How do I find which sites ChatGPT cites about my brand? How we do it for clients

This is one of the first things we do with a new client, and the process is simple enough to start this afternoon. We do it in two passes: a manual check to get a feel for the answers, then a tracked set to see how they change.

Pass one: ask ChatGPT yourself

Turn on ChatGPT search and ask questions the way a buyer would. Skip the vanity query. "Tell me about [brand]" tells you little. Better:

"Best [category] for [specific use case]"
"[Brand] vs [competitor]"
"Is [brand] worth it for [type of company]"
"Alternatives to [competitor]"

When ChatGPT cites sources, it lists them as links and you can open the sources view. For each prompt we record whether the brand was named, which domains and exact pages were cited, whether the client's own site appeared, and which competitors showed up. After 15 or 20 prompts a pattern forms. The same handful of domains turns up repeatedly: a comparison site, a Reddit thread, a YouTube review, a trade publication, a competitor's blog.

That pattern is already valuable. It tells you where the conversation about your brand and category lives.

Why one check is not a measurement

We never hand a client a manual check as the final answer, for four reasons.

The same prompt can return different sources on another run, so a single pass is a sample.
Location and personalization shift what you see.
There is no history, so we cannot say whether a source is gaining or losing ground.
It does not scale past a few prompts and one engine.

One more point: the sources ChatGPT cites are not everything it read. It runs sub-queries behind the prompt, called query fan-outs, and fetches pages it never credits. Seeing the fetched pages explains cases where a client's page was read but not cited.

Pass two: track it properly

We run client programs on Promptwatch, and the citation tools are why. Here is how we use them for this question.

Citations by page and domain. We see the exact URLs cited for each prompt, their position in the answer, and how often each domain appears. That gives us the list we were building by hand, with history.

Citation Trends. Each citation is classified by content type and source type. For each page we see citation rate over time: first cited, ramp-up days, peak rate, and decay half-life. This separates a durable source from a one-week blip.

Reddit, YouTube, and LinkedIn reports. Where outside voices shape a category, these show which threads, videos, and posts the engines use. "My channels" filtering lets us see how the client's own channels perform.

Offsite mentions. These flag third-party pages that name the brand without linking to it. They often become outreach targets.

Page Tracker. We can track specific URLs, the client's or a competitor's, and see which prompts cite them.

Crawler logs. Agent Analytics shows IP-verified AI crawler hits and a crawl-to-citation path with a citation rate per page. If a page is fetched often and never cited, the problem is the content. If it is never fetched, the problem is access. Crawler logs start on Professional at $245 a month.

For a free start, the AI Visibility Report runs a no-login scan, and the GEO Inspector Chrome extension shows fan-outs and fetched versus cited sources as you browse. It runs locally.

Turning the list into a plan

We sort the cited sources into three buckets.

First, the client's own pages. We keep and strengthen what is cited, and investigate pages that should be cited and are not.

Second, third-party pages we can influence: reviews, directories, comparison articles where information can be corrected or supplied. We prioritize by how often the engines cite them.

Third, sources we cannot influence. We note them and compete by publishing a clearer answer on the client's site.

Reddit and YouTube get extra attention in our reviews, because for comparison questions they often appear. Honest participation in relevant threads, and accurate video coverage, can matter more than another blog post.

What to avoid

Do not buy citations or seed fake threads. It is a bad idea ethically, and engines and communities catch it. Also do not read one run as the truth. Check trends across days and prompts before acting.

And remember the limit. Seeing who ChatGPT cites tells you what influences answers. It does not let you control them. We tell clients that tracking and fixes improve the odds, and we do not promise placement.

A simple routine

Weekly: scan the citation changes on the top prompts. Monthly: review the Reddit, YouTube, and offsite lists and pick two actions. Quarterly: revisit the prompt set, since buyer questions drift.

That cadence keeps the work small and the reporting honest.