Skip to content
Writing

AI Prompts for Writing: Articles, Essays and Copy

Most AI writing reads flat because the prompt was flat. Here is how to brief AI for articles, essays, emails and fiction — with prompts you can reuse today.

Illustration of AI-assisted writing turning notes into polished text

Ask an AI to "write something about productivity" and you get exactly what you asked for: something. Vague in, vague out. The sentences are grammatical, the paragraphs balanced, none of it sounds like a person who's ever missed a deadline. That gap — between text that's technically fine and writing anyone would want to read — is what most people are actually fighting when they say AI writing doesn't work.

Good AI prompts for writing close that gap the way a solid brief does for a human writer. You decide the angle, name the reader, set the voice, and cap the length before a word gets generated. Do that and the model stops averaging. Here's how it plays out across the writing you actually produce.

Why AI writing comes out generic

A model fed a thin instruction guesses at everything you left out, and it guesses toward the middle of everything it has read. That middle is the bland, faintly corporate voice everyone recognizes now.

Four decisions steer it off the average:

  • Angle — the specific take, not the topic. "Remote work is lonely and companies pretend otherwise," not "remote work."
  • Audience — who reads this and what they already know. A prompt written for skeptical CFOs produces different sentences than one for new grads.
  • Voice — plain and dry, warm and personal, sharp and opinionated. Name it, then anchor it to something concrete.
  • Length — a word count changes the structure. "Write about X" sprawls; "220 words" forces choices.

Miss these and you're editing mush; get them in and the first draft is worth editing. The pattern underneath — role, context, constraints — is covered in the fundamentals of writing prompts, and it holds for every format below.

Articles and blog posts: outline, then draft, then edit

The single worst prompt for a blog post is "write a 1,000-word blog post about X." You get a wall of text where everything's wrong at once. Split the job into three passes and each one becomes steerable.

Start with structure only:

You're an editor planning a blog post. TOPIC: why small teams should write fewer meeting notes, not more. READER: startup founders who feel buried in documentation. ANGLE: most notes are written for no one and read by no one. Give me an H2 outline only — 6 sections. For each: the one point it makes and the single example or number that would prove it. No prose yet.

Argue with the outline, cut the weak section, then draft one section at a time with a word budget attached. Working in chunks keeps each part on topic and lets you redo a flat section on its own. The marketing version of this workflow lives in prompts for marketing writing; the outline-first habit carries straight over.

Essays and academic-style writing, honestly

AI writing prompts can genuinely help you think through an essay — surfacing counterarguments, testing whether your thesis holds, tightening a paragraph that circles. What they should not do is write the essay you'll submit with your name on it.

Most schools and many employers treat AI-authored work as plagiarism or misconduct, and the policies are getting sharper, not looser. Using AI as an aid — sparring partner, research map, editor — is defensible. Passing off its prose as your own thinking is what gets people in trouble, and it quietly means you never learn to write.

Prompt it as a thinking tool:

Act as a demanding essay tutor. My thesis: "public libraries should be funded as digital infrastructure, not just book repositories." Don't write the essay for me. Instead: - Name the 3 strongest objections a skeptic would raise. - Point out where my thesis is doing too much and should narrow. - List 5 questions I need to answer before this argument holds.

That gets you a better essay you actually wrote. If a prompt keeps returning shallow pushback, the instruction is underspecified — a prompt optimizer usually finds what you left vague.

Emails and messages

Email is where AI prompts for writing pay off fastest — low stakes per message, high volume. The trick is handing the model the situation, not just the topic.

Write a short email. CONTEXT: I'm declining a vendor's proposal. We liked the demo but the price is 3x our budget with no path down. GOAL: close the door clearly, keep the relationship warm for later. TONE: friendly, direct, no corporate padding. LENGTH: under 90 words. No "I hope this finds you well."

Name the tone and ban the filler, and you get something you can send after a one-line edit. Keep the prompts that work — a "warm no," an overdue-invoice chase, a reschedule without groveling — in a saved prompt library so you're not rewriting the brief every Tuesday.

Fiction: constraints beat "write a story"

"Write a story about a robot" produces a synopsis with the life sanded off. Fiction needs constraints, and the more specific the constraint, the more the model has to invent instead of retrieving a cliché.

