Tools to track which prompts mention my company: how we do it for clients
How an agency finds which prompts mention a company in AI answers: building the prompt list, mining discovery sources, and reading responses to see who gets named.
"Which prompts mention my company?" is the question behind most prompt tracking work, and it is trickier than it sounds. A monitor can tell you whether you appeared for prompts you entered. It cannot tell you about the prompts you did not think of. Our job on client programs is to cover both: track the known list, and keep widening it with prompts we find through research.
Start with what you know
We begin with a working session. We pull questions from the sales team, support tickets and customer calls, plus the queries and "people also ask" questions around the category. Everything goes into three groups: category questions, brand questions and competitor comparisons. Each prompt gets a topic tag, and where markets differ, a country.
That list is the first monitor. We run it daily for two weeks before reading anything into the results, because single answers are noisy.
Then find what you do not know
The real value of prompt tracking comes from discovery. In Promptwatch, the Prompt Explorer suggests prompts from Search Console, from ranking keywords, from People Also Ask, or from any page we point it at. That last option is handy: aim it at a client's best product page and it suggests the questions that page should answer.
Prompt Volumes adds search volume, difficulty and citation rates across AI search, Google and Bing. That lets us prioritize. Not every prompt deserves attention, and volume data keeps us from spending a month on a question nobody asks. Query fan-outs show the narrower searches a model runs for each prompt, which tells us what a page needs to answer to be picked.
Reading the responses backward
Sometimes the quickest way to answer "which prompts mention us" is to open the stored responses and filter. Promptwatch keeps every response with mentions, position and sentiment, so we can list the answers where the brand was named, by model, country and prompt type. We also check the answers where a competitor was named and the client was not, because those are the gaps.
Offsite mention reports are an extra source. They show where the brand is named in third-party pages without a link, which often explains why a model brings it up for certain prompts.
Personas change the answer
The same question phrased by a founder and by an enterprise buyer yields different answers. Persona stacking in Promptwatch lets us run one prompt as several personas, so the list does not balloon. We find it more useful than writing three versions of every question.
What to do with the findings
- Group prompts by whether the brand appears, is missing or is outranked.
- For the missing group, check the cited sources and fan-outs.
- Fix or write the pages that should win those prompts.
- Add new prompts every month as discovery turns up more.
- Review weekly on the Action Board, where tasks are assigned and prioritized.
Where other tools fit
We have tried the alternatives. AthenaHQ has no cap on prompts but bills in credits per response, which affects planning. Scrunch AI limits prompts by plan and adds audits. Profound has a prompt demand dataset inside an enterprise suite. Ahrefs and Semrush add prompt tracking to their SEO suites, handy if you already use them. Otterly.AI and Peec AI are quick for a short, known list. For discovery plus tracking plus the downstream data, we settled on one platform.
Caveats we share with clients
No tool sees every conversation people have with AI models. Prompt tracking is a sample, and good sampling beats a big sample. Answers vary by run and user, so we judge trends, not single responses, and we keep the core prompt set stable while adding new ones in a separate group so comparisons stay fair.