A page can rank near the top of Google and still be absent from the answer above it. Since the 2023 Search Generative Experience test, that gap has become the point: ranking still matters, but it no longer earns the citation by itself.

Read the changes in order and the practical response gets clearer. An answer-first page gives Google something usable. A clear business identity helps it understand who supplied that answer. Deep pages and third-party mentions give it more than one place to find you. Ads entering the answer do not make that organic work obsolete. They put another claimant in the same small patch of screen.

10 May 2023: SGE makes the top result a plausible citation

Google put Search Generative Experience into Search Labs as an opt-in test on Chrome desktop and the Google App in the US. You had to turn it on. That matters: Google could test long, synthesized answers with people who had asked to see them, rather than make the format the default for everyone.

Early SGE answers drew heavily from pages that already ranked, stitched a summary above the results, and displayed links alongside it. A strong ranking made a page a plausible source for the snapshot. It was reasonable, then, to treat position as the main route into the answer. Reasonable is not the same as permanent.

The limitation was easy to miss because the old and new systems appeared on one page. The familiar results still sat underneath; the generated answer borrowed from them often enough to look like another reward for ranking well. But a citation was already a separate decision. Google had to select a source for an answer, not merely display the next result in order. The 2023 lesson was to rank well. The warning was that ranking and citation had become different outcomes.

14 May 2024: AI Overviews make citation a separate target

On 14 May 2024, Google rolled AI Overviews out to everyone in the US without a Labs sign-up. The feature used a custom Gemini model for multi-step reasoning. A test shown to willing users had become an answer placed in front of people who might never have requested one.

That wider job called for answers assembled across more sources, not just the most obvious ranking page. The difference shows up in subsequent tracking: overlap between cited pages and organically ranking pages stood at 32.3% in the early months. That figure does not mean rankings stopped mattering. It means a ranking report could no longer tell you, on its own, whether your page appeared in the answer.

This is where the old checklist breaks. Improving a page from fifth to third may help its visibility in the conventional results, but it does not tell Google which sentence answers the question or make the page easy to use in a synthesis. If a team reports position as though it were citation share, it is measuring the wrong shelf. From May 2024 onward, check the answer itself, then make the relevant page capable of supplying one.

May 2024: Google announces ads for the answer

At Marketing Live 2024, Google announced tests of Search and Shopping ads inside AI Overviews. The ads would appear in a sponsored section when relevant to both the query and the generated answer. Existing Search, Shopping, and Performance Max campaigns could be eligible; there was no separate Overview buy.

That distinction matters for advertisers. You cannot treat the answer box as a new placement to purchase and then ignore the rest of the campaign. Google decides whether an existing ad fits the answer. The announcement also changed the organic calculation: a citation would have to earn attention in a space where a sponsored result might appear nearby. Paid access to the answer did not replace the need to be cited in it.

On 28 October 2024, Google brought AI Overviews to more than 100 countries in six languages and said the feature reached more than a billion monthly users. In the same update, desktop links appeared in a right-hand display and inline with the answer text rather than only below it.

The reach is the headline number. The link placement is the change a page editor can act on. When a link sits beside the claim it supports, the cited passage matters. A page whose useful answer is buried under a long introduction makes that job harder than a page that states it plainly beneath a relevant heading.

For businesses outside the first US rollout, the expansion also made the question immediate rather than theoretical. More audiences could encounter a synthesized answer before they reached conventional results. That did not make every query an Overview or every citation a click. It did make a clean, direct answer worth writing even when the page already ranked.

I would start with the sentence a reader actually needs: what the thing is, what it costs, how long it takes, or where it applies, depending on the query. Put the supporting detail below it in an order that makes sense without a search engine. If your key page cannot give the answer clearly, a prominent inline link has nothing useful to point to.

November 2024: Ads appear below the summary

The first live AI Overview ad placements were spotted in November 2024 on mobile in the US. They sat below the summary, separated from organic citations. The distinction was visible: sponsored placement on one side, cited sources on the other. Both were now competing for attention within the answer experience.

For practitioners, this was less a reason to chase a new ad setting than a reason to inspect two existing jobs. On the paid side, the query match, product data, headlines, and assets still had to make sense when shown beside a generated answer. On the organic side, the page still had to give that answer a useful source to cite. An ad might be eligible through an existing campaign; an organic citation had to stand on its own. Do not mistake presence in one part of the box for coverage of the other.

May 2025: Desktop ads make the overlap harder to ignore

At Marketing Live in May 2025, Google expanded AI Overview ads to desktop in the US, with rollout to select English-language countries on mobile and desktop later that year. It also began testing ads in AI Mode for longer conversational queries.

The mechanism stayed important. Advertisers were not bidding directly for an Overview citation or buying a separate answer-box campaign. Existing campaigns supplied ads that Google could show when they matched the query and the answer. A business could therefore be relevant enough to advertise near a response while its own page failed to appear among the cited sources.

That is why I would not respond to these ads by pulling back on organic work. Paid and cited placements do different jobs, and the answer now has room for both. Clean product data and tight query match matter for the ad. A page with a direct answer, clear subject, and supporting detail matters for the citation. As paid entered more of the answer, leaving the organic source to somebody else became a more obvious gap.

September 2025: Ranking overlap recovers, but not evenly

The history gets less tidy here, which is useful. Overlap between cited pages and ranking pages rose from the early 32.3% to 54.5% by September 2025, a climb of about 22 percentage points across nine industries. Anyone claiming rankings had become irrelevant would have to explain that reversal.

