2.3
Prompting That Lands First Time
Why this exists
Two lessons of failure modes — it guesses, and it decays. This one is the lever you actually control.
You said the thing you most wanted was learning to talk to it properly. Good instinct — it's the highest-leverage skill in the course, and it isn't what most people think.
There are no magic words. No phrase unlocks a better model. It's something more boring and far more useful: getting the first message right, because everything after it is you arguing with a decision that's already been made.
The idea
Your first prompt is worth more than every prompt after it combined.
Here's why, and it falls straight out of the last lesson. The model doesn't restart when you correct it — it re-reads everything. Your correction gets added to the pile, and it continues from all of it, including the wrong direction it already committed to. So when you say "no, I meant X", you're not resetting. You're asking it to walk back down a road it's already halfway along, while still staring at the road.
You've felt this. It's the conversation where every reply gets slightly worse and you end up with something clumsier than if you'd started over. That's not the model being stupid. That's four corrections stacked on one bad opening.
Four things make an opening prompt land:
| Part | What it does | Missing it looks like |
|---|---|---|
| Context | Who you are, what this is for | Generic output that fits nobody |
| Task | The specific thing you want done | It does something adjacent |
| Constraints | Length, tone, what to avoid, what to use | Three paragraphs when you wanted three bullets |
| Output shape | What the finished thing looks like | Right content, unusable format |
Most weak prompts have the task and nothing else. "Write me an email about the invoice." Compare: "I'm following up with a client who's two weeks late on a £4k invoice. We want to keep the relationship. Short email, five sentences max, friendly but clear there's a deadline. Give me two versions." Same amount of thinking, thirty seconds more typing, and you'll use the first output instead of the fourth.
The recovery move. When you do end up five prompts deep — and you will — don't push on. Say this:
"Write me the prompt I should have opened with to get this in one go."
Take what it gives you, open a fresh session, paste it in. You'll get a better answer than the one you were grinding towards, and you'll have a reusable prompt out of it. This works because it's genuinely good at describing what it needed — it just can't unsee where it's already been.
One more thing, and it's the one people skip: if you can't say what good looks like, it can't either. Half of all bad outputs are actually unclear requests. Working out what you want before you type is not preparation for the skill. It is the skill.
Do
- Pick a real task from your week. Something you'd genuinely use — an email, a summary, a plan, a comparison.
- Write the lazy version. One line, the way you'd normally type it. Send it. Save the answer.
- Now write the full version in a fresh session — context, task, constraints, output shape. Send it. Save that answer.
- Put them side by side. Which one would you actually use?
- Go back to the lazy conversation and correct it three times, properly, the way you would if you were in a hurry. Note whether answer four is better or worse than answer two.
- In that same messy conversation, run the recovery move: "Write me the prompt I should have opened with to get this in one go." Save what it gives you.
- Turn all of it into
prompt-template.md— your four headings, with your own one-line note under each about what goes there.
Step 5 is the one people skip, and it's the one that changes behaviour. Reading "corrections make it worse" does nothing. Watching answer four be worse than answer two rewires you permanently.
Done when
Never do
Don't push past three corrections. If you've corrected it three times and it's still not right, you're not close — you're stuck to a bad opening and every further message makes it heavier.
Stop. Run the recovery move, open a fresh session, start clean. It feels like giving up and losing your progress. It's the opposite: it's the fastest route to the thing you wanted, and the version you get will be better than the one you were fighting for.