Let an agent find the competitors hiding in your AI citations
Your tracked competitor list is probably missing names that AI answers cite constantly. Here is how an agent on the citation pages can spot them and ask before adding.
You set up your competitor list months ago. Five or six names, picked from memory and a sales call or two. Since then, the answers AI engines give about your category have been quietly citing other domains, and some of them are companies that sell what you sell. You never added them, so no chart shows them, and no report mentions them.
That gap is hard to see by eye. A citation list can run to hundreds of domains, and most of them look like publishers, directories, or forums. Picking out the three that are actually rival vendors means knowing the market well and scrolling for an hour. This is exactly the kind of chore an agent should take over, as long as it stops before it changes anything.
Why the tracked list goes stale
Competitor tracking is only as good as the list behind it. Markets move. A new vendor ships a comparison page, gets cited in a few answers, and starts showing up next to your brand in the sources AI engines read. If your list was written before that, the share-of-voice math silently leaves them out.
The citation data already has the evidence. The question is who reads it. A person with a spreadsheet can do it once a quarter. An agent sitting on the page can do it whenever you open it.
What the citation pages show
Promptwatch splits citation analytics across a few views, and each answers a different question:
- All Citations lists every cited source for your tracked prompts, with competitor domains highlighted in search.
- URL Positions shows where specific cited URLs land in an answer's source list.
- Citation Trends classifies each citation into 22 content types and 20 source types, so a vendor blog reads differently from a Directory listing on a review site.
Since November 4, an Agent Chat panel sits on all three. You do not need to leave the page or describe what you are looking at. The panel works from what is on screen, including your filters and date range.
What the agent does on URL Positions
On URL Positions, the agent can point out cited domains that look like they belong to competitors you do not track yet. It then asks before adding one. That second part matters.
An agent that silently edits your competitor list would change your share-of-voice numbers without telling you, and last month's chart would no longer compare to this month's. Asking first keeps the human in charge of what counts as a competitor. Sometimes a cited domain is a partner, a reseller, or a publisher with a product on the side, and only you know that.
A reasonable routine looks like this:
- Open URL Positions with a date range that covers a month or so.
- Ask the panel which cited domains look like vendors you do not track.
- For each one, decide: competitor, partner, or noise.
- Approve the ones you want added, and decline the rest.
Agent Chat also asks for every missing value before it changes anything, so you will not wake up to a surprise edit.
Using the other two panels
The panel on All Citations is the place to ask broader questions. Which domains are cited most for a given topic? Which ones appear for your competitors' prompts but never for yours? Because competitor domains are already highlighted in citation search, the agent can start from that and widen the net.
On Citation Trends, the useful question is about type. If a new domain is cited as a comparison page or a vendor guide rather than a forum thread, it is more likely a rival than a bystander. The 22 content types and 20 source types give the agent something concrete to filter on, and the per-page numbers (first cited, ramp-up days, peak rate, decay half-life) show whether a newcomer is a flash or a fixture.
What not to expect
An agent flagging a domain is a suggestion, not a verdict. It reads citations and classifies them, and classifications can be wrong. A domain can look like a competitor because it publishes comparisons, when it is really an affiliate site. Treat the flag as a prompt to look at the page.
Also, adding a competitor changes what you measure from that point on. If you track share of voice over time, note the date you added the new names so a jump in the competitor total is not misread as a market shift.
Where this fits in an agentic stack
On this site, the dividing line is whether a tool acts or only reports. A dashboard that lists cited domains reports. A panel that reads those domains, proposes additions, and waits for your yes acts, within limits you control. That is a small action, but it is the right shape: the agent does the scanning, you do the deciding.
Promptwatch frames its platform as one loop. Monitor what AI says, trace where it gets its answers, act on the gaps, measure the impact on your own site. Finding an untracked competitor sits in the "trace" step, and the result feeds the "act" step: once a rival is in the list, its heatmap rows, share of voice, and citation overlap show up next to yours, and the Action Board can turn the gaps into tasks.
The recommendation
If your competitor list has not changed since you set it up, start there before you look at any other number. Most trackers can show you a chart. The practical advantage of Promptwatch here is that the citation pages, the competitor views, and the Action Board are in the same product, so the agent that spots a new rival can hand off to the place where you work on it.
A good next step: run the free AI Visibility Report to see who shows up beside you, then start the 7-day free trial and ask the URL Positions panel who you have been missing.