How the score is calculated

The score runs from 0 to 100. It adds three sub-scores: visibility out of 40, credibility out of 30, understanding out of 30. It is shown only when all three are measured. The three calculation rules are the report's own, word for word.

• 01

  Questions asked

  Questions a buyer asks before purchasing, based on the business read on your website. Free asks 5, Pro 10. Their text is in the report.

• 02

  AI assistants queried

  Free: Perplexity, Gemini (flash-lite) and ChatGPT (mini). Pro: Perplexity, Gemini (flash-lite), ChatGPT, Claude and Mistral. Perplexity and Gemini search the web; the others answer from memory. We use their APIs: this is not what a user sees in the ChatGPT or Gemini app, which adds its own search and history.

• 03

  How many answers

  Each question is asked twice to each assistant on Free and 3 times on Pro. Failed or empty answers are discarded. Cut-off answers are kept and flagged. The report shows the number of answers kept per question. Each scan has a measurement budget. If it is reached, the number of answers per question drops first, and the report says so: "Scan reduced to stay within the measurement budget". If an assistant could not be queried, the report says so too.

• 04

  Are you visible? 40 points

  Presence: share of questions where an AI assistant cites you, 15 points. Frequency: share of answers that cite you, 15 points. Position: your average rank in lists, 10 points. An answer cites you when it contains the name taken from your domain (acme for acme.com) or the domain itself. Position points: 1st 10, 2nd 8, 3rd 6, 4th 4, lower 2, cited outside a list 5, averaged over the answers that cite you. Each part is rounded to the nearest point, halves up.

• 05

  Are you credible? 30 points

  Your website among the sources: 6 points. Each third-party site that mentions you: 3 points, up to 15. Each media outlet, expert site or review site among them: 3 more points, up to 9. A site mentions you when your name is in its title, in the excerpt the assistant read, or in the sentence it supports. Only assistants that search the web return sources: Perplexity and Gemini, on Free and Pro.

• 06

  Are you understandable? 30 points

  Consistency between what AI assistants say and your website: 20 points. Share of your services they cite: 10 points. Two questions are asked: "What does [brand] ([domain]) do?" and "Who is [brand] ([domain]) for, and what is its expertise?". A judge model, Mistral Large, compares each answer with the profile drawn from your website: business, targets, services. On Free, only ChatGPT answers these two questions. If your website lists no service, consistency carries all 30 points.

• 07

  Worked example

  AI consulting firm, Pro scan on 11 September 2026. Visible: none of 145 answers cites it, 0/40. Credible: its website is in no source and none of the 256 sites consulted mentions it, 0/30. Understandable: average consistency 35/100, so 7 points; 2 of 5 services cited on average, so 4 points: 11/30. Score: 11/100.

• 08

  What the score does not measure

  What a user sees in the ChatGPT or Gemini app. Your traffic, sales or market share. The tone of the answers. The score promises no gain after changes. It shows no margin of error: its stability from one scan to the next is not measured.

• 09

  Known limits

  We look for the name taken from your domain: a brand with a different name can be missed, a short or common name can be over-counted. The consistency judge is itself an AI model: the gaps it finds are shown so you can check them. Every report says it: "AI assistants change constantly. This report is a snapshot, not a guaranteed value over time." Reports are written in French or English.

Method of 12 September 2026. Model panel changed on 11 September 2026. Versions are pinned and each change is dated here.