How well do you use AI?
Twelve short situations from everyday use of AI, each with one best answer. You get a score for each of four skills and the one page worth reading next for any skill that is weak. No email and no sign-up. Your answers stay in this browser.
New to all of this? Start with your first 20 minutes, then come back.
New to this? What it does, in plain words
It is a short quiz about judgment, not about how AI works inside.
Each question describes something that happens when people use ChatGPT, Claude, Gemini or Copilot: a vague request, an answer with a statistic in it, a document with a customer’s name, a teacher’s rule. You pick what you would do. It takes about five minutes.
What you need
Nothing. No API key and no account. Your answers are scored in this browser.
How to use it
- Press Start and answer each question. After each one you see the best answer and why.
- At the end you get a score for each skill.
- Follow the link for your weakest skill. Each one is a short read or a free tool.
- Share your result if you like: the link carries the four scores and nothing else.
Four skills, three questions each
- AskingGetting a useful answer the first time, and steering it when it is not.
- CheckingKnowing which parts of an answer to verify, and how.
- ProtectingKeeping private and confidential details out of the chat.
- DecidingKnowing when AI is the wrong tool, and being open about using it.
Your answers stay in this browser. A shared link carries only the four scores, after the # sign, which browsers never send to a server.
Why these four skills
Most AI advice is about writing better prompts. That is one skill of four, and the one people already practice most.
The skill people skip
Anthropic studied 9,830 conversations on Claude.ai from a week in January 2026, each with several back-and-forths. In its AI Fluency Index, published February 23, 2026, 85.7% of those conversations showed people refining what they asked, while 8.7% showed them checking the facts and claims that mattered. Asking is practiced. Checking mostly is not, so this quiz gives it equal weight.
What each skill covers
- Asking. Saying who an answer is for and how long it should be, and steering the second message instead of starting over.
- Checking. Spotting the parts of an answer that are most often wrong (figures, quotes, references, anything recent) and confirming them at a source that would know.
- Protecting. Leaving out names, account numbers and confidential material the AI does not need, and knowing that text an assistant reads can carry instructions of its own.
- Deciding. Keeping decisions about people with people. Following the course’s or employer’s rule, and saying plainly how you used AI.
What the score means
It is a quick self-check, not a certificate, and twelve questions cannot measure anyone’s ability. Each question has one answer that is better than the others for most situations, and the explanation says why. Where your own course, employer or client has a rule, that rule wins. The questions do not test any one product: the same habits work in ChatGPT, Claude, Gemini and Copilot.
Where to go next
For asking, the free prompt course takes a lesson at a time. For checking, the Fact Check List and Edit Check do the tedious part for you. For protecting, Safe Paste swaps names and numbers for placeholders. For deciding, the AI Use Statement writes a clear note of how you used AI.
Questions
Is this AI skills test free?
Yes, with no sign-up or email. The questions are scored in your browser and nothing you answer is sent to thinkhard.si. Take it as often as you like.
Who is it for?
Anyone who uses ChatGPT, Claude, Gemini or Copilot, or is about to: students, people at work, job seekers, teachers and people using AI for everyday questions. The situations are deliberately ordinary, and none needs technical knowledge.
Can I use it with a team or a class?
Yes. Ask everyone to take it and share their result link. The link carries only the four scores, so it shows where a group is weak without showing anyone’s answers. The weakest skill is the place to start a session.
Why is there one right answer when real life is messier?
Each question is written so that one answer holds in most situations. Real rules differ, and the explanations say where a course, employer or client rule would change the answer.
Does it test knowledge of how AI works?
No. It tests what you do with an answer, not what you know about models. For the vocabulary, the glossary explains each AI term in plain English.