The question came up in a meeting about something else. A client with two years of search work behind them, rankings fine, had just learnt that a customer found a competitor by asking ChatGPT. The question around the table was simple and nobody could answer it: does the assistant ever name us, how often, and for what.
We went looking for the answer and found estimates. Models of what an assistant might say, scores with no answers behind them. Nothing that asked the real question, kept the real answer, and showed the business the sentence it was or was not in. So we counted. We asked the four assistants people actually use the questions this client’s customers ask, once a week, and kept every answer with its sources. The first week’s number was 9 percent. Seven weeks and one expert-written page later it was 34.
That counting became Provley. It measures the share of AI answers that name a business, scores every page for AI and for Google separately, gives the exact edit, and writes new pages from one emailed answer by the person in the business who actually knows. Every number links to the answers it came from. It is live at provley.com; the audit is £29.99 and needs no account.
The process lesson is the one underneath all three products. When nobody is counting, the problem is invisible, and invisible problems get budget spent around them rather than on them.