Every SEO team has been asked the same question this year, and most can't answer it: when Google shows an AI Overview, does it mention us?
Rank trackers don't answer that. They tell you where you sit in a list that fewer people scroll to. So we ran a study to find out what actually drives citation in Google's AI answers — and whether the classic ranking playbook still gets you in.
Method
120 commercial-intent keywords across 9 niches (SaaS, finance, health, e-commerce, travel, marketing, B2B services, education, home & DIY). US, English, desktop, single snapshot on 12 July 2026, collected through residential IPs. For every keyword we captured:
- whether an AI Overview appeared,
- the full answer text,
- every visible citation (URL, domain, position),
- the top 10 classic organic results on the same page.
Then we compared the AI's sources against the organic results below them. 120/120 queries completed.
Result 1: 77% of commercial keywords now trigger an AI Overview — but coverage is wildly uneven
| Niche | AI Overview rate |
|---|---|
| Health | 93% |
| Finance / Travel / Home / E-commerce / Education | ~80% |
| SaaS | 75% |
| Marketing | 73% |
| B2B services | 40% |
If you sell B2B services, AI Overviews are still a minority of your SERPs. If you're in health or consumer finance, the AI answer is the SERP.
Result 2: ranking #1 is a coin flip
The page ranking #1 organically was cited in the AI Overview only 49% of the time.
Half of the time, the page Google's own ranking system considers the best result for the query does not appear among the sources its AI chose to cite.
The nuance that saves classic SEO: 87% of AI Overviews cited at least one top-3 organic result. Top-3 presence is nearly necessary — it just isn't sufficient. Something else decides which of those candidates gets pulled into the answer.
Result 3: a third of the AI's sources are invisible to your rank tracker
Of 319 resolvable citations, 33% pointed to pages nowhere in the organic top 10 for that query.
Think about what that means operationally: a third of the sources shaping the answer your buyers read are not in the dataset your reporting is built on. You cannot see them, so you cannot compete with them.
Result 4: platforms beat publishers
Most-cited domains across all 92 AI Overviews:
- YouTube — 38 citations (11% of all citations)
- Reddit — 31 (9%)
- NerdWallet — 10
- Forbes — 9
- Semrush — 6
YouTube and Reddit together took 20% of every citation we saw — more than any publisher, and far more than any vendor's own product page. Google's AI reaches for demonstration (video) and consensus (community threads) over marketing copy.
Result 5 (the follow-up that surprised us most): Google's two AIs disagree with each other
We re-ran 25 of the keywords against Google AI Mode — the conversational search tab — the same day, and compared its citations with the AI Overview's citations for the identical query.
Mean overlap of the two citation sets: 16%. Every single comparable query had under 25% overlap. One pair shared zero sources.
The overview box and the conversational tab are, for practical purposes, two different search engines with two different opinions about who to trust. Tracking one tells you little about the other.
What we'd actually do with this
- Stop reporting rank alone. Add one line to every report: cited in AI answer — yes/no, per keyword, per surface.
- Build the gap list. Keywords where a competitor is cited and you aren't — especially those where you outrank them. That's not an authority problem, it's a format problem, and format is fixable.
- Format for citability. The pages that get cited look like comparisons, structured lists, and specific claims with numbers — not homepage copy. Get the answer to the question above the fold, in plain language, in a structure a model can lift.
- Get into the platforms. If YouTube and Reddit are a fifth of all citations in your category, a video and an honest community presence are SEO deliverables now, not "brand".
- Track more than one surface. AI Overviews, AI Mode, and other engines (we've measured Brave's answers too) cite different sources. One number isn't the truth.
Reproduce it yourself
The study ran on a tool we built and published on the Apify marketplace: Google AI Overview Brand Monitor — feed it keywords + your domain + competitor domains and it returns, per keyword, whether an AI Overview exists, its full text, every citation, your citation status vs. competitors, share of voice, and the gap/opportunity keyword lists. It covers Google AI Mode as well, so you can see the two surfaces disagree in your own category. There's a sibling actor for Brave's AI answers if you want a second engine.
Top comments (0)