What gets asked
On each run, every active prompt in a project is put to every AI assistant switched on for that project. With 25 prompts and three assistants, that’s 75 answers a run. Prompts are asked in each assistant’s real interface — the same ChatGPT, Gemini or Perplexity a person uses — not through the developer API. The API and the interface often answer the same question differently, and the interface is what your buyers see. Each prompt is asked from your project’s country, because assistants recommend different companies in different markets.When it runs
By default, prompts run once a day. You can run them every 3, 7, 14 or 30 days instead, pause them, and choose the hour they run (in UTC) in the project’s Settings. See Models and schedule. New prompts don’t wait for the schedule: they’re asked as soon as you save them. Answers are filed under the UTC day they were collected.When an assistant is having trouble
AI assistants go down or block automated traffic from time to time. When one stops returning answers, Growth Mention pauses it and resumes automatically once it’s answering again. Runs that fail are retried later the same day. A paused assistant leaves a gap in that day’s data rather than a wrong number. You don’t need to do anything.What counts as an answer
Not everything an assistant returns is an answer. Two things are left out of every metric:- Empty responses — nothing, or almost nothing, came back.
- Refusals and errors — the assistant declined, reported a problem (“I’m having trouble responding right now”), or showed a sign-up wall instead of answering.
These are excluded because every metric divides by the number of answers. Counting a failed response as “an answer that didn’t mention you” would pull every brand’s visibility down for a reason that has nothing to do with them.
What is read out of each answer
Each answer is read by an AI model that records, for every brand named:
Your brand is found by its name, by any aliases you add in Settings (abbreviations, former names), and by spelling and spacing variants — Growth Test and Growthtest count as the same brand without an alias.
Alongside the brands, each answer’s cited sources are recorded. For ChatGPT, so are the web searches it ran before answering — see Query fan-outs.