The Beginner's Guide to Writing Better AI Prompts

Why most first prompts disappoint

A prompt like “write me a marketing email” gives the model almost nothing to work with, so it fills in the gaps with generic assumptions — a generic audience, a generic tone, a generic length. The output isn’t wrong, exactly. It’s just the statistical average of every marketing email it’s ever seen, which is rarely what you actually needed.

The fix isn’t a magic phrase or a secret trick. It’s giving the model the same context a competent human freelancer would need before they could do the task well. If you handed that one-line brief to a real copywriter, they’d ask you a dozen clarifying questions before writing a word. A good prompt answers those questions upfront.

Six habits that work

1. State the role and audience. “You’re writing for busy small-business owners with no technical background” changes the output more than almost any other single instruction, because it tells the model who it’s writing for, not just what to write.

2. Give it an example. Paste one example of the tone, format, or style you want. Models are very good at matching a pattern you show them — often better than they are at interpreting a written description of that same pattern. If you have an old email you liked the tone of, paste it in and say “match this tone.”

3. Ask for a specific structure. “Give me three options, each under 50 words” is easier for the model to satisfy — and easier for you to judge — than an open-ended request. Constraints aren’t limiting; they’re what narrows an infinite space of possible answers down to something you can actually evaluate.

4. Iterate instead of restarting. If the first output is 80% right, say exactly what to change rather than rewriting the whole prompt from scratch. “Keep the structure, but make the second paragraph shorter and cut the exclamation points” gets you there in one step. Restarting from zero throws away the 80% that was already working.

5. Tell it what to avoid. “Don’t use corporate jargon like ‘synergy’ or ‘leverage’” is often more effective than describing the tone you do want in the abstract — it’s much easier for a model (and a person) to avoid a concrete thing than to hit a vague target.

6. Separate instructions from the content you’re working on. When you’re asking the model to edit, summarize, or analyze something, make it visually obvious where your instructions end and the source material begins — for example, put the material after a line that says “Here’s the text:” or wrap it in quotes. This matters more than people expect; without a clear boundary, models sometimes get confused about which part is the instruction and which part is the thing to act on.

A full before/after example

Before: “Write a tweet about our new app.”

After: “Write a tweet announcing our new budgeting app for college students. Playful tone, no emoji, under 200 characters, and end with a question to drive replies. Here’s an example of the voice we use on social: ‘saving money is a skill, not a personality trait. we built the app that teaches the skill.’”

The second version gives the model a role and audience (college students), constraints (length, no emoji), a goal (drive replies), and a concrete voice sample to match — exactly what narrows a near-infinite space of possible tweets down to something usable on the first try.

Fixing a bad answer without starting over

This is the habit that saves the most time in practice, so it’s worth a dedicated example. Say you asked for a product description and got something too long and too formal.

Don’t write a brand-new prompt. Just reply:

“Cut this to 2 sentences and make the tone more casual — like you’re texting a friend, not writing ad copy.”

The model still has the full context of what it wrote and why; you’re editing, not re-briefing. This back-and-forth is usually faster than crafting one “perfect” prompt upfront — treat the first response as a draft, not a final answer.

A quick pre-send checklist

Before sending a prompt for anything beyond a trivial question, glance down this list:

Have I said who this is for?
Have I said what format or length I want back?
Would an example of the style I want help more than another sentence of description?
Am I about to restart from scratch when I could just ask for a targeted edit instead?

Next step

Once prompting feels natural, the next skill worth building is picking the right tool for a given job — different tools are better at different things, and knowing which is which saves you from fighting a tool that was never going to be great at your task. Our AI Tools Directory is organized by category to help with exactly that.