How AI Overviews Are Changing What Counts as SEO
Google's own summary now sits above the results it used to send traffic to. Here is what that does to a click, and what actually still earns one.
10 min readBy Somangsu Mukherjee
For twenty years, ranking first meant a click. That link between position and traffic is the thing AI Overviews are breaking. Google now answers a growing share of searches inside the results page itself, before a searcher ever reaches a list of blue links, and a page that used to earn a visit by ranking well can now lose that visit to a summary built from several pages at once, including, sometimes, its own.
This article sets out what actually changed, in plain terms, and what still earns a click once the summary above the fold has already answered the easy version of the question.
Key Takeaways
- An AI Overview answers the query directly, above the ordinary results, which is why more searches now end without a single click going anywhere.
- The model behind it does not just match the top query. It fans a single search out into several related sub-questions and pulls a source for each one, so a page can be cited without ranking first for the headline phrase.
- E-E-A-T and other authority signals matter more under this model, not less, because being selected as a trustworthy source is now a precondition for being cited at all.
- Structural clarity, direct answers near the top, and clearly formatted lists and steps are what a summarising model can lift cleanly. A page written for skimming humans and a page written for extraction by a model reward largely the same habits.
- Branded search and long-tail, specific questions are where a click still survives, because a generic query is exactly the kind an AI Overview is most likely to answer completely on its own.
- Google’s own name for the underlying shift is generative AI search. AI Overviews are the most visible surface of it, not the whole of it.
AI Overviews Are Turning Ranking Into Something Else
Search Engine Optimisation was built on a simple mechanic: rank higher, get more clicks. That mechanic assumed the searcher had to visit a page to get an answer. An AI Overview removes that assumption for a large share of informational queries, because Gemini, the model family behind Google’s own summary, reads across several ranking pages and writes the answer itself, directly on the results page, before the searcher scrolls any further. The same underlying shift shows up beyond Google’s own results page too: ChatGPT and Perplexity answer a search directly inside their own interface rather than sending a searcher out to a ranked list at all, which is why this is a change to how a search gets answered generally, not a single new box bolted onto one results page.
The immediate, measurable effect is a rise in zero-click searches: queries that end with the searcher satisfied and no page visited at all. That is not a ranking problem in the traditional sense. A page can hold its position, or even improve it, and still see fewer visits, because the position itself stopped being the thing that decided whether a click happened. What decides it now is whether the AI Overview already gave a complete enough answer that nobody needed to go further.
That does not mean visibility has stopped mattering. It means visibility has split into two separate questions: does this page still rank, and separately, does this page get cited as one of the sources behind the generated answer. A business optimising only for the first question is optimising for a mechanic that decides fewer outcomes than it used to.
Underneath that shift sits query fan-out, the technique that changes what “ranking for a keyword” even means. The model writing an AI Overview does not simply summarise whichever page ranks first for the exact phrase typed into the search box. It breaks that phrase into a set of related sub-questions the searcher probably also has, a technique referred to as query fan-out, and it selects a source for each of those sub-questions independently. A single search can trigger several small retrievals behind the scenes, each one an opportunity for a different page to be the source cited for a different piece of the final answer.
The practical consequence is that a page no longer needs to win the single, most competitive version of a query to be part of the answer. It needs to be the clearest, most specific source for one of the several sub-questions a fan-out generates around that query. That rewards depth on a narrow point over breadth attempted across a broad one, and it is part of why long-tail keywords, specific enough that the model has not already synthesised an answer to them from elsewhere, are worth more deliberate attention than they were when a single head-term ranking was the whole game.
Underneath the fan-out sits a mechanism worth naming plainly: retrieval-augmented generation, the process of pulling live, specific source material into an answer rather than relying only on what a large language model already learned during training. That is also what makes grounding a meaningful check on the result. A generated answer that is properly grounded points back to a real, checkable source for each claim it makes; one that isn’t can still read as confident while stating something wrong, which is exactly what a hallucination looks like from the outside, fluent and plausible right up until someone checks it.
Authority and Structure Decide Who Actually Gets Cited
A search engine ranking page and a generative model choosing what to cite are answering related but different questions. Ranking has long weighed relevance and authority together. A model assembling a generated answer has an additional, narrower filter on top of that: it is choosing which handful of sources to actually attribute a claim to, in an answer it is presenting as reliable. That is closer to an editorial decision than a ranking decision, and it leans harder on the same signals search engines have described for years under E-E-A-T: demonstrated experience, subject-matter expertise, authoritativeness, and trustworthiness.
A page can be technically well-optimised and still not read, to a model deciding what to cite, as a source worth attributing a claim to. Clear authorship, evidence of direct experience with the subject rather than a rewritten summary of other pages, and a track record the model can find corroborated elsewhere all function as the trust signal that gets a page selected as a source in the first place. Being well-written is necessary but no longer sufficient; being credibly the kind of source an answer should be attributed to is the harder bar underneath it.
Clearing the authority bar is only half the job. A separate and more mechanical question decides whether it actually gets used: can the model extract a clean, self-contained answer from it without doing a lot of interpretive work. A direct answer stated plainly near the top of a section, a clearly labelled list or set of steps, and specific facts stated in a sentence that stands on its own without needing the paragraph before it, are all easier for a summarising model to lift intact than the same information buried in a long, discursive paragraph a human reader might still enjoy.
