Chapter 2 · Part 1

Say what you actually want

Here's a mental model that fixes most bad prompts on its own: treat the AI like a brilliant intern on their first day. Enormously capable, endlessly patient — and knowing nothing about your company, your reader, or what "make it good" means to you. You wouldn't tell that intern "write a post about the launch" and walk away. You'd brief them.

Chapter 1 showed why: every detail you don't specify, the model fills with the most average choice. Specificity is how you take those choices back.

Scroll to brief the intern, one line at a time.

'Write a post about our launch.' The model has to guess the reader, the tone, the length, the shape…

scroll

The briefing checklist

You don't need all of these every time — but each one you include is a guess the model no longer has to make:

  • Goal — what is this for? ("convince CTOs to book a demo", not "write a post")
  • Audience — who reads it, and what do they already know?
  • Constraints — length, tone, things to avoid, things that must appear.
  • Format — bullets? a table? an email? code with comments? JSON?
  • Context — the background facts the model can't possibly know: your product, the previous email in the thread, the code around the function.

That last one matters most. The model has read most of the internet, but it has read nothing about your situation. Paste in the relevant context — generously.

the same request, before and after a briefing
BEFORE:
Write a post about our product launch.

AFTER:
Write a LinkedIn post announcing our launch.
Audience: busy CTOs deciding on deployment tools.
Goal: get them to book a demo.
Tone: direct, concrete, zero hype. Max 120 words.
Format: one hook line, 3 bullet benefits, one CTA.
Context: we're DeployKit — one-command rollbacks,
SOC 2 certified, cuts deploy time ~40% in pilots.

"But I shouldn't have to spell it out"

Sure — and a great senior colleague eventually learns your taste. But within a single conversation, the model only knows what's in the context window. The five extra lines of briefing cost you thirty seconds and routinely save five rounds of "no, not like that."

Sometimes, though, describing what you want in words is genuinely hard — tone and format especially. For those, there's a better tool than description: examples. That's next.