Search Google for "ai overview optimization" and Google answers with an AI Overview. It cites thirteen sources. Not one of them is the page you are about to read, and not one of them belongs to a company selling AI Overview optimization.
That is the whole problem in one SERP. Almost every guide on how to rank in AI Overviews tells you the same things: add schema, write clear answers, structure your headings, chunk your content. All of that is real. None of it explains why the citations in front of you point at Reddit, Search Engine Land, Semrush, and a handful of agency blogs instead of the brands being discussed.
The answer sits off your site. Below is what three large-scale studies and our own SERP analysis actually show about where AI Overview citations come from, what correlates with getting cited, and what that changes about how you build links in 2026.
Most AI Overview Citations Do Not Come From the Top 10
The first assumption worth killing is that ranking first gets you cited. It helps. It is nowhere near sufficient.
Conductor analysed 167,867,680 AI Overview citations from US Google searches across every industry between 15 June and 15 August 2026. Their finding:
- 57% of AI Overview citations go to pages outside the organic top 10
- 43% come from top-10 pages
- Position 1 is cited 24.9% of the time, falling to 10.2% at position 10
- Position 1 alone accounts for just 8.5% of all citations
Pages that never crack the top 10 are, in Conductor's words, the largest single source of citations — bigger than every ranked position combined.
Other studies land in the same territory with different numbers. Coverage of Ahrefs' data puts top-10 pages at 38% of citations. Originality.ai's study found 52% of citations appear in the top 10. The spread is wide, because sample, date, and industry all move the figure.
So we ran our own.
Our Analysis: 48 Citations Across 8 US SERPs
In October 2026 we pulled the AI Overview for eight US queries spanning commercial, comparison, and informational intent, then checked every cited domain against the organic top 10 for that same query.
All eight queries returned an AI Overview. They produced 48 citations in total.
- 46% of citations came from domains outside the organic top 10
- Excluding YouTube, that figure held at 44%
- Per-query overlap with the top 10 ranged from 20% to 75%
- 48 citations came from 39 distinct domains. Only 5 domains were cited on more than one SERP
Our sample is small and we are not presenting it as a replacement for a 168-million-citation dataset. It is a check, and it agrees: somewhere between four and six citations in ten are handed to pages that do not rank on page one.
The more useful finding came from cutting the same data a different way.
71% of Citations Pointed Somewhere Other Than the Vendor
For each query we separated citations that pointed at a vendor — a company that could itself be the answer — from citations that pointed at a third party writing about those vendors.
- 71% of the 48 citations pointed at a domain other than the vendor being discussed
- 19% were UGC or reference sites: Reddit, Quora, YouTube, Wikipedia
And the two agency-service queries were absolute:
- "link building services" — 7 citations, 0 agency websites
- "digital pr agency" — 7 citations, 0 agency websites
Fourteen citations across two high-intent commercial queries, and not a single one went to a company that sells the service. Google cited comparison posts, listicles, and editorial roundups that name those companies.
Read that back as a strategy and it stops being depressing. If AI will not cite your site for your own money query, then the path to being the answer runs through the pages it will cite. Getting your brand onto those pages is not content marketing. It is link building.
What Actually Correlates With AI Overview Visibility
Ahrefs studied 75,000 brands, measuring each one's AI Overview presence against its off-site and on-site signals. The Spearman correlations:
- Branded web mentions — 0.664
- Branded anchors — 0.527
- Branded search volume — 0.392
- Domain Rating — 0.326
- Referring domains — 0.295
- Number of backlinks — 0.218
- URL Rating — 0.18
Brands in the top quartile for web mentions averaged 169 AI Overview mentions. The next quartile down averaged 14. A twelve-fold gap.
Seer Interactive found a similar shape testing ChatGPT brand mentions: Domain Rating at 0.25, raw backlink count at 0.10. Kevin Indig measured a 0.334 correlation between chatbot mentions and branded search volume.
Everyone reporting this data is careful to say correlation is not causation, and so are we. Ahrefs notes that every factor it measured was moderate to very weak, and that AI Overview visibility has been demonstrably manipulable.
But the ordering is consistent across three independent datasets, and it is the opposite of how most link building is sold:
- How often your brand is named on other sites correlates roughly three times more strongly than how many backlinks you have
- Branded anchors — the text people use when they link to you — sit second, above Domain Rating and well above link volume
- Raw backlink count is the weakest link signal in the set
Link volume is not the lever. Being named, in the right context, on pages that cover your category, is. That distinction is the difference between a campaign that moves AI visibility and one that just moves a dashboard — which is also why we argue for measuring real SEO impact instead of DR.
Why Standard AI Overview Optimization Advice Is Only Half the Job
The on-page half of AI Overview optimization is genuinely necessary. If a model cannot parse what your page claims, it will not quote you. Clear question-and-answer structure, specific and extractable sentences, consistent entity naming, and clean markup all raise your odds. We have written the full on-page version of this separately: how to design content AI can actually use.