This is largely the same discipline good web writing has always rewarded: say the answer first, format for skimming, keep a claim specific rather than vague. What has changed is who is doing the skimming. A page written to be extracted cleanly by a model and a page written to be scanned quickly by an impatient human turn out to want nearly the same thing, which is one of the few pieces of this shift that does not require an entirely new set of habits, only more consistent adherence to the old ones. Content freshness and factual specificity sit alongside structure here: a model choosing between two structurally similar sources will lean towards the one with a checkable date and a specific figure over the one making the same claim in vaguer terms.
Branded Search Is Where the Click Still Survives
When a generic, informational query gets fully answered inside an AI Overview, the click that used to go to whichever page ranked first for that phrase often does not happen at all. What does not disappear is a searcher’s tendency to search for a business by name once they already know it, and that branded search still reliably sends a visit to that business’s own site, because a model has far less reason to intercept a query that is already asking for one specific, named source.
That makes brand recognition a more direct SEO asset than it used to be, not just a marketing nice-to-have sitting next to it. A business a searcher has never heard of has nothing to fall back on once the generic version of its category’s query gets absorbed into a generated summary. A business searchers already know how to name still gets found deliberately, on a query an AI Overview has much less reason to try to answer for them.
What to Track, and What It Means for a Content Strategy
The old dashboard, rank tracked against traffic, no longer tells the whole story, and a business that keeps watching only those two numbers will misread what is actually happening to its own visibility. Search Console already separates out impressions and clicks specifically tied to appearances inside an AI Overview, and the shape worth watching for is impressions holding steady, or even climbing, while clicks on the same query drift down. That is not a tracking error. It is the query being answered on the results page itself, and it is the single clearest signal that a page’s visibility and its traffic have started to move independently of each other.
A second split worth building into any regular reporting is branded against non-branded query performance, tracked separately rather than blended into one traffic total. A page can lose ground on its generic, category-level queries while holding or gaining on searches for the business by name, and blending both into a single line hides exactly the shift this article describes. Watching them apart is what shows whether a drop in overall clicks is the market changing or the brand-building actually working.
The third habit is simply checking, periodically and by hand, what an AI Overview actually says for the handful of queries that matter most to a given business, and which sources it names while saying it. A page that used to rank first can find a competitor cited in its place, or find itself cited only for a narrower sub-question than the one it was written to answer. Neither shows up in a rank tracker. Both show up the moment someone actually reads the generated answer instead of only the position underneath it.
None of this is a reason to abandon the discipline that has always underpinned good SEO: relevant content, clearly written, technically sound, and built around what a real searcher actually needs. What changes is where the marginal effort should go. Depth on specific, long-tail questions earns citations a broad head-term chase will not. Clear authorship and demonstrated experience decide whether a page is even in the running to be a cited source. Direct, well-formatted answers near the top of a page are what get lifted cleanly once it is. And branded search, built over time rather than optimised for in a single sprint, is the query type a generated summary is least likely to intercept.
Most of that work is not a rewrite. It is an audit of what a site already has, run through the five-step audit methodology and checked against a genuinely different question than the one SEO used to ask: not only “does this rank,” but “would a model reading across the top sources choose this page as one worth citing.” Being in the running to be cited is a separate question from what gets content cited in AI-generated answers once it is, and a business getting the first question wrong never reaches the second. That distinction is what an AI audit is built to check, and every audit here starts with a scoping conversation: get an audit to have a site checked against generative AI search directly, rather than against the ranking mechanic it is steadily displacing.
Sources
- Search Engine Land’s guide to optimising for AI Overviews
- Search Engine Journal on the impact of AI Overviews on publishers, and how to adapt into 2026
- WordStream on how AI Overviews are impacting SEO, and what to do about it
- Semrush’s complete guide to AI content optimisation
- Google Search Central’s guide to optimising a website for generative AI features
FAQ
Questions we get asked
What is an AI Overview?
An AI Overview is the generated summary Google places above its ordinary results, written by a model that reads across several sources and answers the query directly rather than sending a searcher to a page to find the answer themselves. Google's own documentation calls the underlying category generative AI search, which is the more accurate label since AI Overviews are one surface within it, not the whole of it.
What is a zero-click search?
A zero-click search is one where the searcher gets a satisfying answer without visiting any page, because the summary above the results already answered the question. AI Overviews increase how often that happens for informational queries, which is why a ranking position that used to guarantee a click no longer guarantees one.
What is query fan-out and why does it matter for SEO?
Query fan-out is what happens when the model behind an AI Overview breaks one search into several related sub-questions and pulls a source for each, rather than answering from a single top-ranking page. A page can lose the main query and still earn a citation through one of the sub-questions the fan-out generates, which is why targeting only the exact phrase in the search box is no longer the whole job.
Does E-E-A-T still matter with AI Overviews?
It matters more, not less. A model deciding which sources to cite in a generated answer is making the same judgement a human editor makes about which source to trust, so the signals that make a page look experienced, expert, authoritative and trustworthy are what get it selected as a source in the first place, before the citation question is even reached.
Why has branded search become more important since AI Overviews?
When a generated summary answers the generic question directly, the click that used to go to whichever page ranked first now often goes nowhere, or goes to whichever brand the searcher already recognises and searches for by name. A business that never earns that name recognition has nothing left to catch once the generic query itself stops sending traffic.
Should long-tail keywords replace short, high-volume ones?
Not replace, but they are worth more attention than before. A short, high-volume query is exactly the kind an AI Overview is most likely to fully answer itself, leaving nothing for a ranking page to add. A specific, long-tail question is more likely to need a source the model has not already synthesised, which is where an actual click still survives.