Good creative writing prompts pin down five things: point of view, genre, a scene (not a whole plot), a character who wants something, and an obstacle in the way.

Write a scene, not a summary. ~400 words. POV: first person, present tense. GENRE: quiet literary sci-fi. SCENE: a night-shift keeper on a colony moon realizes the automated fog signal is answering a call that isn't the ship she expected. SHE WANTS: to clock off on time and not get involved. OBSTACLE: the signal knows her name. End on an image, not a resolution.

"Scene, not summary" and "goal plus obstacle" do most of the work. They force conflict onto the page instead of a plot description hovering above it. Vary POV and tense between runs to find the version that reads like a book.

Rewriting and editing prompts

The most useful AI prompts for writing often aren't about generating — they're about fixing something that already exists, whether the model wrote it or you did. Editing prompts want one job at a time. "Make this better" gets a lateral move; naming the specific fix gets the fix.

  • Tone: "rewrite this in a plainer, less salesy voice — no adjectives you couldn't defend in court."
  • Clarity: "same meaning, half the jargon; a smart 12-year-old should follow it."
  • Cutting: "cut this by 30% without losing a single concrete detail; kill the adverbs first."

When a whole passage is off rather than one line, stop editing the output and fix the instruction. Rewriting the prompt itself beats nudging a bad draft toward good. The errors that quietly wreck edits — vague verbs, contradictory instructions — are catalogued in the common prompting mistakes guide.

Feed the model a sample of your voice

You can describe your voice for a paragraph and still miss it. Showing beats telling: paste a few hundred words you wrote and label them clearly.

Below is a sample of my writing. Match its rhythm, sentence length, and vocabulary — not its topic. Notice what I do and don't do (never exclamation points; short fragments for emphasis; no corporate verbs). SAMPLE: [paste 200–400 of your own words] Now write a 150-word intro for a post about switching to a standing desk, in that voice.

Two real samples steer tone harder than any list of adjectives you could write. For a running start on structure before you drop your voice in, a ChatGPT prompt generator gives you a scaffold to paste the sample into.

Tip: Save your voice sample once with the label "voice reference — match rhythm and vocabulary, not content," and prepend it to any writing task. You stop re-explaining who you are, and every draft lands closer on the first try.

Strip the AI tells before you hit send

Every AI draft carries the same fingerprints. One deliberate editing pass removes most of them:

  • Delete throat-clearing first sentences. "In the world of writing, there are many..." — gone.
  • Kill the triads. Models love "clear, concise, and compelling." Keep one, prove it.
  • Break the rhythm. When every sentence runs 18 words, the page hums like a machine. Drop in a four-word one.
  • Cut hedges: "it's worth noting," "arguably," "in many ways."
  • Add one thing the model couldn't know — a real number, a real name, something that actually happened to you.

You can even make the model do the first sweep:

Edit this draft to remove AI tells: delete filler openers, break up uniform sentence length, cut hedging phrases, and replace abstract nouns with the concrete thing they stand for. Show the edited version, then list what you changed and why.

Strong AI prompts for writing don't replace the writer. They get you a better first draft in a tenth of the time, leaving the part that matters — deciding what's true, what's yours, what's worth saying — where it belongs: with you.

References

Put this into practice. Apply what you just read with our free tool: Prompt Optimizer →
FAQ

Frequently asked questions

Give it a specific angle, audience and voice, feed it a sample of the tone you want, and ban filler phrases. Generic input produces generic prose, so the more concrete your brief, the more human the result.
It can draft one quickly, but the best results come from you supplying the outline, key points and facts, then editing the draft. Unedited AI articles tend to be shallow and repetitive.
Prompts that set constraints — genre, point of view, tone, length and a concrete situation — beat open-ended ones. Ask for a scene rather than ’a story’, and give the character a clear goal and obstacle.
It depends on the rules that apply to you. Many workplaces allow AI as a drafting aid; many schools restrict it. Always check the policy, and treat AI as an assistant rather than the author.

Write your next prompt in seconds

Turn a rough idea into a clear, structured prompt any AI can follow. Free, private, and no account needed.

Open the Prompt OptimizerSee all tools