For maths, logic and multi-part decisions, asking the model to work step by step improves accuracy, and lets you see where it went wrong.
Ask for the steps
“A laptop costs ₹42,000 + 18% GST with a 5% discount on the base price. What’s the final price? Show each step, then the final answer.” Now you can check: 42,000 × 0.95 = 39,900; GST 7,182; total ₹47,082.
Many newer models reason internally anyway, but showing the steps still helps you verify them.
Break big tasks into stages
“Give me an outline for a 4-week Excel training plan.” Review it.
“Now write week 1 in detail.” And so on.
Smaller steps you approve along the way beat one giant prompt.
Ask it to check
“Check your answer for errors and list any assumptions you made.” Models often catch their own slips, and the assumptions list shows what you didn’t specify.
⚠️ Always re-check numbers that matter (tax, invoices, salaries) with Excel or a calculator. Language models can make arithmetic mistakes with complete confidence.
💡 For data work, ask for an Excel formula instead of the answer. The formula is checkable and reusable; the number isn’t.
Practice
Do the three experiments and verify the arithmetic yourself.
📎 Practice files for this article
📄
Lesson 4 prompt worksheetCopy-paste prompts and exercises to try in ChatGPT, Copilot, Gemini or Claude.
