Ask the awkward ones first

Plenty of tools will sell you that, and if you want a live chart you should buy one.

We think it's the wrong instrument for the job. The daily number moves mostly on its own, the category leaders barely move at all, and AI is currently about one percent of the traffic to a typical site. Watching it hourly is a lot of attention spent on a small and noisy signal.

Once a month, read properly, with the actual answers attached and three ranked actions, is a better use of your time. If that turns out to be wrong we'll say so.

You can, and honestly you should. Go and ask it the question your best customer would ask, right now. Takes ten minutes and it's usually enough to ruin your afternoon. What takes longer is running it across five platforms, scoring it the same way every month, working out who's beating you and why, and knowing what to do about it.

We don't, until you say so. You approve them before anything runs.

Yes. We set the market when you sign up, because the answers genuinely differ by country. Most of our questions are written for a specific market on purpose. If your buyers are in one town rather than one country, say so and your five questions name the town. Three of the five platforms will also be searched from there, and your report says which.

Then we'll tell you that. We're not in the business of manufacturing a problem. Plenty of businesses come up well on one question and vanish on the next, which is its own useful thing to know.

Because we won't estimate an answer and call it a reading. ChatGPT, Gemini, Grok, Perplexity and Google's AI answers can all be asked directly and captured word for word, the same way, every month. Claude and Copilot can't, not without swapping in a different system and hoping you don't check. So we ask those two by hand, once a quarter, and we tell you that's what we did. Five measured identically every month. Seven once a quarter, two of them read by a person.

Nearly, and it makes no difference. Copilot runs mostly on OpenAI's models, sometimes Anthropic's, sometimes Microsoft's own. It still answers differently to ChatGPT, because it searches a different index and it's been given different instructions. Same engine, different car. That's the whole reason we measure the places people actually ask, not the models underneath them.

Word of Model was built by a marketer with thirty years in digital and creative, who went looking for what the models were saying about a handful of real businesses and found one of them inventing customer reviews that had never existed. That was enough to keep going.

No. If you can describe what you sell, you can use this.