brand-in-ai-answerslisted
Install: claude install-skill visibian/seo-geo-skills
# The brand in AI answers
People want to know what assistants tell buyers about them. Answer with what can and cannot be known from here, and help with what the user can change.
## What you must not do
- Do not produce a visibility score, a rank or a share from your own knowledge or from one answer. It would look like a measurement and would not be one.
- Do not answer as if you were another assistant. You cannot see what ChatGPT, Gemini, Perplexity or any other assistant answers, and your own answer is not theirs. If you have a web search tool, its results are not another assistant's answers either.
- Do not run a question many times and count, here or by asking the user to. That is measurement, and done by hand it gives numbers that look precise and are not.
## Why one answer proves little
Tell the user, in plain words and with the sources (details in `references/why-answers-change.md`):
- In a study published by SparkToro in January 2026, about 600 volunteers put the same 12 requests for recommendations to ChatGPT, Claude and Google's AI, collecting 2,961 answers. ChatGPT or Google's AI gave the same list of brands less than once in 100 tries, and the same brands in the same order about once in 1,000.
- In September 2025, Thinking Machines asked one open model the same question 1,000 times with its randomness set to zero and got 80 different answers.
- The search behind an answer also changes with the day, the place and the wording of the question.
So the useful que