You'll leave with: a repeatable way to improve any answer, a verification checklist, and a place to keep your best prompts.
Your first prompt is a draft
Almost nobody gets the best answer on the first try, including people who write prompts for a living. The skill is not “writing the perfect prompt”. It's noticing what's wrong with the answer and changing one thing.
The 5-step iteration loop
- Run it. Read the answer once, quickly.
- Name the gap. One sentence: “too long”, “too generic”, “wrong tone”, “missed my budget”, “made up a number”.
- Change one thing. Add a missing context fact, tighten the format, add a constraint, or add an example. Not five things at once. Otherwise you won't know what worked.
- Run again in the same chat (to give feedback) or a fresh chat (to test the prompt on its own).
- Save the winner in your prompt log (below).
Useful follow-ups that work in any chat
That's too generic. Use my details: [FACTS]. Give me one concrete example I could use today.
Before you answer, ask me up to 3 questions that would most improve your answer. Then wait.
Critique your own answer. List the 3 weakest points, then give me an improved version that fixes them.
Rewrite that in half the length. Keep the numbers and the next step, cut everything else.
Verify like a skeptic (kindly)
AI writes fluent text whether or not it's right. Fluent is not the same as true. In the HumAIn $40 dinner video, the interesting thing was that the total ChatGPT showed didn't match the sum of its own list. Nothing looked broken until we added the numbers ourselves.
A 60-second verification checklist
- Numbers: add them yourself, or ask the AI to show each step and recalculate.
- Facts and names: find the official page (company, government, original study) and compare.
- Quotes and sources: if it cites a source, open it. Models can invent plausible-looking references.
- Dates and prices: they change. Ask “as of when?” and check the live page.
- Confidence: ask “which parts of your answer are you least sure about?”. It often tells you where to look.
Check your last answer. Recalculate every number step by step, list any difference from what you wrote, and mark each factual claim as confident / unsure / can't verify.
Build your prompt log
A prompt that works is an asset. Keep a simple note (a doc, a notes app, a text file) with one entry per prompt:
| Name | What it's for | Prompt | Last tested | What to watch out for |
|---|---|---|---|---|
| Dinner planner | Weekly meal plan | (paste) | Nov 2026 | Check the total |
When a prompt becomes something you use weekly, promote it to a saved instruction in a Project or skill (Lesson 8). The Prompt Library is a good starting set.
Exercise
Run the loop once, for real.
- Pick a prompt from the Prompt Library or one you use already. Run it.
- Write the gap in one sentence.
- Change one thing and run it again. Did it improve?
- Ask the AI to recalculate or critique itself. Did it find anything?
- Add the winning version to your prompt log, with one “watch out for” note.
Bonus: ask the AI a question where you already know the answer (your own birthday plus 40 days, the total of three receipts). See whether it gets it right. It's a cheap way to calibrate your trust.
Key takeaways
- Prompts are drafts. Change one thing at a time.
- Verify anything involving money, health, law, safety, dates, or quotes.
- Save what works in a prompt log.
Related: Safety prompts: double-check the maths and facts
Sources
Checked November 10, 2026 (Eastern Time). Plans, limits and features change: confirm on the official page.