How the score is calculated

Every prompt in an analysis is sent independently to each active model. There is no aggregation layer, no proxy model and no cached corpus — each score reflects what that specific model returned at the time of the run.

Paid plans query all six. The free plan queries the first two available (normally Claude and GPT-4o), which is why a free score is directionally useful but narrower than a paid one. Each model reply is capped at 400 tokens, so the score reflects a model's primary answer, not an exhaustive one.

For every response we count case-insensitive occurrences of your brand name, and of each competitor name, in the returned text. The count is a plain string match — no fuzzy matching, no synonym expansion, no LLM judging whether a mention “really counts”.

That choice is deliberate: it means the same response always produces the same count, and you can verify any number yourself by reading the raw response, which we store and show you.

Each prompt is scored per model:

A model's score is the points it awarded divided by the maximum possible across all prompts, expressed out of 100. Your overall score is the plain average of the per-model scores — every model counts equally, so a strong showing in ChatGPT cannot hide a zero in Perplexity.

The second half of the formula is why the score moves when a competitor gains ground even if your own mentions are unchanged. Being named is table stakes; being named more is the position that wins the recommendation.

For each mention we read a window of roughly 150 characters either side and weigh it against a fixed lexicon of positive and negative terms. The verdict is positive, neutral or negative per response, then summarised across the run.

This is a blunt instrument by design — it is transparent and stable rather than clever. It will miss sarcasm and subtle framing. Sentiment is available on paid plans; free analyses report every mention as neutral rather than guessing.

A visibility score answers whether AI sends people to you. It says nothing about what happens when they arrive — and a brand that is recommended but not believed converts no better than one nobody hears about.

That is what the AI Focus Group measures: a panel of synthetic customers, generated from evidence on your own site, walks your pages and reports where they stop believing you. Those results are clearly labelled simulations — they are directional, and they do not replace research with real customers.

The two combine into a single AI Customer Readiness score, which weights discovery and persuasion separately so you can see which half is actually failing.