is priced rather than the way a first position used to be.

When AI Mode is not the problem to solve

When the pages are not indexed. Nothing is selected from a document Google has not stored, and no amount of rewriting changes that order.

When the demand is transactional or brand. A written answer cannot complete a purchase, quote your inventory or book an appointment, so the click survives where the decision does.

When the fall has a date on it. Google announces and dates its core updates, and a drop that begins on one of those days has a cheaper explanation than a new surface, which is the first branch in why organic traffic drops.

When nobody will keep the log. The only instrument that matches this surface is a manual monthly check, and building one to abandon it in two months leaves less behind than not starting.

Opting out, and what that decision costs

Google publishes one control here, it applies to the whole property, and it only removes.

The Search generative AI control offers three settings: "Include my site's links and content in Search generative AI features", "Exclude my site's links and content from Search generative AI features", and "Inherit control from parent". Inclusion is the default for all properties. On exclusion, Google is blunt about the consequence: "You won't receive any traffic or impressions from these features." Two clarifications sit alongside it and both matter to the decision. The control "isn't used as a ranking or inclusion signal affecting other parts of Search", and it "doesn't affect AI training", which runs through Google-Extended instead.

So the asymmetry is documented rather than inferred: there is a switch that takes a site out and none that puts it further in.

The second cost of pulling it is the one nobody prices. Excluding the site empties the generative AI performance report along with the appearances, so a publisher who opts out to protect traffic also destroys the only instrument that would have shown what the exposure was worth. The decision becomes permanent in a specific way: no measurement survives with which to reverse it.

Mistakes that repeat

• Reading it as AI Overviews with a worse temper. One is a block above a list that stays. The other is a surface entered instead of the list.
• Waiting for a clicks column in the generative AI report. It carries impressions and four ways to group them, and that is the whole report.
• Subtracting one report's impressions from the other's. Two results from the same site count once in the generative AI chart total, so the two are not on the same scale.
• Tracking only the query somebody typed. The appearance answers a subquery the system wrote, and that string is not published in any tool.
• Excluding the site to protect traffic. It removes the appearances, returns none of the clicks and empties the only report that measures the surface.
• Judging a quarter on one screenshot. The answer is composed per query, so the same question can return a different set of links the next day.

Data and transparency

Six Google pages carry every quotation in this article and all six were opened on 6 September

1. From the documentation on AI features in Search, last updated 10 December 2025: the eligibility sentence, the statement that no additional requirements or special optimizations exist, the two sentences about machine readable files, AI text files, markup and schema.org structured data, query fan-out described as issuing multiple related searches across subtopics and data sources, the note that AI Mode and AI Overviews may use different models and techniques, and the placement of AI feature traffic inside the performance report under the Web search type. From the Search Console performance report reference: the click, impression and position rules for both surfaces, including the single-position rule for AI Overviews and the results-page methodology for AI Mode. From the generative AI performance report help page: the definition of an impression there, its four grouping tabs, the absence of clicks, click-through rate, position and queries, and the rule that two results from the same site count as a single impression in the chart total; the announcement date of 3 June 2026 comes from the Search Central post that introduced those reports. From the guide to optimizing for generative AI features, last updated 10 July 2026: the sentence directing site owners to that report and the quotation about prioritizing effective SEO over AEO and GEO hacks, which names chunking, unnecessary AI text files such as llms.txt and inauthentic mentions. From the Search generative AI control help page: its three settings, its default, the sentence about receiving no traffic or impressions after exclusion, and the two clarifications about ranking signals and about model training running through Google-Extended.

Two things in this article are reasoning rather than documentation, and they are marked as such where they appear. That the impression totals of the two reports cannot be subtracted from each other follows from the two counting rules read together, and Google states neither the conclusion nor its opposite. That a query dimension is unlikely to arrive for this surface follows from fan-out being machine-composed, and it is an expectation rather than a roadmap.

The figure of 93 % of AI Mode queries ending without a click to an external site has two independent measurements behind it, which is what makes it the most solid number in this subject: Semrush, over 69 million US desktop sessions between 1 May and 5 July 2025, and Seer Interactive, over 25.1 million impressions; both were recorded for this project's data file on 12 August 2026, and the Seer page linked below is that firm's published study, from which its sample is taken. The 58 % figure comes from Ahrefs, published 4 February 2026, over 300,000 keywords split evenly between queries with and without an AI Overview, comparing December 2023 with December 2025 on aggregated Search Console data. It is a correlation between the presence of a summary and average click-through, not a loss followed on a fixed set of pages, and it belongs to AI Overviews: no published measurement gives an equivalent figure for what AI Mode costs an individual site, and none is invented here. The Pew Research Center study of 22 July 2025, on the browsing data of 900 US adults covering 1 to 31 March 2025 with results pages collected 7 to 17 April 2025, is the only one in this area that watched real navigation rather than keyword lists: 68,879 unique Google searches, 12,593 of which produced a summary, 18 % presence, a click on a link inside the summary in 1 % of visits, and 88 % of summaries citing three or more sources. It measured summaries and not AI Mode, and is cited here for that boundary as much as for its numbers. The relative improvement of 30 to 40 % on Position-Adjusted Word Count from adding sourced figures comes from the GEO paper by Aggarwal, Murahari, Rajpurohit, Kalyan, Narasimhan and Deshpande, arXiv 2311.09735, presented at KDD 2024, over roughly 10,000 queries.

This site's own numbers are exact because they are its own, queried on 6 September 2026: 159 impressions across eight named queries in the ai overview family between 1 August and 3 September 2026, in positions 87.8 to 100, and no named query containing ai mode in the property's entire history, which is zero named queries and not proof of zero impressions, since Search Console anonymizes low-volume queries. The search demand for this article's query, 600 monthly searches in the United States with no difficulty score returned and a cost per click near 1.20 USD, and for the head term google ai mode, 29,000 monthly US searches at a keyword difficulty of 76, was measured in Ahrefs on 6 September 2026 for this blog's keyword research. That five pages on the first results page explain the product to its user and that none of them names the generative AI performance report is my own reading of those five pages on 6 September 2026, and it is a statement about those five rather than about the wider literature. The operational judgment throughout comes from work across a portfolio recording more than 300 million impressions a year in Search Console. Verified as of September 2026.

Primary sources, opened on 6 September 2026: Google's documentation on AI features in Search; the Search Console performance report reference; the generative AI performance report help page; the Search Central announcement of those reports; Google's guide to optimizing for generative AI features; the Search generative AI control; the Ahrefs February 2026 update; the Seer Interactive study; the Pew Research Center short read; the GEO paper (arXiv 2311.09735).

What this changes

For twenty years the bargain between Google and a site was legible, and the legibility was half the product. Google told you the query, the position and the click, and the performance report was a receipt you could argue with.

The generative AI report is the first Google surface where the company confirms that you appeared and declines to say for what, to whom, or with what result. The interesting part is not that a column is missing. It is which column: the query, absent because on this surface a person never wrote one. Everything else in that report follows from it, and so does most of the confusion around the subject. The work that comes next is not harder than the old work. It is the same editing, judged by an instrument honest enough to publish what it cannot see.

Frequently asked questions