But the industry split matters more than the average. Overlap reached 75.3% in healthcare and 72.6% in education, while ecommerce sat at 22.9%. Ranking remained a useful signal, particularly in some categories. It was not a universal shortcut into the answer, and ecommerce made that hard to wave away.

Other citation patterns show why a homepage and a rank tracker are a thin plan. In one teardown of roughly 8,000 citations, 82.5% pointed to deep nested pages, not homepages. The mix included about 46% blogs, 20% news, and 4% community pages. Reddit was the single most-cited site in that analysis, and LinkedIn articles ranked fourth. Vendor product blogs accounted for around 7%; Wikipedia, under 1%.

The same teardown found a split between markets. B2C answers cited sources such as YouTube, PCMag, CNET, Forbes, and Reddit, while official company sites appeared under 4% of the time. B2B answers drew about 17% from company sites and blogs, alongside sources including Clutch, TechTarget, Gartner, Statista, and LinkedIn.

I would not turn one sample into a rule for every query. I would use it to challenge a lazy assumption: that polishing the homepage and moving up two positions settles citation work. Build the deep page that answers the question, and know which other voices the answer uses in your category.

January 2026: AI Mode turns one answer into several searches

In January 2026, the AI Mode tab opened from mobile Show more. AI Mode holds a conversation rather than answering once. It breaks a longer query into sub-questions, searches across them, and can cite different pages at different steps.

A single excellent page may address the first question but leave the follow-up to someone else. A pricing page can cover cost; a how-it-works guide can cover the process; a discussion can surface the gotchas. The point is not to manufacture pages for every imaginable prompt. It is to notice when your own useful answers live in different places, or nowhere at all.

For a practitioner, this changes the unit of work. One cornerstone page is easier to maintain, but it cannot be assumed to carry an entire conversation. Make the related pages clear enough to stand alone, then connect them so a reader can follow the subject.

February 2026: More Overviews make the citation gap harder to dismiss

SE Ranking’s 100,000-keyword panel found AI Overviews on around 28% of queries in May 2025 and 59.73% by February 2026. Its average links per Overview also rose from 6.82 in November 2024 to 15.22. That is one panel, not a promise about what any individual business will see. It does show why citation coverage deserves its own attention.

More links do not mean your page automatically gets one. They mean an answer may draw on a wider set of sources while the number of queries showing that format grows. A rank-tracking report can still tell you where you appear in conventional results. It cannot show which parts of a generated response cite you, which cite a competitor, or which question your site has not answered at all.

The practical move is to compare the response with the pages you already have. Look for the missing answer before commissioning another general guide. More citation slots reward useful coverage, not a larger pile of interchangeable pages.

October 2026: The present job is to earn the source, not just the position

The line from SGE to today is not that Google stopped caring about ranked pages. The September overlap figures rule that out. The line is that ranking became one input to a different contest: which source helps answer the question in front of the reader. Ads added another claim on that same screen. AI Mode stretched the contest across follow-up questions. Answer-first, entity-clear pages are a better bet than treating a top-three position as the finish line.

Say you spend $20k a month on search and your best service page ranks fifth. I would not assume rewriting it to rank third earns an Overview citation. First, I would read the answer that appears for the query and the pages it cites. Then I would make the relevant page easier to use: one direct answer near the top, under a heading that matches the question, with the price range, timing, inclusions, or limitations the reader needs underneath. The aim is not a block written for a machine at the expense of a person. It is a passage that does not make either one hunt for the point.

A short answer paragraph on a deep guide page beside an AI answer box

Next I would check two less glamorous things:

  • Entity clarity: Use a consistent business name, location, and service across the page, title, and schema. Internal links should name the subject consistently too. A reader should not have to guess which business or offer the page describes.
  • Crawl and page clarity: Check that the answer is accessible, not hidden behind a crawl block or buried in JavaScript. Keep a page focused enough that its answer is identifiable, even when it covers the necessary detail.

Neither fix guarantees a citation. Both remove avoidable reasons for a useful page to be passed over. Then look beyond your own site. The citation mix differs between B2B and B2C, but in both cases outside voices appear in the answer. A deep page of your own and a credible mention elsewhere do different work; a polished homepage cannot substitute for either.

A report can show the miss; it cannot fix the page

Overview-monitoring tools can tell you whether your domain appeared and chart its share of voice. Useful, as far as it goes. The chart does not rewrite a vague opening, clear a crawl block, or make your business easier to identify. I keep a short list of capabilities that matter when getting a business featured in AI Overviews because reporting and execution are different purchases. groas belongs on the execution side of that divide: continuous work on search visibility, with a human strategist responsible for direction, rather than another weekly chart to interpret.

Where does the line point next? I expect the paid result, the cited source, and the follow-up question to keep pressing closer together. That is a reading of the trajectory, not a dated Google announcement. The response now is less speculative. Keep a direct answer on the page that deserves to be cited. Make the entity and offer clear. Cover the related questions on deep pages instead of asking one cornerstone to do every job.

If you do nothing else this quarter, pull five queries that trigger an Overview in your category. Read each answer and its citations line by line. Note the question each cited passage answers, then find the page of yours that should answer it. If that page makes the reader dig, fix the page. A rank tells you where you stood in the results; the citation shows whether you helped build the answer.