AI Overviews: what they are and what they cost in clicks

Google's documentation answers this query in one sentence, and the four non-Google pages on its first results page, all opened on 5 September 2026, do not quote it. What circulates instead is a presence percentage, and there are at least four of those in play: 25.11%, 43%, 48% and 60.32%, from four studies that are not comparable with one another. The number that would change what you do is in none of them.

What are AI Overviews?

An AI Overview is the summary Google writes above the search results, with links to the pages it drew on. Google requires only that a page be indexed and eligible to be shown with a snippet. A new site is rarely cited; a mid-authority site loses informational clicks; a consolidated site gets cited and loses clicks at the same time.

Quick answer

• The requirement is one sentence and it is not new. Google documents that no additional optimization, machine readable file or markup is needed to appear.
• The click evidence is a correlation, not a loss you can trace. Ahrefs measured a 58% lower average click-through rate on keywords carrying a summary, comparing December 2023 with December 2025.
• Being cited is not being visited. Pew Research Center recorded a click on a link inside the summary in 1% of visits to search pages that had one.
• Search Console will not separate it out. An AI Overview occupies a single position and every link inside it is assigned that same position.

Percentages from the Pew Research Center browsing study of March 2025, described below with its sample. They are shares of searches of that shape, not shares of your traffic.

What kind of problem this is

The model most people bring to this topic has one number in it: AI Overviews take some percentage of organic traffic, the percentage is knowable, and the response is to get cited so the traffic comes back.

There are three separate events in that sentence and it collapses them into one. A summary appearing on your query. Your page being named inside it. Somebody clicking the name. The published measurements cover the first and the third, on different populations in different years, and the second is the only part anybody sells. A plan that assumes the three move together will buy citations and count visits, which is the one combination guaranteed to look like failure whatever happens.

The correct model is duller. This is a change in what a results page is for, and the useful question is which of your queries now get answered before anyone reaches the list.

What Google documents, and what it counts

Two pages settle most of the arguments and take four minutes to read.

On eligibility, Google's documentation on AI features in Search states that "there are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary". AI Mode is the other surface named in that sentence and it is not this one: it is a separate conversational tab, and what it does to a report is different enough to have its own page in what Google AI Mode is, that "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", and that "you don't need to create new machine readable files, AI text files, or markup to appear in these features". The controls that limit what gets shown already existed: nosnippet, data-nosnippet, max-snippet and noindex. That is the whole entry fee, and it is the fee what SEO is has always charged.

On counting, the Search Console performance report reference is blunter than most descriptions of it. Clicking a link to an external page in an AI Overview counts as a click. Standard impression rules apply. And the line that matters operationally: an AI Overview occupies a single position in search results, and all links in the AI Overview are assigned that same position.

That last rule has a consequence nobody mentions. Cited third inside a summary sitting above everything, your page is not reported third: it is reported at whatever position the summary occupies, along with every other link in it. Average position for that query can improve while clicks fall, and both numbers are correct. Reading those two columns as one story is the same failure mode as diagnosing a drop in organic traffic from a total.

Three measurements, three different answers

The three most cited studies measured three different things, and knowing which is which is most of the skill.

Ahrefs, published 4 February 2026. Three hundred thousand keywords, half of them carrying an AI Overview, on aggregated Search Console data, December 2023 against December 2025. The presence of an AI Overview now correlates with a 58% lower average click-through rate for the top-ranking page. The word doing the work is "correlates": no fixed set of pages was followed through time, so it is not a receipt for clicks a particular page lost. In the same update Ahrefs superseded a smaller figure of its own from April 2025, which is worth knowing before quoting anything here as settled.

Seer Interactive, published 4 November 2025. June 2024 to September 2025, across 3,119 informational queries and 25.1 million organic impressions. Click-through rate on queries with an AI Overview went from 1.76% to 0.61%, which Seer reports as a 61% decline. On queries without one it went from 2.72% to 1.62%, a 41% decline. The second series changes the reading: click-through rate fell where no summary appeared either, so the part attributable to the summary is the difference between the two declines, not the whole of the first.

