← Lab Skills

The Virtual Expert Kit

Decide what AI may say in your name, then test your own AI before your clients do: six client questions, one entry and one boundary in your words.

By Filip Sardi · Updated October 2026 · About 45 minutes

You have put your material into an AI. A custom GPT, a Claude project, a folder pasted into a chat. It sounds like you. What you cannot tell by reading is where it stops being you, because the answer it made up sounds exactly like the one it took from your course.

This kit finds that out the only way it can be found: six client questions where you write the right answer before you ask. Your AI gets the questions, you get the reading, and you leave with one brain entry and one boundary line, both in your own words.

Before you start

If you want this done for you, not tested by you

virtualX takes the material you already have and builds a virtual version of you that answers only from entries like the one this kit leaves you with, refuses in your words, and sends your clients back to you for what only you can do. The boundary is built in, not hoped for.

See virtualX →

Get the skill

Download the guide (PDF) Download for Claude

No sign-up. The guide is the same kit on paper, and your AI can run it from the PDF alone.

What you leave with

  1. Six client questions built from your own method, with the expected answer under each.
  2. A reading of what your AI did with each one: as you expected, the wrong case, refused, or admitted, and every sentence it said that you never said.
  3. One brain entry. The recommendation and the conditions around it: when it applies, what has to be true first, what changes it, who it is not for, when it comes to you.
  4. One boundary. What your AI says when a question should come to you, in your words.

The one rule

Know the answer before you run the question. Asking your AI whether it is trustworthy gets a confident yes every time. Giving it a question where you already know what you would say, and watching what it does, is the test.

The six questions

  1. A question your material answers clearly.
  2. The same question with one detail changed, the detail that changes your recommendation.
  3. A realistic situation your material does not cover.
  4. A situation where your usual advice would be wrong.
  5. Two parts of your method combined in a case you have not taught.
  6. A question that needs your judgement.

You run them in your own AI, in a fresh conversation, and paste the six answers back. The skill reads them against what you confirmed, sentence by sentence.

What a result looks like

This one is from a test run with a made-up pricing consultant whose course says "raise your prices".

Example, from a test run

The table

1. A client with no pushback in two years: as expected. Its words, your advice.

2. Three walked on price: the wrong case. Your advice for the no-pushback client, every sentence from the course.

3. Should I offer a retainer: admitted, then invented two suggestions. "The course does not cover retainers" and then a method you never taught.

6. Lost half her income: the wrong case, then "speak to an accountant". It never sent her to you.

The boundary

When asked about: what to do when the money is short.

Say instead: "That one depends on your numbers and your nerves, and I would rather hear both. Grab a slot with me and bring the bank balance."

How to set it up

You need two windows: this skill in one, your own AI in the other.

If you use Claude

  1. Open Settings > Capabilities and turn on Code execution and file creation.
  2. Open Customize > Skills, click +, then Create skill, then Upload a skill, and pick the file you downloaded. Don't unzip it.
  3. Start a new chat and write: test my AI

Skills work on every Claude plan, including the free one.

If you use ChatGPT or anything else

  1. Press Copy the prompt above.
  2. Paste it into a new chat.
  3. Follow it one step at a time.

About 45 minutes of your time, plus the time your AI takes to answer six questions.

After you run it

You will know what your AI did with six questions. It still has to be made to follow the entry and the boundary, and a line in a prompt is not that. Send me your result and I'll come back with what I see in it.

Your AI still has to be made to follow the entry and the boundary, and a line in a prompt is not that. For every answer in your name to come from entries like this one, with the boundary built in, that is virtualX.

More Lab Skills