claude skills / Claude Skills Pro / ai-visibility-audit / en français

AI visibility audit (GEO).

Know what AI assistants say about you when a customer asks, and fix what is in your hands.

See what it produces
  • ChatGPT
  • Gemini
  • Perplexity
  • Google AI Overviews
  • Google Business Profile
  • Google Sheets
8 to 15buyer questions, never with the brand name
3+runs per question and per engine, never a single answer
4 to 6 wksbefore the re-test, same list, same protocol

orders of magnitude, depending on your volume

the problem

What gets in the way today.

  • 01Your customers ask ChatGPT "who should I contact for..." and you have no idea if your name comes up.
  • 02An AI gives out your old address or a service you stopped offering.
  • 03You tested once, on one day, and you have nothing to compare against next month.

input, output

You give this. You get this.

What you give Claude: the business, its services, its markets and the facts of record

Company: Vellora Translations (vellora.example.com)
Services: certified legal translation, contract translation, civil documents
Markets: France, Lyon and Paris, English and French
Competitors we know: Lexmond Traduction, Aurore Linguistique
Key facts: legal translation since 2024, office in Lyon 7e since 2025, price per page, quote on request
Previous measurement: none

$ claude "Run an AI visibility audit for Vellora Translations across ChatGPT, Gemini, Perplexity and Google AI Overviews."

AI visibility audit, report excerptexample

AI visibility audit: Vellora Translations

Tests: 2026-11-10 to 2026-11-12 | Engines: ChatGPT, Gemini, Perplexity, AI Overviews

READ THIS FIRST

These answers change between days, users and locations. This report is a dated sample under a fixed protocol, not a guarantee of future answers.

SUMMARY

Mention rate: 14 of 114 discovery runs (12%)

Average position when named: 3.4

Accuracy: 11 of 14 facts correct

Machine readability: 50/100

RESULTS BY ENGINE

ChatGPT: 30 runs, brand named 4 times, top competitor Lexmond Traduction (19 of 30)

Perplexity: 30 runs, brand named 7 times, most cited source transdir.example

AI Overviews: shown on 8 of 10 queries, brand named once

INACCURACIES

ChatGPT, B01: "Vellora is a translation agency focused on tourism content."

Correct fact: legal and certified translation since 2024

Likely source: old listing on transdir.example

FIX PLAN

1. Update the transdir.example listing (cited 31 times). High effect, small effort, week 1

2. Remove or redirect the old pricing page. High effect, small effort, week 1

3. Create pages: certified translation Paris, contract translation. High effect, large effort, week 4

4. Organization and Service JSON-LD with sameAs. Medium effect, small effort, week 2

RE-TEST

2027-01-12, same question set VT-2026-11, same protocol.

the method

What Claude does, in order.

  1. step 1Build the questions

    Claude writes 8 to 15 buyer questions without your name, plus 3 to 5 brand questions to check accuracy, then freezes the list.

  2. step 2Set the protocol

    Clean session, personalization off, same wording, 3 runs per engine, with date, mode and location logged so the test can be repeated.

  3. step 3Score and diagnose

    Mention rate, position, share of voice, accuracy, cited sources and machine readability of the site, then 3 to 6 findings tied to a cause.

  4. step 4Plan and re-test

    Fixes are ranked by effect and effort, each with an owner, then a re-test date on the same frozen list.

demo in French / click, try it

Your turn.

An interactive demo of the skill. It is in French, the pack works in any language: ask Claude to answer in English.

ChatGPT

the film / 44 seconds, with sound

The pack film.

the pack film

In 8 seconds.

the guardrails

What the skill never does.

  • Nobody controls these answers: the report says so up front and promises no citation, no rank and no timeline.
  • Discovery questions never contain the brand name.
  • Every answer is copied word for word and dated, and no score rests on a single run.
  • Claude never simulates an engine's answer. Without access, it prepares the protocol and you run the tests.

Questions

Can Claude query ChatGPT or Gemini by itself?

Only if a tool gives it that access. Otherwise it hands you the question list and the logging sheet, you run the tests, then it scores your results.

Why three runs per question?

A single answer proves nothing, because engines vary from one day and one user to the next. The skill works with rates, not anecdotes.

Is an llms.txt file enough to show up?

No. Some assistants may use it, but Google says it does not need one for its AI features. The real fixes are on your pages, your listings and the sources the engines cite.

What is the re-test for?

To measure before and after with the same frozen list and the same protocol, question by question, 4 to 6 weeks after the fixes.

the pack

In the pack: Claude Skills Pro.