One ChatGPT screenshot is not a baseline

Someone in the firm types a prompt, gets a name, forwards a screenshot to the partners. Useful as a spark. Not a measurement.

UK mid-market firms that already invest in being found know this pattern well. A marketing lead pastes ChatGPT. A partner replies with Perplexity from their phone. Another says Google still sends work. All three can be true the same week. None of them is a baseline. Anecdote is cheap because it feels like evidence while remaining unrepeatable.

Engines change. The same prompt on a different day, a different model, or with a different preface returns a different shortlist. A capture taken after “act as a buyer” is not the same as a bare category question. A brand prompt that already contains your name is not the same as a category prompt that does not. One image cannot tell you whether you are generally absent, generally named, or named only when the buyer already knows you.

That is why partner meetings stall. The screenshot says a rival appeared once. It does not say on how many prompts, against which buyer, on which engine, or whether the result holds tomorrow. People argue about a single frame. Nobody can say “four of eighteen category prompts” because nobody locked eighteen prompts and re-ran them.

A baseline is a versioned set. Locked questions. A fixed date. The engines we can actually score that week — not a live demo of whatever is open on someone’s laptop. For each prompt you record mention, who is named instead, and the sources the models lean on. Then the conversation changes. You stop saying “we think we are invisible.” You say where you appear, where you do not, and which firms hold the slot when the prompt is the job, not the brand.

Accountancy examples make this concrete without claiming every reader is an accountant. “Best accountant for a £10m UK firm” on Tuesday in one chat window is a spark. The same line, plus seventeen related prompts, re-run on a fixed date across the engines in scope, is a measurement. A law firm or consultancy uses the same discipline with different nouns. The unit does not change.

We will not run a few prompts live on a sales call and call it evidence. That theatre produces the same problem as the partner screenshot: unreproducible, easy to game with preface, impossible to defend in a board pack. The Geo Audit exists to replace that with a written pack and a working session — locked set, scored engines, named-instead list, order of work.

Firms that already pay for service pages and content often already have ranking reports. Those reports answer a different question. They do not tell you whether a serious buyer who never searches your brand gets your name when they ask an engine who to hire. Mixing the two surfaces without a locked set is how anecdote keeps winning the room.

The discipline sounds slower than a screenshot. It is. Speed without a locked set only accelerates a wrong conclusion. Partners who want a decision need a date, a set, and a re-run — not another capture from someone’s phone.

If someone has already forwarded a screenshot, keep it. Treat it as the spark that opened the conversation. Then write the buyer prompt in one line, expand it into a set, and re-run on a date you can cite. Until then you have a story, not a baseline.