AI Overviews is a summary at the top of a regular results page, with the classic links still below it. AI Mode is a standalone conversational mode where the answer stands on its own and you can keep asking follow-ups. Both build an answer from multiple sources, but they often cite different URLs — so you can’t treat them as one surface.
From the outside it looks like one thing: “Google answers questions itself now.” In practice these are two different places with different behavior — and that has a fairly concrete consequence for how you measure and what you can promise.
Where you’ll run into each one
| Aspekt | AI Overviews | AI Mode |
|---|---|---|
| Where it lives | A summary at the top of a regular results page | A standalone conversational mode |
| Links alongside | Yes — the results stay below the summary | No — the answer stands alone |
| Interaction | A one-shot summary | A conversation, with follow-ups |
| Length | Shorter | Considerably longer |
| Query decomposition | Uses query fan-out | Uses query fan-out |
The common denominator matters more than it looks: based on what we’ve seen so far, both break the query into sub-questions and assemble the answer from several sources. How that works is covered in the piece on query fan-out.
Same answer, different sources
This is the most interesting finding — and also where you need to be careful with the numbers.
Working from US data collected in September 2025 that captures individual generations, Ahrefs compared answers from both surfaces across a large sample: roughly 730,000 answer pairs for content similarity and 540,000 query pairs for citations. The answers were close in meaning — average semantic similarity around 86 percent, while word-level overlap came in around 16 percent. The cited URLs overlapped by 13.7 percent, and by 16.3 percent for the top three citations. In other words: both said much the same thing, but they backed it with different pages.
Before you file that number away as fact, it comes with three caveats — either flagged by the study itself or baked into how it was set up.
The conclusion you can safely draw from the study is more sober than the headline number: even where more than one source could support an answer, the two surfaces often reach for different ones.
What this changes for your site
Measure them separately. Finding yourself in one tells you nothing about the other. How to check your visibility is covered in the AI visibility test; for Search Console data, see the AI report in GSC.
One run isn’t enough. When source selection shifts between runs, a single test is an anecdote. Repeat it over time and across more queries.
But the optimization doesn’t split in two. The foundation is the same for both: answer a specific sub-question clearly, keep your sources verifiable, and keep your data consistent. The practical content prep is covered in Google AI Mode.
Common mistakes
Treating them as one surface
“We’re visible in AI” says almost nothing. These are two different places that behave — and cite — differently.
Porting US numbers to your market
Different market, different rollout stage, different volume of source material in the language. Treat numbers from a study run elsewhere as a guide, not a forecast.
Drawing a conclusion from one run
Citations fluctuate between runs. Anyone who infers a trend from a single test is measuring noise, not reality.
Special tricks for one surface
Techniques promised exclusively for AI Mode or exclusively for AI Overviews usually can’t be backed up. The shared foundation is still the same.
Where this article stops
It doesn’t tell you how to get into them. That’s a content-prep topic, not a comparison; for that, see Google AI Mode and the pieces that follow.
It doesn’t give numbers for your market. There’s no publicly documented data beyond the US analysis, so we’d rather leave it out than estimate.
It doesn’t promise that one surface carries over to the other. If anything, the opposite — the available analysis shows the citations of the two places often diverge.
What to take away
AI Overviews and AI Mode blur easily into a single “Google answers questions itself now.” But the difference is practical: the first is a summary above the results that leaves the links in place below it, the second is a standalone conversation without them.
And above all — the available analysis suggests that even when they say similar things, they often back them with different sources. So track them separately, don’t take one measurement as proof, and with numbers from elsewhere, always ask which market they came from and when.
Want to know where you stand across the individual AI surfaces and what you can do about it? Sniper Design runs AI SEO audits that go through both visibility and content and help name the next steps.
Source for the numbers: an Ahrefs analysis comparing AI Overviews and AI Mode answers over Brand Radar data (US, September 2025). We cite it along with the limits the study itself flags.
