Stop Prompting Like It’s 2023

Get Better AI Answers in Fewer Attempts

A practical guide to prompt engineering for ChatGPT, Claude and Gemini

Almost everyone believes a disappointing answer is a wording problem. The wrong phrasing, the wrong opening line, the wrong role assigned. So they reword and fire again. And again.

Chapter 4 takes a prompt that does everything the guides ask for — a task, a role, a format, a constraint on tone — and shows it producing something close to worthless. Not because the sentence was weak. Because the model had never seen the document the question was actually about, and answered anyway, in the same confident voice it uses when it knows. No amount of rewriting reaches that. Prompting turns out to be mostly a matter of supplying context and only a little a matter of phrasing, which is the reversal the whole book is built on.

Chapter 2 handles the other one: the assumption that something in the process checks a fact. Nothing does. That single idea explains hallucination without requiring anyone to understand how the machinery works.

The book also refuses the theatre. No secret formulas, no megaprompts, no promise that a role-play opening unlocks a hidden model. Where the popular techniques hold up, it says so. Where they don’t, it says that too.

And it counts the cost. Most prompting advice was written by people who never worry about what an attempt costs; regenerate it, they say, try another phrasing. That’s free advice on an unmetered connection and an expensive habit on a bundle that has to last the week. Aiming for a usable answer by the second attempt rather than the fifth changes what’s worth teaching: front-load the context, correct rather than regenerate, compose offline and paste once.

What’s in it

• Nine chapters, written to be read in order or raided for the one thing you need
• Twelve templates you can copy and adapt, each one for a job rather than a profession: drafting, rewriting, compressing, explaining, planning, checking
• Three appendices, including the full template library in one place
• Worked Nigerian examples throughout, in Naira, on the connections people actually have
• No named model versions, so it doesn’t go stale the week a new one ships

If you want a sample of the argument before buying, the article Why good prompts still give bad answers is drawn from Chapter 4.

There’s also The Expanded Template Pack, which takes all forty templates further with worked examples.