Lesson 1 of 5 · 8 minutes · Module 1: Say what you mean
Why AI answers come back vague
When an AI reply comes back vague or generic, it’s usually because the prompt left gaps and the AI filled them with guesses. In this lesson you’ll send the same job twice, once in one line and once with the details, and see exactly what it guessed.
You need
- A free ChatGPT or Claude account
- About ten minutes
A narrated walkthrough of this lesson is coming. The steps below cover everything in it.
Step 1 of 5
Send a one-line prompt
Start a new chat and send the one-line prompt below. It’s the kind of thing most people type first. Our example is a home bakery replying to a customer who asked about a gluten-free cake.
The prompt · works in ChatGPT, Claude or any AI chat toolWrite a reply to a customer asking about a gluten-free cake.The reply we got on 2026-10-03, unedited. Read our reply.
Screenshot to come
We haven’t captured this step yet. We’d rather show nothing than an old or made-up screen.
CheckpointYou’ve got a reply to a one-line prompt.
Step 2 of 5
See what it guessed
Our one-line prompt got a long reply full of guesses. It said yes, the bakery could make a gluten-free cake, which wasn’t true. It left gaps in square brackets for flavours, sizes and prices, used dollars, and offered two different kitchen notes to choose from. It wasn’t being careless. It just had nothing else to go on.
Screenshot captured 3 October 2026. - Vague: one line, no details
- It guessed: says yes, you can make one
- Gaps to fill in, and prices in dollars
- Specific: who, what to say, length, sign-off
- It said what you asked it to say
- Still check: it promised something on someone else's behalf
CheckpointYou can point to at least three things in your reply that it guessed.
Step 3 of 5
Send a specific prompt
Start a new chat and send the specific version below. It says who we are, what the customer asked, what we want to say, how long it should be and who signs it.
The prompt · works in ChatGPT, Claude or any AI chat toolI run Brook Lane Bakes, a small home bakery. A customer called Priya has emailed to ask if I can make a gluten-free birthday cake for her daughter's party. Write my reply email. - I can't make gluten-free cakes: I use wheat flour in my kitchen every day, so I can't rule out traces - I'd still like to help, so suggest she looks for a specialist gluten-free baker - Keep it warm and short: under 120 words - Sign it from Jo at Brook Lane Bakes - Don't promise anything I haven't said hereThe reply we got on 2026-10-03, unedited. Read our reply.
Screenshot to come
We haven’t captured this step yet. We’d rather show nothing than an old or made-up screen.
CheckpointYou’ve got a reply to the specific prompt.
Step 4 of 5
Compare the two, then check the good one too
The specific prompt got a short, warm reply that said what we wanted. But read it closely, because it also said a specialist baker “will be able to guarantee a safe bake”. We never said that, and we can’t promise it for someone else. A better prompt means fewer guesses, not none.
CheckpointYou’ve read the good reply line by line and found anything you didn’t say.
Step 5 of 5
The rule for the rest of this course
AI fills every gap you leave, so the more you tell it, the less it guesses. Anthropic, which makes Claude, puts it this way: show your prompt to a colleague who knows nothing about the job. If they’d be confused, the AI will be too.
CheckpointYou can explain why a vague prompt gets a vague answer.
Common problems
When you have finished
Want to keep your place? Get a sign-in link by email and we’ll save your progress. You can keep reading either way. Sign in to save your progress.
The screenshot is our own page showing the two prompts and replies word for word, captured by tools/capture/prompting-for-beginners.mjs; it is not the chat screen. The chat screens themselves need a signed-in account, so they have no screenshot yet.