Pew Research Center, published 22 July 2025. This one did not read Search Console at all. It used the browsing data of 900 US adults who agreed to install a tracker, covering every URL they visited during March 2025, and recorded what each did next after a search. The dataset holds 68,879 unique Google searches, of which 12,593 produced a summary. People who saw one clicked a traditional result in 8% of visits, against 15% for those who did not, and 26% of visits to a page with a summary ended the browsing session outright, against 16% without one.

Direction agrees across all three. Magnitude does not, and averaging them produces a number that describes no population.

The same split runs through the presence figures. Pew found 18% of the Google searches in its March 2025 dataset produced a summary. Keyword-list studies for 2026 run from 25.11% (Conductor, 21.9 million queries, first quarter of 2026) to 60.32% (Xponent21, United States, April 2026), with Similarweb at 43% and BrightEdge at 48% between them. Two things separate the low figure from the high ones and both are real: Pew counted searches people actually ran, where short navigational queries dominate, while the others counted keywords off a list; and a year passed, during which the feature expanded.

The gap between being cited and being visited

Here is the number the citation market does not put on its slides.

People who encountered a summary clicked a link inside it in 1% of all visits to pages with one. Not 1% of the summaries, not 1% of the citations: 1% of the visits. In the same dataset, 88% of summaries cited three or more sources and only 1% cited a single one, and the typical summary ran to 67 words.

A citation is therefore not a placement. It is a share of something small: one of at least three names attached to an event that happens once in a hundred visits. The mention still has value, closer to what getting cited by ChatGPT returns than to what a first position used to return. What it is not is a replacement for the session, and a plan priced as if it were is priced against a 1% event.

The source mix says something similar. Wikipedia, YouTube and Reddit together supplied 15% of the sources listed in the summaries examined, close to their 17% share of standard results, and government sites appeared in 6% of summary sources against 2% of standard ones. This is not a new game with new entrants. It is the old index, quoted, with a tilt toward institutions.

What a two-month-old site sees from the bottom

This blog is a useful instrument here precisely because it has no authority yet.

Between 1 August and 3 September 2026, its Search Console recorded 159 impressions across the ai overviews family: 50 for ai overviews, 44 for google ai overviews, 24 for ai overviews google, 21 for ai overview, and the rest across variants. Every one landed between positions 87 and 96, and at the time of that measurement the site had no page in English on the subject.

Two things follow, and the second is easy to miss. Demand reaches a site that has never written about the term, which is what 4,600 monthly US searches at a keyword difficulty of 5 look like from underneath. And 159 impressions at position 90 are worth zero visits, so a report showing them as growth would be describing a queue rather than an audience. Impressions say the query exists and that Google has something of yours it could show. They are not traffic, and the distance between the two is the whole subject of how to tell whether SEO is working.

Where the work moves

Split the queries before reading any total

A single traffic line mixes queries that lost most of their clicks with queries nothing touched, and produces an average that happened to nobody. The minimum split is informational against transactional and brand, compared on a fixed list across two months. Impressions holding while clicks fall in the informational group puts the cause above the results rather than in them, and that comparison starts with the tool: setting up Google Search Console.

Write the block a summary can lift

The extracted unit is a self-contained passage: entity named inside it, claim in the first sentence, no pronoun leaning on the paragraph above. Pew's median summary ran to 67 words, so that is roughly how much of an answer this surface carries. A page can be excellent and contain no such block, in which case it competes for the click and never for the quote. That formatting layer is the whole of answer engine optimization.

Systems that compose an answer out of several sources at once, rather than lifting one passage from one page, reward the same writing plus something extra: a claim specific enough that a machine can attribute it to you by name. That wider version of the job is generative engine optimization.

Spend where the summary is rare

Pew's breakdown says where that is: 8% of one and two word searches produced a summary against 53% of searches of ten words or more. Short, navigational, brand and transactional demand sits on the low side. Local intent splits differently again, and that split is covered in AI Overviews in local search.

When this is not your problem

When the pages in question are nowhere near the first results page. Nothing is extracted from a position nobody reaches, so the prior job is ranking.

When the drop coincides with a core update. Google announces those and dates them, so the check takes a minute and settles the question before any rewriting starts. Blaming a summary for a fall that began on the day a core update rolled out leads to rebuilding pages that had nothing wrong with them.

