How can you tell if your competitors are getting cited more than you in AI answers?

This is the question clients ask us most often once they have seen a competitor named in a ChatGPT answer. They want a yes or no. What we can give them is a number, and the number only means something if the method behind it is sound.

The test is a panel, not a screenshot

One answer proves nothing. AI engines vary their output between runs, between countries and between phrasings. So we build a panel: a fixed set of prompts, run daily, across the engines the client's buyers use. Then we read the rates.

We split prompts into three groups. Category prompts ("best tools for X"), comparison prompts ("A versus B") and problem prompts ("how do I stop doing Y"). Each group tells a different story. Clients are often ahead on comparison prompts about themselves and behind on category prompts, where the buying decision begins.

Mentioned versus cited

We track two things separately.

A mention is the brand name appearing in the answer. A citation is a source link or listed source behind the answer. Competitors can win on one and lose on the other. A rival that is named often but cited rarely is riding on third-party content. A rival that is cited often without being named is giving the engines the facts, which tends to turn into mentions later.

Share of voice sits on top: your mentions as a share of all brand mentions across the panel. Citation share is the same idea applied to cited pages and domains.

What we use to run the panel

For client programs we use Promptwatch. The relevant pieces for this question:

• Competitor visibility and heatmaps. The heatmap compares any two dimensions, so we can lay competitors against engines, topics or countries and see where the gap is concentrated.
• Auto-discovered competitors. Clients often name the wrong rivals. The tool surfaces brands the engines actually put next to the client, which sometimes changes the whole conversation.
• Brand merging and aliases. If a competitor appears under three spellings, we want one line in the report.
• Citation analytics. Page and domain level citations with URL positions show which pages win. Page Tracker lets us follow a specific competitor URL over time, and Citation Trends shows how fast a page ramps up and decays.
• Offsite mentions and community reports. Reddit, YouTube and LinkedIn citation reports show where the argument is being made outside anyone's own site.

Brand plans start free on Explore and run $95, $245 and $579 a month for Essential, Professional and Business. Agency plans run $199, $399 and $799 a month for Kick-off, Growth and Scale, with unlimited projects and 10 seats.

Reading the source list

The most useful output is often boring. When we list the domains cited for a client's top twenty prompts, patterns show up fast: a single comparison site, a couple of forum threads, an industry directory. A competitor that leads on citations usually has presence on two or three of those sources that the client lacks.

That list becomes work. Which sources can we get the client onto honestly, with real reviews, real contributions and accurate listings? Which pages on the client's own site should exist and do not?

Check whether the engines can read the client

Sometimes a competitor leads for a dull reason: the client's site is partly invisible to AI crawlers. Promptwatch's Agent Analytics shows IP-verified crawler logs, so spoofed user agents are filtered out. We look at GPTBot, OAI-SearchBot, ChatGPT-User, ClaudeBot, PerplexityBot and the others, then at error rates by page. Crawler logs begin on Professional on brand plans and are on every agency plan.

If bots cannot fetch the pages, no amount of content work will fix the citation gap.

Cautions we give every client

Do not conclude from twenty prompts. Wait for the trend.

Do not build the panel only from prompts you expect to win. Include the awkward ones.

Do not treat citations as traffic. We add visitor analytics so we can see whether AI-referred visits turn into conversions, because a cited page can still send little traffic if the answer satisfies the reader.

And do not promise a client that a fix will produce a recommendation. We can show where they trail and what is likely to close the gap. We cannot control what an engine says.

A first-week checklist

1. Write 50 prompts from sales calls and support tickets.
2. Name five real competitors.
3. Run the panel daily for at least two weeks before reading conclusions.
4. Export the top cited domains for the competitor set and for the client.
5. Mark each domain as "we can get on this", "we cannot", or "unclear".
6. Check crawler logs and fix access problems first.

If you want to try it on your own brand, the free plan on Promptwatch is enough to see whether a competitor is out-citing you on your highest-value prompts.