How AI cites sources: what gets quoted and why

If you have used ChatGPT or Perplexity recently, you will have noticed the little numbered citations next to claims in the answer. Click one and it takes you to the source the AI relied on. Those citations are not decorative. They are the mechanism by which AI engines build trust, and they tell you exactly which sources you need to be in. This article explains how citation works, which sources get quoted, and what your business can do about it.

Why AI engines cite sources

AI models do not know everything. They generate text by predicting likely next words, which means they can be confident and wrong at the same time. To ground their answers in fact, modern AI search engines retrieve real web pages, read them, and cite them as evidence for each claim.

The citation serves two purposes. It lets the user check the claim, which builds trust in the answer. And it forces the model to only make claims it can back up with a source, which reduces hallucination. The result is that the businesses and facts that appear in AI answers are the ones that appear in the cited sources.

Which sources get cited

AI engines do not cite every page they read. They cite the ones that most directly support a specific claim. Through analysing many answers across sectors, a pattern emerges in what tends to get cited:

Review and comparison sites. TripAdvisor, Trustpilot, Yelp and sector-specific review platforms are cited heavily for “best” and “recommend” questions, because they aggregate independent opinions.

Directories and registers. Professional registers, Google Business Profile data, and industry directories are cited for factual claims about what a business is, where it is, and what it does.

Press and media. News articles, features and guides from recognised publications are cited for authority claims, because the model treats them as independent and trustworthy.

Official websites. A business’s own website is cited for factual details about that business, but it carries less weight for evaluative claims like “the best”, because the model knows it is self-serving.

Forums and communities. Reddit threads, expat forums and specialist communities are cited surprisingly often, because they contain direct, specific answers to narrow questions.

The key insight is this: for evaluative questions, the model wants independent sources. For factual questions, your own website is fine. Knowing which type of question your buyers ask tells you which sources to target.

How a business becomes a cited source

You cannot ask an AI engine to cite you. There is no submission form. You become a cited source by being present, accurate and consistent in the places the model already reads. That means a few concrete things.

Be in the directories the model reads. For clinics, that means Doctoralia and Top Doctors. For estate agencies, that means Idealista and the major portals. For hotels, that means Booking.com and TripAdvisor. Find the directories your sector relies on and make sure your listing is complete and accurate.

Earn reviews that mention specifics. A review that says “the dentist explained everything in perfect English” is citable evidence for the claim “English-speaking dentist in Malaga”. A review that says “great service” is not. Encourage customers to mention what they bought and how you helped.

Get mentioned in independent media. A feature in a local English-language publication, a sector guide, or a travel blog gives the model an authoritative, independent source to cite. One good mention can feed many answers.

Make your own website citable. Clear, factual pages that answer questions directly are more likely to be retrieved and cited than vague marketing copy. “Implants from 900 euros, Dr Ana García, English spoken” is citable. “Excellence in dental care” is not.

The consistency multiplier

One source mentioning you is good. Several sources describing you the same way is much better. When the model sees your business described consistently across your website, your directory listings, your reviews and a press article, it becomes confident enough to state it as fact and cite any of those sources as evidence.

When sources conflict, the model hedges. It might say “some sources describe X as a boutique hotel, while others list it as an apartment complex”. That hedging pushes you down the answer. Consistency is what turns a maybe into a confident mention.

What to check this week

Run three of your buyer questions through ChatGPT and Perplexity. For each answer, click every citation and note where it leads. You will quickly see which sources the engines rely on for your sector. Then check whether your business is accurately represented in each of those sources. The gaps you find are your priority list.

This is the most practical way to understand AI citation: see it for yourself, source by source. Once you know which sources the engines cite for your questions, you know exactly where to focus your effort.