Where AI citations come from and how to earn them
Assistants do not quote random pages. They lean on a narrow set of sources they treat as trustworthy, and that set is rarely where companies expect it to be.
For a model to recommend your company, it has to run into your company in a source it already trusts. Everything else in generative engine optimisation follows from that one sentence.
In short
- Models frequently quote communities: Reddit, YouTube, LinkedIn, Wikipedia.
- What earns that trust is real opinions and specific data, not marketing copy.
- Your own structured pages do get quoted, but only when they answer the question outright.
- The strategy that works is both at once: presence in the communities, plus answer-first content you control.
Assistants do not quote random pages
When a model assembles an answer, it reaches for sources it treats as reliable. In practice that very often means communities and services carrying a large volume of genuine opinions, rather than the polished pages a brand would nominate for itself.
The logic is not mysterious. A model asked to recommend a supplier is looking for evidence that other people found the supplier acceptable. A page written by the supplier is weak evidence for that specific claim, however well written it is. A thread where practitioners argue about the supplier is strong evidence, even when it is messier.
Where assistants look most often
- Reddit, YouTube and LinkedIn are among the sources large models cite most.
- Wikipedia, plus trade publications and sector sources that carry concrete numbers.
- Company pages that are well structured, provided they answer the question directly rather than circling it.
The pattern across all three groups is the same: they contain something a model can attribute. A number, a comparison, a stated experience, a definition. Prose that could describe any company in the category gives a model nothing to hold onto, so it moves on to a page that does.
A model cannot quote an adjective. It quotes a fact with an owner.
What this means for your company
Your website alone is not enough. You need to be present in the conversation where your industry is actually discussed, and you need your own content to be quotable when the model does land on it. Neither half works on its own. Community presence without a solid site sends the model to a page it cannot use. A perfect site with no external footprint leaves the model with one source, which is exactly the situation where it prefers somebody else.
This also explains why brands that spent years on ordinary search visibility tend to appear in AI answers sooner. Retrieval runs over live search results, so indexation and technical health are still the entry condition, not an optional extra.
An action plan that holds up
- Build a presence in the communities that matter to your sector. Answer questions as a named person from a named company, not as an anonymous account dropping links. The value comes from being useful in public, repeatedly.
- Keep the reviews and the brand description consistent. Same company name, same address, same description of what you sell, everywhere it appears. Contradictions across sources make a model hedge, and a hedging model names somebody else.
- Publish answer-first content with data and an FAQ. Question-shaped headings, the answer in the first two sentences, then the evidence. Back it with structured data that matches what is visible on the page.
- Measure it. Keep a fixed list of prompts your customers would realistically type, run them regularly, and record whether you are named and which source the model credits. That log is the only feedback loop available in GEO right now.
What to expect from the timeline
None of this pays off in a fortnight. Community presence compounds slowly, and models refresh what they know about a brand on their own schedule. Treat it as the same category of investment as ordinary search visibility: a few months before the pattern shifts, then a position that is hard for a competitor to buy their way past. The mechanics of how those fragments are retrieved in the first place are covered in query fan-out, and the wider picture in SEO vs AEO vs GEO.
Common questions
Which sites do AI assistants quote most often?
Very often communities and services full of real opinions, among them Reddit, YouTube, LinkedIn and Wikipedia, alongside trade sources that carry specific data.
Is a good website enough on its own?
Not always. Presence where your industry is actually discussed counts as well. The workable approach combines your own content with a real presence in those communities.
How do we raise the odds of our own page being quoted?
Write answer-first content, add data and an FAQ, keep the company details consistent everywhere and back the page with structured data. All of it makes the page easier for a model to understand and attribute.
Do customer reviews matter to AI?
Yes. Genuine reviews and mentions on trusted services build the credibility a model weighs before it recommends a brand.
Read next
Find out whether AI recommends your company.
Start with the free SEO and GEO audit, delivered in 5 working days. We check how the models describe your brand and hand back a prioritised list of changes.