How to Write Prompts That Get Usable Results From AI

Most weak AI results come from weak instructions or missing information. A good prompt gives the AI what it needs and asks it to show its work. These habits come straight from the course lessons.

Give it the right information

An AI model answers based on the context you give it in the conversation. Include your current goal, the files that matter, the exact error message, the steps to reproduce the problem, and what you have already tried. Keep your lasting project rules short, and save decisions in your project files instead of relying on the chat alone.

A weak prompt and a better one

Weak: "Make my idea into a complete platform with every useful feature."

Better: "Help me narrow this idea to one user and one painful workflow. Separate known evidence from assumptions. State the current workaround, first useful result, non-goals, and an observable success criterion. Ask only about missing decisions that affect this scope. Suggest one inexpensive next test and a finding that would change the plan; do not invent customer demand."

Why the better one works

It names one task, separates what you know from what you are guessing, says what to leave out, asks for something you can check, and tells the AI not to make things up.

Ask it to show its checks

Don't accept a confident answer by itself. Ask for the files it changed, the checks it ran, a screenshot, or its reasoning. Review what was actually changed rather than how sure the explanation sounds.

Keep requests small

One agent and one small task is a good default. Say what result you want and what is off limits, agree on checks you can see, and review the differences before you accept a change.

When a long conversation drifts

Long conversations can lose details, and a summary can turn an untested theory into a "confirmed" cause. Start a fresh conversation with a short handoff: the goal, what is confirmed, what failed, what is unknown, and the next step. Don't paste your whole project or any secrets.

Go deeper with the free lessons

• LLMs, Tokens, and Context Windows
• Ideation and Product Shape
• Verification-First Development
• Agent Workflow and Human Review
• Best Starting Practices

Start the Foundation. Last updated 2026-10-04.