Write AI Prompts That Keep You in Control

Use the job-boundary-check pattern

Start with the job: what should the AI produce? Add the boundary: what should it avoid, preserve, or flag? Add the check: how should you evaluate the result? This pattern works for emails, summaries, research, policy drafts, and learning prompts.

Example for a summary

Weak prompt: "Summarize this."

Controlled prompt: "Summarize these notes for a project lead. Use five bullets. Keep action items separate from background. Do not invent owners or dates. If an owner is missing, write 'owner not stated'."

Example for a decision

Instead of asking, "What should we do?" ask: "List three options, the tradeoff for each, the information still missing, and the decision that would be easiest to reverse." This keeps you from accepting a single confident recommendation too quickly.

Use negative instructions carefully

Instructions like "do not be vague" are less useful than specific standards. Say "include dates, owners, and next steps where provided." Say "if the source does not include a number, do not estimate one." Specific boundaries are easier for the model to follow and easier for you to check.

Save reusable prompts

When a prompt works, save the structure, not only the words. A reusable prompt should have placeholders for audience, purpose, source material, output format, and review criteria. This turns prompting into a repeatable work process.

How to apply this guide

Use this guide when your AI results feel vague, generic, too long, or hard to review. Better prompting is not about secret phrases. It is about designing a clearer task with boundaries and a review path.

• Write the task in one sentence before opening an AI tool.
• Decide which parts need human review: job, boundary, check.
• Remove private or unnecessary context before prompting.
• Check whether the final output changes a fact, promise, number, date, or decision.

A prompt should reduce ambiguity without adding private details the model does not need. If the prompt becomes a long dump of context, stop and separate the source, task, and checks.

A safer prompt to try

Use this starter prompt when you want help with the idea in this guide but still want the model to show its limits.