Fix the prompt
Learn to write prompts that get a usable answer the first time. First watch Infra show one task done well, then change one thing at a time to see what breaks. Then it is your turn.
Read the transcript
You asked the AI a simple question and got a useless answer. What if the problem was never the AI, but the ask?
We think a good prompt is clever wording. It isn't. A good prompt is a complete brief, the kind you'd hand a new colleague.
It has parts: the task, the facts, the shape of the answer, the limits, and what to do when unsure.
Take one out and watch. No facts, and it fills the gap with filler. No limits, and it promises things you never offered.
And the role everyone starts with? It matters least. Spend your words on the facts and the limits instead.
A better answer starts with a better brief. Practise on five real tasks in our free lab. Which part do you always forget?
Checking your account…
This lab and its certificate are free. Full certification programs, the Certified AI Practitioner and role masterclasses, are coming. Tell us which you want first.
About this lab
Most weak answers from an AI tool start with a weak prompt. A good prompt has up to 6 parts, and the lab shows what each one does by taking it out. You then rewrite 5 real-looking tasks and see which parts your prompts contain.
- A clear task. A model does what it is asked, so a vague verb gets a vague answer.
- The facts it needs. It cannot know your numbers, your audience or what breaks. Without them it fills the gap with plausible filler.
- The shape of the answer. Without a format you get whatever length and layout the model likes, which is rarely what you can paste and use.
- Limits and exclusions. Saying what to avoid removes the habits you would otherwise edit out by hand.
- What to do when unsure. A model that is not told it may say "I do not know" will guess, and a guess reads the same as a fact.
- A role. A role sets tone. It helps least of the six, so add it after the others, never instead of them.
Questions
What is the Fix the prompt lab?
It is a practice exercise in writing prompts. You watch one task answered well and then answered badly as parts of the prompt are removed, and then you rewrite 5 weak prompts and see which of 6 parts each of yours contains.
What are the six parts?
A clear task, the facts the model needs, the shape of the answer, limits and exclusions, what to do when unsure, and a role. The lab shows what the answer does when each one is missing, and why the role matters least.
Does an AI mark my prompt?
No. A fixed checklist on our server looks for each part. It tells you whether the parts are there, not whether your wording is wise, so run your prompt in the tool you use and read the answer as well. No AI service is called and nothing you type is saved.
Are the sample answers real model output?
No. We wrote them to show the usual effect of each missing part. A real model will vary, which is why the lesson is about the parts, not the exact words.
Do I need an account?
Yes, a free one. The account keeps your progress: when a prompt is strong we record which task and the day, never your prompt. When you are ready, the certified round on this page gives a certificate anyone can check.