When brand queries are most of the traffic. Summaries barely appear on them, and any conclusion drawn from that segment will be wrong about everything else.

Mistakes that repeat

• Quoting one presence percentage. The published range runs from 18% to 60.32% depending on who measured, on what and when, so a bare number carries a precision that does not exist.
• Reading the correlation as a receipt. A 58% lower average click-through rate on keywords with a summary is not a measurement of clicks a specific page lost.
• Buying citations and counting sessions. The two are different events and the second is a 1% event, so the plan reports failure even when it worked.
• Blocking AI crawlers to protect traffic. It removes the page from the citation pool and returns none of the clicks.
• Using a rank tracker to assess this. A tracker records a position; it does not record whether a summary appeared or who it named.
• Rewriting the whole site. The reformatting pays only where impressions hold and clicks fall, and that is a subset, not a sitewide project.

Data and transparency

The eligibility quotes, the statement that no new machine readable files, AI text files or markup are required, and the list of nosnippet, data-nosnippet, max-snippet and noindex as the controls that limit what is shown, come from Google's documentation on AI features in Search, opened on 5 September 2026. The single-position rule, that clicking an external link in an overview counts as a click, and that standard impression rules apply, come from Google's Search Console performance report reference, opened the same day.

The 58% comes from the Ahrefs study of 300,000 keywords, 150,000 with an AI Overview and 150,000 informational keywords without one, on aggregated Search Console data, December 2023 against December 2025, published 4 February 2026. It is a correlation with the presence of a summary, not a loss measured on a fixed set of pages, and it is quoted here in those terms. The paired decline comes from Seer Interactive, published 4 November 2025, over 3,119 informational queries across 42 organizations and 25.1 million organic impressions from June 2024 to September 2025: 1.76% to 0.61% with a summary, reported by Seer as a 61% decline, and 2.72% to 1.62% without one, a 41% decline, cited together because the second changes how the first reads.

The behavioural figures come from the Pew Research Center short read of 22 July 2025, on the browsing data of 900 US adults from the KnowledgePanel Digital panel covering 1 to 31 March 2025, with results pages collected 7 to 17 April 2025: 68,879 unique Google searches of which 12,593 produced a summary, the 8% against 15% click rate on traditional results, the 1% click rate on links inside the summary, 26% against 16% session endings, 18% of searches producing a summary, 88% citing three or more sources, the median of 67 words, the source-mix shares and the breakdown by query shape. Presence figures for 2026 are Conductor 25.11% (21.9 million queries, first quarter of 2026), Similarweb 43%, BrightEdge 48% and Xponent21 60.32% (United States, April 2026), recorded on 12 August 2026 and quoted with their fiches because they are not comparable and are not averaged here. No figure is given for what an individual site loses to this feature: none of the three studies measures that, and the number that would answer it can only come from the reader's own report.

This site's own numbers are exact because they are its own: 159 impressions across the ai overviews family between 1 August and 3 September 2026, in positions 87 to 96, from its Search Console, queried on 5 September 2026. The demand for this query, 4,600 monthly US searches at a keyword difficulty of 5 and a cost per click near 1.10 USD, was measured in Ahrefs on 5 September 2026 for this blog's keyword research. That the four non-Google pages on the first results page (Wikipedia, Botify, Semrush and Zelst) do not quote Google's eligibility sentence is my own reading of that results page the same day, and is a statement about those four pages 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 5 September 2026: Google's documentation on AI features in Search; the Search Console performance report reference; the Ahrefs February 2026 update; the Seer Interactive study; the Pew Research Center short read.

What this changes

The interesting thing about the three studies is not that they disagree. It is where they agree.

Ahrefs and Seer measured ratios out of aggregated reporting and produced large, quotable declines. Pew watched what nine hundred people actually did and found that around two thirds of all searches, with a summary or without one, ended without anyone clicking a result. The summary made that worse by a measurable amount. It did not invent it.

So the strategic question was never whether AI Overviews take traffic. It is whether a business was ever entitled to the visit, or whether it had been living on a transaction Google was under no obligation to keep offering. The sites that will be fine are the ones whose demand consists of decisions nobody can make from a paragraph. That was true before any of this, and all the summaries did was make it expensive to keep pretending otherwise.

Frequently asked questions