How do AI assistants choose who to cite?

Four signals dominate. Extraction quality — can the model pull a clean, self-contained answer from your page? Entity confidence — does the model know who you are and what you do, consistently, from multiple sources? Topical coverage — do you answer the whole question cluster, or one query? And corroboration — do other credible sites describe you the same way?

Notice that only one of those is about your website. The rest is about how the web describes you.

The uncomfortable implication is that most of this work happens off your own site. You can write the best page in your category and still go uncited if nothing else on the web confirms you are who you say you are. Optimisation here is as much a public relations exercise as a technical one.

What content structure gets extracted?

Answer engines reward structure that survives being quoted out of context.

The test is whether a paragraph can be lifted without its neighbours. Read each section of your highest-value page and ask whether it makes sense alone. Where it does not, add the missing context explicitly — name the product, the place and the outcome rather than relying on 'it' and 'this'.

  • Lead with a 40–60 word direct answer under a question-shaped heading
  • Keep paragraphs to 2–3 sentences so any one can stand alone
  • Use tables for comparisons and numbered lists for processes
  • Add a labelled FAQ block for every question your customers actually ask
  • Name entities explicitly — the business, the place, the product — rather than 'we' and 'it'

How do you make your business machine-verifiable?

Verifiability comes from consistency. Use one canonical description of your business — what you do, who you serve, where you operate — and repeat it across your website, Google Business Profile, LinkedIn, industry directories and social profiles.

Then make it machine-readable: Organization and LocalBusiness schema sitewide with address, geo coordinates, hours and sameAs links; Article schema on posts; FAQPage on question content; BreadcrumbList for structure. Validate everything, and make sure the markup agrees with what a human sees.

Keep a single source of truth document for your business facts — legal name, trading name, address, phone, services, service areas, founding date — and check every profile against it once a quarter. Most entity confusion is caused by one stale listing nobody has looked at since it was created.

What is llms.txt and does it help?

llms.txt is a plain-text file at your domain root that gives AI systems a curated map of your most useful pages, with short descriptions. It is not an official standard and no major provider guarantees using it, but it costs almost nothing and gives crawlers an unambiguous summary of your site — including what you want understood and what you would rather they ignore.

Treat it as a well-organised index for machines, and keep it current alongside your sitemap.

Two caveats. It does not influence search rankings, and letting it drift out of date is worse than not having one, because it points crawlers at content that no longer exists. Add it to the same maintenance routine as your sitemap.

How do you earn corroboration?

Models weight consensus. If your website says you are the leading paediatric clinic in a city and no other site agrees, the claim is weak. If reviews, local press, professional listings and partner sites all describe you consistently, the claim becomes a fact the model can rely on.

Practical moves: a systematic review request routine, an accurate and complete presence on the three to five directories that matter in your industry, and outreach that earns genuine mentions rather than paid links.

Earned mentions beat paid placements here for a specific reason: a model trying to verify a claim weights independent agreement. A sponsored placement describing you in your own words adds little. A journalist, a customer or a partner describing you in theirs adds a great deal.

Does anyone actually control what AI assistants cite?

No. There is no submission form, no ranking dashboard and no guarantee. Each assistant uses its own retrieval and ranking logic, and providers change it without notice. Anyone selling a guaranteed citation is selling something they cannot deliver.

What you control is the probability. Clean structure, consistent entity data and independent corroboration all raise the likelihood that a retrieval system finds you, understands you and trusts you enough to quote you.

What does a citation audit look like?

Pick ten buying questions your customers actually ask — not the ones you wish they asked. Run each through ChatGPT, Perplexity and Google's AI features, and record four things: whether your brand appears, whether a competitor does, which sources are cited, and how your business is described.

Do this monthly and keep the results in a simple spreadsheet. The value is in the trend rather than any single month, and the citation lists tell you which third-party sites the assistants trust in your category — which is a list worth being on.

  • Ten buying questions, asked the same way each month
  • Whether you appear, and how prominently
  • Which sources are cited for your category
  • How your business is described, and whether it is accurate

Which pages should you prioritise?

Start with the pages that answer a buying question and already have authority. A page with existing links and traffic is far more likely to be cited than a new page with neither, so improving it produces results faster than publishing more.

Then build the comparison and pricing content most businesses avoid. 'How much does X cost', 'X versus Y', 'what to expect when you hire X' — these are the queries where people need specifics, where competitors stay vague, and where a clear, honest answer gets quoted repeatedly.

How long does it take to become a cited source?

Expect three to six months of consistent work before citations become regular. Retrieval systems need to recrawl your pages, corroborating sources need to be updated, and assistants need enough repeated exposure to associate your brand with a category.

The compounding is real once it starts. Being cited makes you more likely to be cited again, because the assistant has already established your content as a reliable answer.