What actually drives AI citations
The short answer
Third-party coverage, overwhelmingly. A University of Toronto study found AI search exhibits a systematic and overwhelming bias toward earned media over brand-owned and social content, with earned media outperforming owned content by roughly 325% for citation rates. After that: entity clarity across trusted sources, an active review profile, and content structured to be quoted. Your own website matters least of the four, which is the opposite of how most AEO budgets are spent.
- Earned media beats brand-owned content by ~325% for citation rates.
- Brands with no review profile appeared in 1% of answers; active ones in 75.3% — a 75× gap.
- Community platforms like Reddit account for roughly 52.5% of citations versus 47.5% for brand domains.
- Only 37.9% of cited URLs now rank in the first ten organic results, down from ~76% in July 2025.
The ranked list
| Lever | Measured effect | Source |
|---|---|---|
| Earned third-party media | ~325% better than owned content | University of Toronto |
Deep entity linking (sameAs) |
+34% overall, +52% Gemini/Bing | GEO Measurement Study |
| Active review profile | 1% → 75.3% of answers | Onely |
| Quotations in content | ~+41% | Princeton GEO |
| Statistics in content | ~+32% | Princeton GEO |
| Inline citations | ~+30% | Princeton GEO |
| Ungating content | 14 → 1,847 citations | GEO Measurement Study |
| Schema markup alone | ~0 | Ahrefs |
| Keyword density | 0 | GEO Measurement Study |
Why earned media dominates
The University of Toronto finding is the one to internalise. AI search shows what the authors describe as a systematic and overwhelming bias toward earned media over brand-owned and social content.
The mechanism is straightforward once stated. A language model answering “who is the best X in Brisbane” has no way to verify a brand’s claims about itself, and every brand claims to be the best. What it can do is observe what independent sources say. So it weights the roundup, the trade publication, the review site and the directory far above your homepage — not because your homepage is bad, but because it is not evidence.
This is why the citation map matters more than the site audit. In most categories, a surprisingly short list of twenty or thirty pages decides whether you exist in AI answers, and almost none of them are yours.
The review profile finding
The gap here is the largest single number in the research:
Brands with no active review profile were cited in only 1% of answers, while brands that actively collected and responded to feedback were cited in 75.3% of answers — a 75× gap.
Two things worth separating. “Active” means collecting reviews and responding to them, not simply having a profile that exists. And this must be done legitimately — soliciting genuine reviews from real customers is fine and effective; generating, incentivising or filtering them is both against every platform’s rules and, in Australia, a potential breach of Australian Consumer Law provisions on misleading conduct.
The Reddit caveat
Community platforms have been widely reported as accounting for roughly 52.5% of citations against 47.5% for brand domains, which produced a wave of Reddit-centric advice through 2025 and early 2026.
That advice aged badly. ChatGPT cut Reddit citations by at least 73% in August 2026. Anyone whose visibility strategy rested on Reddit presence lost most of it in a single unannounced retrieval change.
The lesson is not “ignore communities”. It is that any strategy dependent on a single source is one vendor decision away from zero. Spread across publications, directories, review platforms and communities, and measure which ones your engines actually quote in your category rather than assuming.
What this means for budget
If you are spending on AI visibility and the work is happening entirely on your own website, the evidence says the budget is pointed at the weakest of the four levers. The uncomfortable implication for agencies — including us — is that the effective version of this work looks much more like digital PR than like SEO.
Questions people actually ask
If Reddit drives citations, should I post about my own brand there?
No. Undisclosed self-promotion violates Reddit's rules and most communities will remove it and ban the account, which is worse than doing nothing. The legitimate version is being genuinely useful in communities where your expertise is relevant, under a disclosed identity, and accepting that it works slowly. Also note ChatGPT cut Reddit citations by at least 73% in August 2026, so a Reddit-dependent strategy is fragile.
How long does it take to see movement?
Earned placements typically show up in citation sets within four to twelve weeks of going live, because engines need to recrawl and the retrieval index needs to update. Entity work can move faster. Anyone promising movement in the first month is either lucky or not measuring properly — AI answers are non-deterministic and a single month of variation is noise.
Does my page-one Google ranking carry over to AI answers?
Less than it used to. Ahrefs analysed 863,000 keyword SERPs in March 2026 and found only 37.9% of cited URLs ranked in the first ten organic result blocks, down from roughly 76% in July 2025. Citation and ranking are coming apart, which is why AEO is becoming a separate discipline rather than an SEO add-on.