What gets a brand cited by AI search
Two things decide whether an AI answer cites you: where you rank, and how often other people mention you. Nearly everything else being sold as AI optimization has weak evidence behind it, or evidence against it.
This is what the research shows, with the numbers and where they came from.
Does ranking still matter?
A lot, although the link has weakened. In an analysis of 16,851 queries and 353,799 pages, a page retrieved in position one was cited 58.4% of the time, compared with 14.2% at position ten. When the data was filtered to pages that matched the query well, the top spot rose to 79.6%, against 21.5% for anything beyond position eleven.
That weakening is real, though. Ahrefs found that in July 2025, 76% of AI Overview citations came from pages ranking in the top ten. By March 2026, that was down to 38%, with 31% of citations coming from pages beyond position one hundred. If someone's still quoting the 76% figure, they're a year behind.
If not the top ten, then where from?
Sub-questions. These systems take a query and break it into lots of smaller ones behind the scenes, then find a source for each. You end up getting cited for questions you never targeted.
The correlation between ranking for those fan-out queries and being cited was 0.77. Ranking for them made a page 161% more likely to be cited than ranking for the headline term alone. And around 95% of those fan-out queries have no measurable search volume, which means keyword tools will never show them to you.
In practice, that changes the shape of your content. Covering lots of questions beats going deep on one term. One real question per page, answered properly, will beat a 4,000-word guide that tries to cover everything.
How long should a page be?
Shorter than most SEO advice would tell you. The sweet spot in the data is 500 to 2,000 words. Pages under 500 words underperform, and pages over 5,000 do even worse.
Separate research into how Google pulls together its grounding data found a limit of roughly 540 words per source. Pages under 1,000 words had 61% of their content available for citation, while pages over 3,000 words had just 13%.
Long-form content doesn't struggle because length gets penalized. It struggles because only a small part of it ever gets looked at.
What is the strongest on-page factor?
Headings that match the question. As a heading got closer to the wording of the query, the citation rate climbed from 30.2% to 41.0%, and the increase held steady across every band. For well-ranked pages the gap was even wider, going from 55.9% to 75.3%.
It's the cheapest fix most sites have, and the one most brands get wrong, because a good brand heading and a good retrieval heading aren't the same thing. "Strategy before screens" is memorable, but it doesn't match anything anyone types.
What gets lifted off the page?
Something smaller than a page. An analysis of 15.7 million Google AI Mode citations, which show the exact passage quoted, found the median cited passage was 117 words. About 80% put the answer in the first few sentences, and roughly 85% made sense on their own without the surrounding context.
Cited text was also about twice as likely to use definitional phrasing (the "X is…" construction), and it had far more proper nouns than typical writing.
So the thing to write isn't really the article. It's a short, self-contained block near the top that defines the answer, sitting under a heading that matches the question.
Do backlinks help?
Apparently not, and that's the most surprising finding here. Once relevance was accounted for, domain authority and backlinks showed no positive link with citation, and even a slightly negative one. Pages that were always cited averaged 1.1 million backlinks. Pages that were never cited averaged 3.2 million.
Unlinked brand mentions are a different story. Across 75,000 brands, YouTube mentions had a correlation of 0.737 and branded web mentions 0.664, compared with 0.266 for domain rating. Those are correlations rather than proof, and brand size muddies them. But they all point the same way: being talked about matters more than being linked to.
If you've got a new domain, that's good news. Starting from scratch is much less of a handicap here than it is in classic search.
Where do citations come from?
Mostly not from your website. For commercial business queries, a brand's own site makes up roughly 8% to 17% of what gets cited. The rest is third-party.
Roundups dominate. Across roughly 400 million citations, 63% pointed to listicles, rising to around 71% for business questions, and about half of the cited titles included the year.
Each engine behaves very differently, though. One study of 379,000 citations on technology questions found Claude cited company websites 64% of the time and Reddit not once, which is close to the opposite of Google's pattern. The engines only shared 13% of their sources. No single strategy works for all of them.
What about freshness?
It's overrated. The median cited result in ChatGPT search was around 500 days old. Content less than 30 days old was the worst-performing age group, and citations peaked somewhere between one and three months.
Updating beats publishing. In one analysis, 27% of cited pages were older URLs that had been refreshed, not newly written.
What is worth doing?
- Get listed on the review platforms and roundups your buyers read. It has the most impact, and it happens off your site.
- Write one real question per page, 800 to 1,200 words, with the answer in the first line.
- Make your headings match questions instead of trying to sound clever.
- Keep everything server-rendered. No major AI crawler except Google's runs JavaScript. Here's how to check what your Shopify store shows them.
- Add Organization markup once, then leave it. Our post on schema and AI citations explains why that's enough.
One word of caution about all of this. Running the same prompt twice on the same day only gives you 32% to 43% overlap in which sources get cited. Any vendor showing you a single before-and-after run is really showing you noise.
We follow the same rules in the content we write for brands, including this post. If you'd like help working out which of this applies to your store, start with a free CRO audit.