AEO Ranking Factors: What Gets Brands Cited by AI in 2026

No answer engine publishes its ranking function. Anyone who claims to know the AEO algorithm is selling something. But 2025 and 2026 produced an unusual amount of good evidence: brand-level correlation studies across 75,000 brands, citation analyses across 680 million citations, a prompt-level study of 1.4 million ChatGPT queries, and Google’s own optimization guide published in May 2026.

None of it is a leaked algorithm. All of it points in the same directions. Here are the factors, ranked by the strength of the evidence behind them — and the myths the same evidence buries.

First, how citation actually works

An answer engine does not rank a list of pages the way classic search does. It retrieves candidate material, then synthesises an answer from it. Two consequences follow, and both are measurable.

First, retrieval leans on the search index you already know. Google’s own guide states that AI Overviews and AI Mode are “rooted in our core Search ranking and quality systems.” On the ChatGPT side, Ahrefs’ analysis of 1.4 million prompts found that 88% of cited URLs entered through its search layer, not through special AI files or training data. Your classic SEO foundation is still the substrate.

Second, the model often decides before opening your page. In the same Ahrefs study, cited pages’ titles matched the prompt’s meaning far more closely than retrieved-but-uncited pages (cosine similarity 0.60 vs 0.48), and pages with natural-language URL slugs were cited 90% of the time versus 81% for opaque ones. Your title, URL and snippet are your first pitch — the retrieval layer grades them before the content gets a say.

Factor 1: Being talked about beats being linked to

The strongest correlation in the largest study: brand mentions, not backlinks.

Ahrefs analysed 75,000 brands and found web brand mentions correlated with AI Overview visibility at 0.66 — three times the correlation of backlinks (0.22). Brands in the top quartile of mentions averaged 169 AI Overview appearances; the next quartile down averaged 14. Brands in the bottom half of mention volume were effectively invisible to AI — and 26% of the studied brands appeared in AI Overviews never.

A follow-up Ahrefs analysis found YouTube mentions correlated even higher — 0.74, the strongest single signal they measured. SE Ranking’s study of 129,000 domains points the same way: brands present on multiple review platforms (G2, Capterra, Trustpilot) averaged 4.6–6.3 ChatGPT citations, versus 1.8 for brands absent from them.

The uncomfortable part: this is mostly not work you can do on your own site. Mentions live on other people’s domains — which is why it tops the list and why it’s the hardest.

Factor 2: Reddit and community platforms are the new authority layer

Seer Interactive analysed 8,500 keywords in May 2026 and found Reddit alone took 20.4% of first-citation slots in AI Overviews — 10.5× more than fourteen major publishers combined, including NYT, WSJ, BBC, Forbes and Reuters (0.6% between them).

But there’s a twist that most AEO advice misses. In Ahrefs’ 1.4M-prompt study, Reddit dominated what ChatGPT retrieved (67.8% of retrieved-but-uncited URLs) while its actual citation rate was under 2%. The model learns from the crowd, then cites another institution. Community threads shape what AI believes; review pages and established media are what AI credits.

For a brand, that’s two different jobs: be present and accurate where the crowd talks about you, and be listed where the citations come from.

Factor 3: Google rankings still matter — but they’re decoupling fast

In July 2025, 76% of AI Overview citations went to pages ranking in Google’s top 10. By March 2026, that share had fallen to 38% (Ahrefs, 4M cited URLs). The mechanism is Google’s query fan-out — the model generates its own sub-queries and cites from those result pools, which widens the net well beyond page one.

ChatGPT decoupled even further: Profound’s cross-platform analysis put its overlap with Google results at roughly 8–12%.

Meanwhile, the citation itself has become the traffic event. Seer’s CTR study found that when a page is cited in an AI Overview, organic clicks rise ~35% and paid clicks ~91% versus the same query without the citation — even as overall organic CTR on AI Overview queries fell by two-thirds. With 68% of Google searches now ending without a click (SparkToro, 2026), being the cited source is what’s left of the click.

Factor 4: Quotable beats comprehensive

The only controlled experiment on this list: the Princeton-led GEO paper (KDD 2024) tested content edits on a live generative engine. Adding quotations lifted visibility 28–41%. Adding statistics: +30–40%. Citing your own sources: +30–40%. Keyword stuffing performed worse than doing nothing.

