Ask six assistants about your company.
Before you fix what AI tools say about you, read what they say. This is how we ask, and why the transcript matters more than the score.
1. Why ask at all
Assistants describe your brand to your customers whether or not you have given them anything to describe it from. What they say comes from whatever they found.
Most teams have never read those answers. It is the one part of the job nobody can do from the inside: what the tools say about you when you are not in the room.
2. Six assistants, 20 questions
We ask ChatGPT, Claude, Gemini, Perplexity, Grok and Copilot, the same six every run. Each gets the same 20 questions, so one run is 120 answers.
The questions cover five dimensions, and each asks something different of the assistant:
- Description — what it says you are, and who you are for.
- Recommendation — whether it suggests you when somebody asks for what you do.
- Competitive position — where it puts you against the alternatives.
- Contamination — what has leaked in from elsewhere.
- Evidence gaps — where it has nothing of yours to go on.
One run: 6 assistants × 20 questions
3. Asked cold
Every assistant is asked logged out, with no memory, no custom instructions and no earlier turns. The questions go in order, one session per assistant, with no follow-ups.
That is deliberately unhelpful to the assistant. It is also exactly what a customer gets, and the customer is who the answer is for.
4. What a single question turns up
Asked what a company sells and who it is for, six assistants need not give six versions of the same answer. We have not published a run yet, so these are the kinds of answer the method is built to catch:
- Describes you as an agency.
- Correct, with a price from two years ago.
- Confuses you with a competitor.
- Quotes a page you deleted.
- Correct.
- Will not answer.
Every one of those is a different problem with a different fix, which is why a single number is never the finding.
5. What you get back
Dated, verbatim transcripts of all 120 answers. A source trace — which stale page or missing evidence drives each bad answer. And a one-page list of findings and fixes.
The transcript is the deliverable; the score is a reading of it.
The Context Diagnostic is free, once per brand.
