Skip to content
FREE ACCESS
Verified 30 Sep 2026
Silent Overseer
Guide

7 Prompting Techniques That Actually Work

Published 30 Sep 2026 Updated 30 Sep 2026 Last verified 30 Sep 2026 7 min read
The short version

You don’t need “magic words”. Small structural habits, like giving an example or asking for steps, reliably improve results in ChatGPT, Claude and Grok. Start with our basic prompt guide, then add these.

1. Give it a role

“Act as an experienced small-business accountant explaining to a beginner.” A role sets vocabulary, depth and tone. It doesn’t make the AI a real expert, so still check facts.

2. Show an example

Paste one sample of what “good” looks like: “Write three more in this style: [example].” Examples are often more powerful than descriptions.

3. Ask for steps

“Think through this step by step, then give your final answer.” Helpful for maths, logic, plans and decisions.

4. Control the format

“Answer as a table with columns X, Y, Z”, “as five bullets”, “in under 100 words”. Say it upfront.

5. Set boundaries

“Use only the text below.” “Don’t invent facts; say ‘not stated’.” Boundaries reduce guessing.

6. Let it ask you questions

End with “Ask me any questions you need answered first.” It often produces a much better first draft.

7. Iterate and critique

Ask for a first version, then: “What’s weak about this? Improve it.” Two rounds beat one long prompt.

All seven in one prompt
Act as [ROLE]. Task: [TASK]. Here's an example of the style I want: [EXAMPLE]. Think step by step, then give the final answer as [FORMAT]. Use only the information I've provided and say "not stated" if something is missing. Ask me any questions you need first. After your draft, list the 3 weakest points and fix them.

Common mistakes

  • Stacking every technique into every prompt. Use what the task needs.
  • Not testing. Try the same prompt in two tools and compare.

What to try next

Use the free Prompt Builder or browse the prompt vault.

Sources

Silent Overseer uses AI tools to help research and draft, and every article is reviewed before publishing. How we work. Spotted a mistake? We correct and date changes.

The Overseer Brief

Important AI developments, practical tools and useful freebies without the noise.

One useful email at a time, unsubscribe any time. We store your email to send the Brief; see the privacy policy.