Why Rankings Matter Less in AI Overviews

What changed when AI Overviews arrived

For twenty years the model of search engine optimization was stable: earn a position on the results page, then earn the click. Position and traffic moved together closely enough that average position worked as a proxy for visibility. Rank higher, get more clicks, get more enquiries.

AI Overviews break the second half of that chain. Google now composes an answer at the top of the search engine results page for a growing share of informational queries, drawing on multiple sources and linking to them. The panel evolved from what Google originally launched as Search Generative Experience, and it sits above the traditional blue links that SEO has always competed for.

The result is that the position you occupy and the visibility you actually receive have come apart. You can hold the top organic position on a query and still watch the click-through rate fall, because the searcher got a sufficient answer before reaching your listing. You can also be cited inside the panel while ranking well below the first position, and see referral traffic that your rank tracker cannot explain.

Ranking first and being cited are two different selections

This is the point most reporting misses. Ranking is a judgement about which pages best satisfy a query overall. Citation inside an AI Overview is a judgement about which passages best support a specific statement the model is composing. These are related but separate decisions, made at different levels of granularity.

A comprehensive guide can rank first because it covers a topic thoroughly, while a shorter page that answers one sub-question in a single clean sentence is the more quotable source for that sentence. Depth wins the ranking; clarity wins the citation. Both matter, but they reward different things, and a page optimised only for the first can lose the second.

Google has been consistent that there is no separate submission process, no markup and no paid placement that gets a page into an AI Overview. Eligibility follows from ordinary indexing: the page has to be crawlable, indexable and permitted to show a snippet. Publishers who use the nosnippet or max-snippet robots directives, or the data-nosnippet attribute, restrict the text Google may display, which also restricts what can be used in the panel. That is the only real lever, and it works by exclusion rather than inclusion.

The measurement problem nobody warns you about

Google Search Console does not report AI Overview impressions and clicks as their own dimension. Appearances within the panel are folded into the standard performance report alongside conventional web results, which means you cannot cleanly separate the two in the interface you already trust.

This produces a specific and confusing pattern. Impressions hold steady or rise. Average position looks healthy, sometimes better than before. Clicks fall. Nothing in the report explains the gap, and the instinctive response is to chase the ranking that already looks fine.

The same disconnect appears outside Google. ChatGPT search, Perplexity and Microsoft Copilot all compose answers over retrieved sources, and none of them report to your analytics the way a search engine referral does. Traffic that arrives after someone read about you in a generated answer usually shows up as direct or branded search, well after the interaction that caused it.

What still depends on rankings

It would be a mistake to read any of this as the end of SEO. AI Overviews draw on the same index, and the retrieval that feeds them favours pages the search engine already considers relevant and trustworthy for the topic. Ranking well remains the most reliable route into the candidate set.

Rankings also still convert directly on the queries that matter most commercially. Transactional and navigational searches — someone looking for a supplier, a price, a location or a specific brand — are far less likely to be resolved by a generated summary, because the searcher wants to reach a business rather than learn a fact. Informational queries at the top of the funnel are where the panel intercepts most traffic.

So the honest framing is not that rankings stopped mattering. It is that ranking is now a necessary condition rather than a sufficient one, and that it predicts traffic much less reliably than it used to on one specific class of query.

What to optimise for instead

The work that improves citation odds is unglamorous and largely overlaps with writing well.

• Answer the question in the first one or two sentences under a heading that matches how the question is actually asked, then supply context beneath it.
• Write definitions that stand alone, so a quoted sentence still makes sense when it is lifted away from the surrounding page.
• Prefer specific entities, figures and dates over vague qualifiers, because a claim that can be checked is a claim that can be safely repeated.
• Use structured data from schema.org — Article, FAQPage, Service, Organization and BreadcrumbList — so the boundaries and subject of a page are explicit rather than inferred.
• Keep business and product information consistent everywhere it appears, so retrieval systems resolve you to one entity instead of several partial ones.
• Show who wrote the content and on what basis. First-hand experience, named authorship and cited evidence are what Google's own quality guidance describes under E-E-A-T, and they are also what makes a source safe to quote.
• Fix the technical foundations first. A page that cannot be crawled, indexed or snippeted cannot be cited, no matter how good the writing is.

How to measure visibility when position stops predicting traffic

If average position no longer answers the question of whether search is working, replace it with measures that survive the change.

• Organic clicks and click-through rate by query group, watched for the divergence between stable impressions and falling clicks.
• Branded search volume, which tends to rise when a brand is being named in generated answers people do not click.
• Direct and unattributed traffic, read as a trend rather than a number, since assistant referrals often arrive this way.
• Assisted conversions and enquiry quality, because a smaller volume of better-qualified visitors can be a better outcome than the traffic it replaced.
• A manual sample of the questions that matter to your buyers, run against AI Overviews and the major assistants on a fixed schedule, recording whether you appear and how you are described.

What this does not mean

Two overcorrections are worth avoiding. The first is abandoning depth in favour of short answer-shaped pages. Thin content does not rank, and a page that never enters the candidate set cannot be cited from it. Depth earns the position; structure earns the quote; you need both.

The second is treating citation as a controllable placement. No agency can guarantee that a business will appear in an AI Overview or in any assistant's answer, because the selection is made by a system that nobody outside the platform controls and that changes without notice. Anyone offering that guarantee is describing something they cannot deliver.

The realistic goal is to be the clearest, most verifiable source on the questions your buyers actually ask, and to measure whether that is producing enquiries. That was good practice before AI Overviews existed. What has changed is that the reward for doing it well now shows up somewhere other than your average position.