Chapter 04 · ~15 min · outline

Tips & tricks.

The pro moves, in plain English. Steal what's useful.

Outline only: full lesson coming next. Each tip below will get a worked example and a "try it" button in v2.

The tips

Tip 1 · Give it a role

"Act as a strict editor." "Be a patient biology teacher." "You're a skeptical interviewer." Roles are the single highest-leverage prompting move.

Tip 2 · Show, don't tell

Want output in a specific style? Paste in 2-3 examples of what "good" looks like. The model will match the pattern.

Tip 3 · Constrain the shape

"In 100 words." "As 3 bullets." "Step-by-step." Constraints prevent default mush.

Tip 4 · Iterate, don't restart

If the first answer is 80% right, refine in the same chat. Restarting from scratch loses the context the model has already built up.

Tip 5 · Tell it what NOT to do

"Don't use the word 'unique.'" "Don't apologise." "Don't list pros and cons: just give your opinion." Negative constraints are powerful.

Tip 6 · Long projects? Ask for a summary, then continue

For multi-day or multi-session work, periodically ask: "Summarise everything we've decided so far in 10 bullets, so I can paste it into a fresh chat." That's how you maintain continuity.

Tip 7 · Pick the right model for the job

  • Reasoning-heavy tasks (math, hard logic, code review): use the slower "thinking" models.
  • Quick chat / casual: use the faster, cheaper models.
  • Creative writing: most frontier models work well; try a few and see which voice you like.

Tip 8 · Save your best prompts

If a prompt works really well, save it. Build a personal library. The best prompts are templates you reuse across hundreds of conversations.

Tip 9 · Ask for the underlying principle

When you get a good answer, follow up with "What's the underlying principle? When does this not apply?" That turns a one-time answer into a permanent piece of understanding.

Tip 10 · Treat the AI like a smart friend, not an oracle

Sometimes it's right, sometimes it's bluffing. Verify the things that matter.