How to be found in AI search

What is evidenced

Three findings survive contact with their own primary sources, each from a large, disclosed sample. None of them is a switch you flip. They describe a position, built over time, that either exists or does not.

Read together, these three studies point at the same underlying position: being mentioned by third parties, independent of anything you publish yourself, correlates with AI visibility more than classic ranking signals like backlinks do; being cited as a source does not guarantee your brand name appears in the answer text, so citation and mention are not the same outcome and both matter; and the retrieval surface AI systems draw from has moved measurably away from Google's organic top 10, which means ranking for your target query addresses a shrinking share of where citations now originate.

What is plausible but unproven

These ideas are repeated constantly, sound reasonable, and have no traceable study behind the specific claim. That does not make them false. It means nobody, including this site, has evidence that they are true, and anyone stating them as settled fact is going beyond what is known.

• Adding an llms.txt file improves citation odds. No study we can find demonstrates this. Google's John Mueller has said no AI service has confirmed using the file at all. Graded UNTRACEABLE.
• Entity presence in a knowledge base helps beyond what mentions already measure. Plausible given how these systems retrieve information, but we found no controlled study isolating a knowledge-graph or entity-database effect separate from the general web-mentions correlation Ahrefs measured. Treat this as a reasonable hypothesis, not a finding.
• A 40% visibility boost from generative engine optimisation. This is the field's most quoted statistic, from a real, peer-reviewed paper. It measured a research benchmark the same authors built, not a commercial AI assistant. Read the full anatomy of the claim. Graded MISLEADING.

What is repeated but refuted

These are the tactics where someone actually ran the experiment, and the result contradicts the popular advice. This is the most useful pile on this page, because it is where following common guidance costs you effort for no measured benefit.

Neither study was published by a party with an incentive to find a null result. Ahrefs sells rank tracking, not schema tooling or a word-count optimiser, and if anything stood to gain from a tidy positive finding on either question, it was Ahrefs. It did not find one, on 1,885 treated pages for schema and 1.68 million cited URLs for word count.

The honest summary

The evidenced factors are about being independently corroborated across the web and sitting inside a retrieval surface that has moved beyond top-10 organic rank. The unproven factors are mostly technical tactics (a file, a schema tag, a markup change) that are easy to sell as a discrete deliverable and easy to implement without evidence they matter. The refuted factors are exactly the tactics that look like classic SEO homework carried over unchanged. If a recommendation in this field is easy to bill for and hard to measure, that correlation is itself worth noticing.

If your brand is not being mentioned at all, start with the diagnostic walkthrough, which grades each possible cause by evidence strength rather than guessing. For the vocabulary this page uses, see AI visibility, GEO, AEO and SEO, explained. For why a single visibility score cannot currently be trusted, see can you measure AI visibility?.

Common questions

What actually determines whether AI assistants mention a brand?

Three things with real, sourced evidence behind them: whether a source is already mentioned across the web independent of your own site (mentions correlate with AI Overview visibility more strongly than backlinks, Ahrefs found, at Spearman 0.664 versus 0.218), whether it is cited at all regardless of whether the brand is named in the answer text (Semrush found 61.7% of citations name no brand), and whether it sits in the retrieval surface AI systems now draw from, which Ahrefs found has moved away from Google's organic top 10. None of these are levers you pull once. They describe a position you are already in or are not.

Does adding schema markup help AI assistants cite a page?

Ahrefs tested this directly on 1,885 pages against 4,000 matched controls and found no meaningful uplift on any platform measured, and a small, statistically significant decline on Google AI Overviews. Schema is not evidenced to increase citation. It remains useful for other reasons, principally that it helps machines parse a page correctly, but "add schema to get cited more" does not survive the study that tested it.

Does long-form content get cited more often?

No. Ahrefs analysed 560,346 AI Overviews and 1,677,876 cited URLs and found a Spearman correlation of 0.04 between word count and citation, effectively zero. Over half of cited pages were under 1,000 words. The belief that AI systems reward exhaustive long-form guides is not supported by the largest dataset we could find that tested it.

Is there a reliable way to measure whether it is working?

Not yet, and this is the honest caveat that belongs on every page in this cluster. The same prompt returns the same list of brands under 1 time in 100 across repeated runs, and only 2.37% of cited URLs appear across all three major AI engines. A single "AI visibility" score is measuring something close to noise. See the full treatment on the measurement page linked below.