But look at what the on-page advice cannot do:
- It cannot put you in a comparison post you do not control
- It cannot get you named in a Reddit thread
- It cannot create the branded mentions that correlate at 0.664
- It cannot get you into the cited set on a query where zero vendor sites are cited
Necessary, not sufficient. Schema is table stakes. If you only optimize for AI Overviews on your own pages, you are competing for the minority of citations that vendor sites receive at all. Off-site presence is the part that decides who gets picked.
This also holds beyond Google. ChatGPT draws almost half its top-10 source share from Wikipedia and Perplexity roughly the same share from Reddit — neither of which you own. We break the cross-engine citation data down in full in our guide to answer engine optimization.
The Off-Page Playbook We Run for AI Visibility
This is what LinkyJuice actually does when a client's goal is AI citation rather than a ranking position.
1. Citation-source mapping
Before building anything, we pull the AI Overview for every query that matters commercially and record which domains get cited. That list — usually 5 to 15 domains per query — is the target list. Not a DR filter. Not a generic prospect scrape. The specific pages the model is already quoting.
Our data showed only 5 of 39 domains repeating across SERPs, which means these source sets are query-specific and shallow. That is good news: they are small enough to win.
2. Placement on the pages that get cited
Once we know a comparison post or category roundup is in the cited set, the job is to get the client accurately represented on that exact page. That is outreach, relationship work, and in some cases a better pitch than the brand currently sitting in the slot. It is the same muscle as competitor backlink analysis, pointed at a different target list.
3. Branded anchor discipline
Branded anchors correlate at 0.527. Most agencies spend that budget on exact-match commercial anchors instead. We weight anchor profiles toward brand and natural phrasing, which is better for AI visibility and lower risk at the same time.
4. Unlinked mention conversion
Mentions correlate more strongly than links, and most brands already have mentions they never claimed. We find them, confirm the context is accurate, and where appropriate turn those mentions into links. The mention already counts. The link adds the second signal.
5. Digital PR for mention volume
If the metric that matters most is how often your brand is named across the web, the discipline that moves it fastest is digital PR — data stories, expert commentary, and original research that journalists and bloggers cite by name.
6. Relevance over authority
A DR 80 placement in an unrelated niche does nothing for whether a model associates your brand with your category. Topical relevance is what builds the association, and the correlation data backs that: context-bearing signals beat raw authority metrics.
7. Measurement against citations, not rankings
We track citation share for target queries — how often the client appears in the AI Overview, and via which source — alongside traditional rankings. A client can hold position 4 and be invisible in the answer, or sit at position 14 and be quoted. Those are different problems with different fixes.
Where This Goes in 2027
Three predictions we are willing to be judged on.
Citation share becomes the primary KPI, and rankings become a diagnostic. If 57% of citations already come from outside the top 10, position tracking is measuring a weaker and weaker proxy. The teams that move first will be the ones reporting on presence in the answer, with rankings as supporting detail rather than the headline.
The cited set per query gets contested and then calcifies. Right now those sets are small — 39 domains produced 48 citations in our sample, with almost no repetition. Small sets are cheap to enter. As soon as more brands realise which specific pages are being quoted, the price of a placement on those pages goes up and the incumbents get harder to dislodge. The window where this is inexpensive is now, not in eighteen months.
Unlinked mentions get priced like links. Mentions already out-correlate links by roughly three to one. Once that is widely understood, the market for a brand being named in a credible roundup will look a lot like the market for a backlink today — and the agencies still selling DR-filtered link packages will be selling the weakest signal in the set.
None of this means links stopped working. Referring domains still correlate at 0.295 and Domain Rating at 0.326 — weaker than mentions, but not zero, and both are upstream of ranking in classic search, which still feeds 43% of citations. The honest read is that link building did not become obsolete. It became a means to a different end, and the end moved from position to presence. We laid out the longer version of that argument in our 2026 link building statistics breakdown.
What To Do This Quarter
- Pull the AI Overview for your ten highest-intent queries and write down every cited domain
- Mark which of those domains already mention you, accurately, and which do not
- Count how many cited sources are vendor sites. If the answer is near zero, stop trying to rank your own page into the answer and start working the sources
- Audit your anchor profile for branded share
- Find your unlinked mentions and claim the ones worth claiming
- Start reporting citation presence separately from rankings, so you can see the two diverge
The pattern underneath all of it is simple. AI answers are assembled from what other people say about you, on pages you do not own. Every credible dataset points the same way, and the strongest signal in every one of them is how often and how well your brand gets named elsewhere.
That has always been what good link building produced. The difference now is that it is the point rather than the side effect.
If you want the source list for your own queries and a plan for getting into it, book a call and we will map it with you.
