AI Mode is not AI Overviews with more text

AI Mode and AI Overviews are two separate retrieval systems inside Google Search, and for the same query they cite mostly different pages, so a single "Google AI visibility" number is averaging two things that do not behave alike.

Most reporting still treats them as one surface. The reasoning is understandable. Both live in Google, both are generative, both draw on the same web index, and Google's own documentation describes them with the same vocabulary. Two large studies measured the actual output, and the pages each surface picks barely overlap.

#The overlap is around a tenth

Ahrefs compared the two surfaces on the same queries using September 2025 US data, running 540,000 query pairs through citation and URL analysis and 730,000 through content similarity. The citation overlap between AI Overviews and AI Mode came out at 13.7%. Restricting the comparison to each system's top three citations barely moved it, to 16.3%.

SE Ranking ran a separate study on 10,000 keywords and landed in the same place. Across the queries where both surfaces appeared, exact URL overlap was 10.7% and domain overlap was 16%. Two different tools, two different samples, two different methodologies, and the answer is the same order of magnitude.

Domain overlap sitting above URL overlap in both studies is worth pausing on. A site that clears the bar for one surface often clears it for the other, just with a different page. That is a retrieval difference, not a trust difference.

#The answers agree; the evidence does not

The strange part is that the two surfaces are not disagreeing about the substance. Ahrefs scored the response pairs for meaning and found 86% average semantic similarity, with 89.7% of pairs scoring above 0.8 on cosine similarity. The same pairs shared only 16% of their unique words, and opened with an identical first sentence just 2.51% of the time.

Two systems reaching the same conclusion from largely disjoint sources tells you something about how much of the web can support a given answer. For most commercial and informational queries, the set of pages that could serve as a citation is much larger than the handful any one system shows. Which pages get picked is close to a draw among qualified candidates, decided by each surface's own retrieval path.

That reframes what a missing citation means. If you are absent from AI Overviews on a query, the useful question is not whether your page is good enough. It is whether the pool of pages that could answer this query is deep, and whether you are in the part of it this particular surface reaches.

#Why the pools differ

Google's documentation says both surfaces may use a query fan-out technique, issuing multiple related searches across subtopics and data sources to build a response, as described in Google's guidance on AI features and your website. Fan-out means the query you typed is not the query that retrieves your page. Each surface generates its own sub-queries, and a longer, more exploratory answer generates more of them.

That shows up in the shape of the output. SE Ranking measured 12.6 source URLs in the average AI Mode response. Ahrefs found AI Mode responses mentioned 3.3 entities on average against 1.3 for AI Overviews, and that 3% of AI Mode responses lacked citations entirely compared with 11% of AI Overviews.

The picture is consistent: AI Mode casts wider and cites more, AI Overviews stays narrow and sometimes cites nothing. Optimizing for one does not automatically place you in the other, because the two are sampling different depths of the same candidate pool.

Neither surface tracks classic rankings closely either. SE Ranking's comparison against the organic top 10 found exact URL overlap of 14%, about 2.5 URLs per query, with domain overlap at 21.9%. A page one position is an input, not a ticket.

#Search Console will not split them for you

This is where measurement gets awkward. Google reports both AI Overviews and AI Mode activity inside the Performance report under the Web search type, per Google's own documentation. There is no filter that isolates either surface. Clicks and impressions from two systems with roughly a tenth of their citations in common arrive in your reports as one undifferentiated line, mixed in with classic blue link traffic.

So the one place with first-party numbers cannot tell you which surface produced them, and the two surfaces behave differently enough that the blend is not informative. A drop in the combined figure could be AI Overviews tightening, AI Mode reshuffling, or ordinary organic movement.

#What to do with this

Stop reporting a single Google AI number. If your tooling collects citations, keep AI Mode and AI Overviews as separate series the way you would keep two different engines apart. Averaging them produces a number that describes neither.

Check them on the same query set. The studies above measured overlap in aggregate; your own overlap on your own thirty or fifty commercial queries is the figure that should drive decisions, and it may sit well away from the published averages.

Read absence as a pool problem first. Given how much the two surfaces disagree while reaching the same conclusions, an uncited page on a competitive query is usually a retrieval outcome rather than a quality verdict. Look at whether the page is reachable, parseable, and specific enough to answer a sub-query, before rewriting it for tone.

Watch domains, not just URLs. Both studies show domain-level overlap running higher than URL-level overlap. If a competitor appears in one surface with page A and the other with page B, a site-wide strength is doing the work, and matching it is a coverage exercise across your section, not a single-page edit.

We keep every engine we track as its own line for exactly this reason. One number for "AI" was always a compromise. Inside a single product, from a single index, with a shared model family, the same compromise now costs you accuracy.