Does your AI know your work? Test it with ten questions

A good answer does not yet mean partnership

AI can write a good email, produce an accurate summary and offer ten useful ideas while still being a stranger to your work.

I learned this quickly while building my personal AI. One impressive answer does not show whether AI understands my role, respects agreed boundaries or carries a correction into the next result. You see the difference only through repetition: do I have to explain the same background again, can AI recognise a good and bad example, and does yesterday’s correction change tomorrow’s output?

If the answer is mostly no, AI is not useless. It is simply a generic tool, not yet a partner adapted to your work. The question is therefore narrow and practical: does AI know one part of your work well enough for the result to be tested and improved repeatedly?

What does it mean for AI to “know”?

AI does not need to keep all of this in built-in memory. The necessary background may live in project instructions, a knowledge file or a collection of examples. What matters is whether the right information is available during the task, the boundaries are clear and a correction reaches the next result.

The test below helps you assess that. It is not a scientifically validated measure, but a practical checklist for examining one part of your work.

10 questions

Give each question a score of 0, 1 or 2. 0 means AI does not know this or cannot yet do it. 1 means AI knows it only when you explain it again each time. 2 means the necessary knowledge or agreement remains available to AI next time.

1. Does AI know your real role?
Not only your job title, but what you are responsible for, who receives the result and which problem you solve.

2. Does AI know which specific part of the work you are improving together?
“Everything” is not a part of the work. “We prepare questions for client meetings” is.

3. Does AI know the language of this work?
Does it know the vocabulary used by you, your client and your company? Can it avoid expressions that sound correct but do not sound like you?

4. Have you given AI examples of good and bad results?
“Write better” does not tell AI what better means to you. Show it a couple of good results, one bad result and explain the difference.

5. Does AI know the boundaries?
Does AI know which data it must not use, what it must not promise and when it must stop?

6. Does AI know when to clarify before answering?
If AI always answers immediately, it may be fastest precisely where it should first ask a question.

7. Does AI distinguish facts, assumptions and judgements?
Fluent text can make an assumption sound like a fact or a personal preference sound universally true. A good answer shows what we know, what we assume and what someone thinks.

8. Does AI know which decisions remain human?
AI can gather, compare and prepare. Pricing, promises, people decisions and other consequential choices remain human.

9. Can AI handle a real work task?
Give it a task that you have already solved well yourself. You can then compare whether the AI result is genuinely usable or merely sounds good.

10. Does a correction change the next result?
If the same mistake returns, the correction has not reached AI in a lasting way. Every new conversation still begins with an AI that is almost a stranger.

Reading your result

The maximum score is 20.

If you score 0–5 points, you are probably using AI as a general-purpose tool. There is nothing wrong with that — it is simply the beginning. You get help with drafts, summaries and ideas, but must reintroduce much of the necessary background each time.

If you score 6–12 points, you have a partial working method. AI may be very useful, but the result still depends on how much you manage to explain again each time.

If you score 13–17 points, you already have a strong AI work practice. Now examine which parts are persistent and which still live only in your head.

If you score 18–20 points, you have something worth building on. The next question is whether the working method can be improved safely when the work changes.

The score does not measure your intelligence or overall “AI readiness”. It shows only how well one part of your work is currently prepared for working with AI.

Start with the weakest question

The goal is not to score 20. The most useful result may be a single zero. If AI knows your tone but does not ask about missing information, start with question six. If it has plenty of background but unclear boundaries, start with question five.

Then choose one recurring part of the work, such as preparing questions for client meetings, decision support or screening an incoming request. Write down on one page:

1. Result and necessary information. What does AI help produce, and what does it need to know?

2. Good and bad examples. How do you recognise a usable result?

3. Boundaries and the human decision. What must AI not do, and which decision remains human?

4. Test and correction. Which old task will you use for comparison, and where will you record the next correction?

This is far more concrete than simply asking AI to “be my better sales assistant”. Nor do you need to give it your whole life or company history. Provide only the information needed for the chosen part of the work — more data does not make AI infallible or remove human responsibility.

Partnership should show in the next result

To me, an AI Partner is not simply a chatbot with a familiar name and warm tone. Partnership begins when AI understands the agreed part of the work, can show what it does not know, respects boundaries and applies a correction next time.

Correct one visible mistake in that part of the work and check whether the correction remains. If it does, you did not merely receive one good answer. You built the first small part of a working partner.