AI Mode and AI Overviews Agree 13.7% of the Time

01They agree on the answer and disagree on who gets the credit

Google runs two AI surfaces on top of search. AI Overviews is the short summary that appears above ordinary results. AI Mode is the longer conversation you open on purpose. A study of 540,000 query pairs put the same question to both and compared what came back, and the two surfaces cited the same pages only 13.7% of the time. Looking only at the top three sources on each side, the overlap rose to 16.3%.

The headline number is about credit, not content

The title of this entry is only half the finding, on purpose. On substance the two surfaces almost always agree: their answers were 86% similar in meaning, and nine times in ten they reached the same conclusion. It is the sources underneath that differ. A reader asking once sees one confident answer with a few links and never learns that the other surface would have credited somebody else. Two answers from the same company, built from different sources, and only you notice.

It also means a single screenshot proves very little. Finding yourself in the Overview for your main query tells you almost nothing about the longer conversation a buyer opens next, and being named in AI Mode says just as little about the summary most people see first.

02One question becomes many searches

The fan behind the answer

Google says both surfaces may use a technique it calls query fan out: issuing several related searches across subtopics and data sources, then building one response from what comes back. How wide that fan opens depends on how much answer there is to write. AI Mode responses run about four times longer than Overviews and name 3.3 entities on average against 1.3, so the fan opens wider and reaches pages the short summary never needed. A longer answer needs more sources, and more sources means different ones.

Different appetites for the same question

The two surfaces also reach for different kinds of page. In the same research YouTube held the top spot among AI Overview citations, with community platforms like Reddit prominent, while Wikipedia appeared in 10% more AI Mode citations and reference style sources did better there generally. That is the argument from a few days ago about who holds the pen, played out inside a single company. The same company can quote a video on one surface and an encyclopaedia on the other, for the same question.

One detail makes AI Mode the more honest of the two to measure. Only 3% of its responses carried no citation at all, against 11% for Overviews, so when AI Mode leaves you out, it has almost always credited somebody instead.

03Write for the sub questions, not the head term

No special markup, and the same eligibility

Google is unusually direct here. There are no additional requirements to appear in either surface, no new machine readable file and no special markup. A page needs to be indexed and eligible to show with a snippet, which is ordinary search hygiene and the reason search is still the supply chain for everything built on top of it. That removes a comfortable excuse. If you appear on one surface and not the other, the gap is the question your page answers, not a missing tag.

Cover the fan

If one question becomes several searches, the page that gets pulled is the one that answers one of those smaller questions cleanly. A guide to moving to a city is really a set of answers about neighbourhoods, costs and schools, and each deserves its own heading with the answer in the sentence directly under it, so a single part can be lifted without the rest. Write for the questions a question turns into.

Then stop measuring the two surfaces as one. Run your question set through each separately, the way the answer audit runs questions through different engines, record which of your pages each one credits, and treat a gap on one side as a missing sub question rather than a penalty. The work that closes it is almost always writing, not code.

One caution about the number itself. The research used US queries from September 2025, both surfaces have changed since, and nothing guarantees the overlap is still exactly 13.7% today. What is unlikely to have changed is the mechanism underneath it: a longer answer assembled from a wider fan of searches will keep reaching pages that a short summary never needed.

Tomorrow: the engine almost nobody optimises for, and a site that went from two citations a day to thirty.