Seer’s 2026 data agrees from the other direction: for definitional queries, winners were bimodal — about a quarter were under 250 words (dictionary-style), another quarter landed in the 1,000–2,000 word range. Five-thousand-word “ultimate guides” captured 4.4% of definitional slots. The playbook is to be quotable, not comprehensive.

Factor 5: Freshness gets you into the pool — not onto the podium

Across 17 million citations on seven platforms, Ahrefs found AI-cited content averaged 1,064 days old versus 1,432 days for organic top-10 results — about 26% fresher. ChatGPT showed the strongest preference (citations ~458 days fresher than its organic baseline); Google’s AI Overviews were the most conservative, behaving like classic search.

The nuance matters: within a single retrieval set, the older, established page tends to win the citation, while fresh pages are often retrieved and discarded. Freshness gets you retrieved. Relevance and entity strength win the cite. Keep content current; don’t expect recency alone to outrank substance.

What the evidence says doesn’t work

llms.txt. Four independent checks, one verdict. Google’s guide says it outright: “Google Search itself doesn’t use them.” SE Ranking modelled 300,000 domains and found no significant citation effect — removing llms.txt from their model improved its accuracy. Ahrefs watched 137,000 domains: 97% of published llms.txt files were never fetched, and 96% of the few fetches came from bots. Limy logged 515 million AI-bot events and counted about 408 llms.txt requests. A file like that can still serve other agent-facing purposes — helping crawlers fetch your content efficiently is a fine goal — but it will not buy you citations.

Schema as a citation hack. The studies conflict: SE Ranking found pages without FAQ schema earned slightly more citations; Seer concluded that beyond Article and Breadcrumb markup, “the rest is wasted markup as far as AIO is concerned”; AirOps claims the opposite. No consensus means no lever. Keep structured data for the rich results it demonstrably powers; don’t expect it to move citations.

E-E-A-T theatre. In Seer’s data, the publisher cohort with the most complete author bios had among the lowest first-citation shares — Reddit, with no author bios at all, took 20% of those slots. Signals written for human quality raters are not what the retrieval layer reads.

Chunking and AI-specific rewriting. Google’s guide addresses this directly: no need to break content into fragments or rewrite for machines — the models handle synonyms and meaning natively.

Each engine has a personality

The factor list above is generic. The citation behaviour is not.

ChatGPT is the brand recommender: BrightEdge found it mentions 3.2× more brands than it cites, and its #1 cited source is Wikipedia (7.8% of all citations, ~48% of top-10 source share — Profound, 680M citations). It also has the strongest freshness bias.

Google AI Overviews cite broadly — 11.4 sources per query on average (Seer) — leaning on Reddit, YouTube and Quora, and fire mostly on long-tail queries, not head terms.

Perplexity is the most citation-dense UI, with Reddit taking nearly half its top-10 source share (46.7%).

And they disagree constantly: only ~17% of queries produce the same brand set across ChatGPT, AI Overviews and AI Mode (BrightEdge). Any AEO strategy — or audit — built on one model is generalising from one witness.

What to actually do

Ranked by evidence strength, not effort:

1. Audit and grow third-party presence. Mentions, review platforms, YouTube, relevant community threads. This is the strongest signal and the slowest to build.
2. Fix titles and URLs to answer real questions. The retrieval layer grades your title, slug and snippet before your content loads.
3. Make key pages quotable. Add real statistics, quotable statements and cited sources. Kill the 5,000-word sprawl that answers nothing in its first screen.
4. Keep doing SEO. The engines are rooted in the classic index — decoupling means page one is no longer sufficient, not that it stopped mattering.
5. Drop the llms.txt project, the schema panic and the chunking experiments. The evidence is in.

The fine print

Three honest caveats. These are correlations, mostly — nobody ran a controlled trial on your market. The studies disagree with each other on details, because Reddit’s measured share depends entirely on how you count (we attributed every number above to its study for that reason). And the landscape moves: the top-10 overlap halved in seven months.

That last caveat is the real lesson of the year. The factors above tell you where to push. Whether the push worked, your buyers’ assistants will tell you — weekly, and differently each week — whether or not anyone is listening.