It read something before it answered.
And it was probably not your website. The cited sources are the most actionable half of any reading, because they turn "be more visible" into a list of specific pages.
An assistant with web search on does not answer from memory. It runs searches, reads a handful of pages, and composes from what it found. Those pages are usually listed underneath the answer, and almost nobody looks at them — which is strange, because they are the only part of the whole exchange you can actually do something about.
What it reads, in practice
In one run against a local hospitality category, a single roundup article was cited 126 times across the run. The next four sources were two review-site domains, a city guide and a hotel’s recommendations page. The brand’s own website was not among the most-read sources at all.
That pattern repeats across categories. The sources that decide an answer are overwhelmingly third-party: roundups and listicles, review aggregators, forum threads, city and category guides, and the occasional video description. Company websites appear, but they lose to a page that compares several companies, because the question being answered is comparative.
The sources it cites are your press list. Not the domains with the best ratings — the pages that actually influenced the answer about your category.
Why your own site loses
Not because it is bad. Because of what it can say. Your site can state that you are excellent; it cannot credibly state that you are the best of five options, and a model composing a recommendation needs the second kind of sentence. A page that names several businesses and distinguishes between them is doing the work the assistant is trying to do.
There is a second reason and it is more mechanical. A roundup is one page that answers the whole question. Your site is one page that answers a fifth of it, and retrieval rewards the page that covers the query.
Using the list
Pull the cited sources for the questions where you are absent, and you have something better than a content strategy: a specific, finite list of pages that are deciding answers about your category without you in them.
- A roundup that omits you. Often the single highest-leverage item on the list, and often reachable — most are maintained by somebody who takes submissions.
- A review site with three reviews on your profile. The aggregate is what gets read, and the threshold for being described at all is lower than people assume.
- A directory or city guide you are not listed in.
- A forum thread asking exactly your buying question, where nobody from your company has ever replied.
What this is not
It is not a ranking factor list, and anybody selling you one for a language model is selling a guess in a confident voice. Citation frequency tells you which pages influenced answers in your category over a measured period. That is a pattern to act on, not a formula, and it will change as the underlying pages change.
The short answers
What sources does ChatGPT use to recommend businesses?
With web search on it runs searches and reads a handful of pages before composing. In practice those are overwhelmingly third-party: roundups and listicles, review aggregators, forum threads and category or city guides. In one of our runs a single roundup article was cited 126 times.
Why does the AI not cite my website?
Because of what your site can say. It can state that you are excellent; it cannot credibly state that you are the best of five options, and a comparative question needs the second kind of sentence. A roundup also answers the whole query on one page, and retrieval rewards that.
How do I find out which pages the AI read?
Engines with search on list their sources under the answer. Collect them for the questions where you are absent and sort by how often each domain appears. That gives a finite list of pages deciding answers in your category without you in them.
Should I try to get onto every cited source?
No. Sort by citation frequency and work down. The page cited a hundred times matters more than the publication you would rather appear in, and most of the list will not be reachable at all.
Do review sites matter for AI recommendations?
They matter more than most businesses expect, because the aggregate is what gets read rather than any single review, and the threshold for being described at all is low. Review domains appear consistently among the most-cited sources.
Is citation frequency a ranking factor?
No, and treating it as one is a mistake. It tells you which pages influenced answers in your category over the period you measured. That is a pattern worth acting on, not a formula, and it moves as the pages themselves move.
Find out in under ten minutes
One free check, no account and no card. You get the answers the assistants gave, not a score.
Every figure quoted here comes from a real question put to a real assistant, saved with the date and the assistant that answered it. This piece was written to answer the search "what sources does chatgpt use".