Prompting that actually works: a practical field guide

You don't need 'prompt engineering.' You need five habits. Every one of these is the difference between a generic answer and a useful one.

1. Give it a role and a goal

Tell the AI who it is and what you're trying to accomplish. 'You're an estimator reviewing a spec for fence scope. Your goal is to find every mention of fencing, gates, and access control.' The role focuses it; the goal tells it what 'good' means.

2. Give it the context, not just the ask

AI can't read your mind or your files unless you show it. Paste the email, the notes, the relevant spec section. The more real context, the more real the answer. Vague in, vague out.

3. Show one example of what you want

One good example beats a paragraph of instructions. If you want scope summaries in a certain format, paste a sample. The AI will match the pattern almost every time.

4. Specify the format

Ask for exactly the shape you need: 'a 3-bullet summary,' 'a table with columns X, Y, Z,' 'under 90 words,' 'just the answer, no preamble.' You'll stop editing its output and start using it.

5. Iterate — don't restart

When the answer is close but off, don't rewrite the whole prompt. Tell it what to change: 'shorter,' 'more direct,' 'add the bid date,' 'drop the sales tone.' It's a conversation, not a slot machine.

Putting it together

"You're a commercial fence estimator. Read the spec section below and list every fence/gate/access-control requirement as bullets, each with the material, height, and any standard cited. If something's unclear, flag it. Here's the section: [paste]."

Role, goal, context, format, and a built-in way to flag uncertainty — all in one prompt. That's a prompt that does real work. Save the ones that land; you're building a toolkit, not typing from scratch every time.