Journal Jul 15, 2026 · 5 min

A short field guide to writing better AI prompts

Five habits that reliably get better answers from ChatGPT, Claude, and Gemini — and what a rough-to-refined rewrite actually looks like.


Models don't read minds; they read prompts. Most disappointing AI answers trace back to a prompt that never said what a good answer would look like. The fix is rarely more words — it's better-organized ones. Five habits cover most of it.

1. Lead with the task, not the backstory

State what you want in the first sentence, then give context. Models weigh the beginning of a prompt heavily, and so do you when you re-read it. “Summarize this contract clause for a non-lawyer” beats three sentences of wind-up that end in “…can you help?”

2. Give the context that changes the answer

Not all context is equal. Who is this for? What do they already know? What’s the constraint that makes this hard? One line of the right context outperforms a paragraph of the wrong kind. If you're asking for code, name the language and version. If you're asking for writing, name the audience and tone.

3. Specify the output shape

Length, format, structure: “three bullet points,” “a table with columns X and Y,” “one paragraph, no hedging.” This is the single highest-leverage line in most prompts, and the one most often missing from rough drafts.

4. One task per prompt

“Summarize this, then rewrite it for LinkedIn, and also list the risks” produces three mediocre outputs. Ask for them one at a time — or, if they must travel together, number them explicitly so the model treats each as a deliverable.

5. Say what to avoid

Negative constraints are cheap and effective: “no marketing language,” “don't explain the basics,” “plain text only.” They prune the failure modes you've already seen.

What a rewrite looks like

A real rough draft:

hey so i have this email i need to send to my team about the deadline moving and people are gonna be annoyed, can you write something, it should be nice but firm i guess, also mention the client stuff

The same intent, refined:

Write a short, professional email to my team announcing that the project deadline has moved earlier. Acknowledge the frustration, explain the client-driven reason, and end with clear next steps. Tone: warm but firm. Length: under 150 words.

Same information, same author — but the second one states the task first, names the audience and tone, shapes the output, and stays a single ask. That's the entire craft.

The lazy version of all five habits

This restructuring is mechanical, which is exactly why it's automatable. RefineIQ applies these habits to your draft in one step: select your rough prompt, press Alt+Shift+R, and review the rewrite before you send it. It preserves your intent, never invents requirements you didn't state, and runs on your device by default — your half-formed thoughts stay yours. Write messy, send clean.

Free. Private. On-device.

Add to Chrome