How to write better AI prompts: a plain-English guide for writers

A good AI prompt isn't a magic phrase. It's just a clear brief. The single biggest upgrade you can make is to stop typing "write me a LinkedIn post about our launch" and start typing "You're our head of marketing. Write a 120-word LinkedIn post announcing our launch to existing customers. Warm, specific, no hype. Here are the three things that changed: …" Same model, wildly different output — because the second one is a brief and the first one is a wish.

I build language models for a living, and I talk to a lot of writers and social-media managers who feel like they're "bad at prompting." Almost none of them are. They were just never told that a prompt is a brief, and that briefs have a shape. Here's the shape, in plain English, with examples you can steal.

Why most prompts produce mush

When you give an AI a vague instruction, it has to guess at everything you left out — audience, length, tone, format, purpose — and it guesses with the blandest, most average option, because that's the safest bet across the millions of texts it learned from.

"Write a product description." For whom? How long? In what voice? Selling what, exactly? The model fills every gap with the median choice, and the median is mush. You didn't get a bad model. You gave it a bad brief.

Good news: you already know how to write a brief. You do it for freelancers, designers, and interns all the time. You just have to say the quiet parts out loud.

The anatomy of a prompt that works

Most strong prompts have four parts. You don't need all four every time, but the more you include, the less the model has to guess.

Role — who it should be. "You're a B2B copywriter who writes like a calm human."
Task — what to produce, with length and format. "Write three subject-line options, max 6 words each."
Context — the raw material only you have. "The webinar is about cutting cloud costs; audience is finance leads who are skeptical of hype."
Constraints — the guardrails. "No exclamation marks. No 'unlock' or 'supercharge.' Plain words."

Put together: "You're a B2B copywriter who writes like a calm human. Write three subject-line options, max 6 words each, for a webinar about cutting cloud costs. Audience: finance leads who distrust hype. No exclamation marks, no buzzwords."

That prompt will beat 95% of what people type, and it took fifteen seconds.

Six prompt patterns worth memorizing

These are the patterns I see produce genuinely useful output for writers and marketers — no jargon, no "act as a world-class 10x ninja."

1. Give it a role and a reader

The fastest quality jump. "You're a [role] writing for [specific reader]." A model writes very differently for "skeptical CFOs" than for "first-time founders," and it will never guess the reader correctly on its own.

2. Show, don't just tell (one example beats ten adjectives)

Instead of describing your tone with adjectives, paste a paragraph you've already written and say "match this voice." This is the highest-leverage move in the whole craft. One real example teaches the model more about your style than a page of "make it punchy but professional but warm." The technical name is few-shot prompting; the plain name is "show it what good looks like."

3. Pin down format and length

Models are wordy by default. Tell it the shape: "exactly 5 bullets," "under 280 characters," "a 2-sentence answer, then a list." You'll cut your editing time in half.

4. Feed it your raw material

The best AI writing isn't invented — it's rearranged from what you already know. Dump your messy notes, the meeting transcript, the three features that actually matter, and ask it to shape them. A prompt with your real specifics will always beat a prompt that makes the model invent generic ones.

5. Ask for options, then choose

Don't ask for "the post." Ask for "five different angles, one line each." Pick the one that sparks something, then expand only that. You stay the editor; the model does the brainstorming grunt work.

6. Iterate in small steps

Treat it like editing a colleague's draft. "Tighter." "Less salesy." "Keep the second one but make it sound less like a brochure." Short, specific follow-ups outperform one giant perfect prompt almost every time.

The part nobody tells you: prompting has a typing tax

Here's the thing I noticed once I started timing my own day. The patterns above work — but they're long. You're constantly retyping the same role, the same brand context, the same constraints, the same "match this voice" preamble. The actual thinking takes ten seconds; the typing takes two minutes, over and over, all day.

That typing tax is invisible until you measure it, and then you can't unsee it.

This is, honestly, part of why I built TypeTab the way I did. It's a Mac app that finishes your sentences inline — in any text field, including the chat box where you write your prompts. When you start typing a prompt you've written a hundred variations of, the rest of it is already there in grey; press Tab for the next word or Ctrl+E for the whole line. Your standard role line, your brand guardrails, your "audience: skeptical finance leads" — they come back fast because the tool has seen how you phrase them.

And there's a detail I'm a little proud of: when we trained TypeTab's base model, we deliberately fed it a slice of how real people actually prompt AI tools — real prompt prose from open datasets — alongside emails, posts, and long-form writing. So it understands the register of a prompt, not just an essay. It finishes "Write a 120-word post for…" as naturally as it finishes "Hi Sarah, thanks for…".

The best prompt is sometimes no prompt at all

Now the contrarian bit, because it's true and it matters for how you spend your day.

A huge amount of what people open a chatbot for doesn't need a chatbot. The quick reply, the caption, the comment, the "thanks, that works for me — let's go with Thursday" — for all of that, prompting is overhead. You leave your app, frame a request, wait, read, copy, paste, fix the formatting. You spent a paragraph of effort to avoid typing a sentence.

For that whole category, inline completion just wins. There's no prompt, no chat, no context switch — you type your sentence and the rest appears where your cursor already is. I think the smart 2026 setup, especially for writers and social managers, is two tools doing two jobs:

A chatbot for the heavy lifts — research, long drafts, restructuring, brainstorming. Worth a good prompt.
Inline autocomplete for the constant stream of everyday writing — emails, DMs, captions, comments, and yes, drafting the prompts themselves. No prompt needed.

You stop paying the prompting tax on the 90% of writing that never deserved a prompt in the first place.

Questions people ask

Do I need to learn "prompt engineering" as a writer? You need to learn to write a clear brief — role, task, context, constraints. That's 90% of it. The fancy term is optional; the clarity isn't.

What's the single most effective prompt tip? Show an example of the voice you want instead of describing it. Pasting one paragraph you've written ("match this tone") beats any list of adjectives.

Is it cheating to use AI for my writing? No more than using a calculator is cheating at math. The judgment — what to say, what to cut, what sounds like you — is still yours. The tool just removes the friction. The writers who win are the ones who stay the editor.

How does TypeTab fit in if it doesn't use prompts? Two ways: it makes typing your chatbot prompts much faster (its model learned from real prompt prose), and it removes the need to prompt at all for everyday writing by completing your sentences inline. More on that approach in How to write faster without sounding like AI.

The takeaway

Better prompts aren't about secret phrases. They're about giving the model the brief you'd give a smart freelancer: who it is, what you want, what you know, and what to avoid. Master that and you'll get good output on the first try far more often.

Then notice how much of your writing never needed a prompt at all — and let an inline tool handle that part so you can save your good prompts for the work that's actually worth one.

Try TypeTab free — your first 100 word completions and 20 full lines are on us, no account. It runs entirely on your Mac, so even your prompts never leave it.