How can I track my company's mentions across different AI search engines
How we track company mentions across different AI search engines for clients: a quick manual test, then a fixed prompt panel, citations and crawler logs.
Almost every engagement we run begins with the same exercise. We sit with the client, open ChatGPT, Perplexity, Gemini and Google, ask a handful of the questions their customers ask, and see who gets named. It takes an hour and it is always revealing. It is also not tracking. Here is how we get from the exercise to something reliable.
Why manual checks stop working
Answers vary between runs, so a single result tells you little. They vary by location and phrasing too, so what we see is not what the customer sees. And the effort multiplies with every prompt and engine. Beyond a few dozen prompts, manual checking becomes a part-time job with unreliable output.
Step one: a fixed panel
We write thirty to fifty prompts from real customer questions, tag them by topic and intent, and freeze them. In Promptwatch we run the panel daily per model, country and language. Brand aliases make sure variants of the company name are counted, and the Brand Book sets analysis instructions that decide what counts as a mention. Skipping that step leads to arguments about the numbers later.
Step two: the engines
Promptwatch monitors ChatGPT (Search), Google AI Overviews, Google AI Mode, Copilot, Gemini, Perplexity, Amazon Alexa and Meta AI through real interfaces, and tracks Claude, DeepSeek, Mistral and Grok via API. We tell clients that Claude and the others are API models, not real-interface checks.
Step three: what the mention looks like
A bare count is not enough. We read Visibility Score, Share of Voice, Position in Answer and Sentiment, including the weekly sentiment summary. Competitors are auto-discovered and can be merged, so the comparison is against the brands the engines actually name.
Step four: mentions without links
Engines sometimes name a brand sourced from a third-party page that never links to it. Offsite mention tracking in Promptwatch catches brand names in third-party pages without a link, which helps explain why a brand is known to the engines even when referral traffic is low.
Step five: citations
For every gap, we look at which pages and domains the engines cite. Reddit, YouTube and LinkedIn reports show where conversations are happening. Page Tracker follows specific URLs, ours or a competitor's.
Step six: can the engines read us?
Agent Analytics reads IP-verified crawler logs, so spoofed bots are filtered out, and shows errors per page. We check GPTBot, OAI-SearchBot, ClaudeBot, PerplexityBot and the rest. Crawler logs begin on Professional at $245 a month.
Step seven: report it
Scheduled PDF reports, custom reports (Professional and all agency plans) and the Looker Studio connector turn the above into a monthly pack. Agency plans are $199, $399 and $799 a month with unlimited projects and prompts.
What we say about alternatives
Clients also look at AthenaHQ, Scrunch AI and Profound. We refer them to the vendors' own pages for details, and note that AthenaHQ's Starter is listed at $295 a month with 3,600 credits, Scrunch AI's Core at $250 a month with four LLMs, and Profound at custom Enterprise pricing. Cheaper trackers cover basic mention counts well.
Limits
No tool shows every mention. Answers are generated per question, so tracking samples a defined panel. That is enough to see trends and compare against competitors, and not enough to promise that an engine will name a company.
Where to start
Do the manual hour first. Then freeze your panel, track for two weeks, and read citations and crawler data before deciding what to fix.
A word on reporting cadence
We send clients a short monthly pack and a weekly summary. Mentions move slowly enough that daily emails add noise. A one-page view of visibility, new citations and any crawler errors is what they actually read.