How to track the third-party sites AI uses to recommend brands (and the tool we use)

Clients usually ask the same question in the first call: "Why does ChatGPT recommend our competitor?" The honest answer is almost never on the competitor's own website. It is on the five or six outside pages the model read before it answered. Tracking those pages is the most useful single thing we do in the first month of an AI visibility program.

Start from the citations, not the mentions

A mention count tells you where you stand. A citation list tells you why. When we set up a new client, we run a prompt set that reflects real buying questions, then look at which domains and exact URLs the engines cite for each prompt. Patterns appear fast. For one category the sources are mostly comparison listicles. For another they are Reddit threads. For a third it is YouTube reviews. The tactic that works is different in each case, and you cannot guess which one applies.

We sort sources into types before doing anything else:

• Comparison and roundup pages. The fix is a conversation with the publisher, or a better product page that gives them something accurate to cite.
• Forum threads, especially Reddit. The fix is honest participation by someone who actually knows the product, not drive-by links.
• Video. The fix is a review or walkthrough that exists in the first place.
• Your own pages. The fix is on-site work, and crawler access matters here.
• Mentions without links. Pages that name the brand but link nowhere. They count in AI answers and are easy to overlook in link-based tools.

What we use and why

We run client programs on Promptwatch, and this part of the work is the reason. Its citation analytics gives page and domain level citations with URL positions. Reddit, YouTube and LinkedIn have their own citation reports. Offsite mentions surfaces pages that name a brand without linking to it. The Page Tracker lets us watch one specific URL, a client's or a competitor's, so we can see whether a placement we earned is actually being cited.

Citation Trends is the report we open most. It classifies each citation by content type and source type and shows how a page's citation rate moves over time: when it was first cited, how long it took to ramp up, its peak, and how quickly it decays. That tells us whether a roundup placement is worth chasing or whether the engines already moved on.

The rest of the platform matters because source tracking alone ends in a spreadsheet. Agent Analytics shows IP-verified AI crawler visits, so we can tell whether a client's own page is being fetched at all. 360 Insights joins crawler logs, citation clicks and AI-referred visits per page. The Action Board turns findings into a prioritized to-do list that we assign and comment on with the client. Most of the pure trackers we compare against stop at the report.

A simple monthly routine

1. Pull the top cited domains per topic. Not overall; per topic. A domain that dominates "best CRM for startups" may be absent from enterprise prompts.
2. List the third-party pages that name competitors but not the client. These are the gap list.
3. Mark each by tactic. Outreach, community, video, on-site fix, or ignore.
4. Track the pages we influence. Add them to the Page Tracker and revisit in four weeks.
5. Compare periods. Promptwatch's A/B period comparison shows whether visibility and citations moved after the work.

Plan fit

Plan choice matters for this workflow. Essential ($95 a month) includes the citation reports and 50 prompts, but not crawler logs. Crawler logs start on Professional ($245). Business ($579) adds the automated Content Agent and Ads Radar. For agencies, the self-serve plans are Kick-off ($199), Growth ($399) and Scale ($799). They include unlimited projects, all AI models and crawler logs, and they do not include AEO articles or the Content Agent. There is a free Explore plan for one project with 10 prompts and ChatGPT only, and a 7-day trial on monthly billing for the paid brand plans.

What to tell clients up front

We say two things early. First, tracking does not guarantee a recommendation. It tells you where to spend effort. Second, the set of cited pages moves. A source that matters this month can disappear next month, so the list is a living document, reviewed monthly. Clients accept that quickly when we show them a trend line rather than a snapshot.

If you only do one thing after reading this, run your ten most commercial prompts, export the cited domains, and read the top five pages yourself. The pattern is usually obvious within an hour, and it will shape your plan more than any dashboard score.