You asked an AI to write an article, and what came back looks like thousands of others: an intro that “dives into the fascinating world of,” smooth-but-empty paragraphs, tips everyone already knows, and a conclusion that repeats the intro. It’s clean, but it’s hollow. Nobody wants to read that all the way through.
The good news: it doesn’t have to be that way. Writing an article with AI can give you a text that’s lively, personal and useful, as long as you don’t hand over the wheel. In this guide, I’ll show you the method I use myself: what you need to bring, what AI does well, ready-to-copy prompts for every step, and the mistakes that make a text generic. It works with ChatGPT, Claude or Gemini.
The short version
- A generic text comes from a generic prompt: the AI fills the gaps with clichés.
- Bring your own material: real experiences, examples, numbers you know, your personal opinion.
- Give a detailed outline and tone guidelines before asking for a single sentence.
- Use AI to structure, suggest variations, headlines, counterarguments or an FAQ, not to think for you.
- Check every fact and rewrite the final version in your own words: raw output is never automatically accurate.
Information checked on September 29, 2026.
Contents
Why AI produces generic writing
A language model predicts the most likely continuation of a text. When you ask it to “write an article about remote work,” it knows nothing about you, your reader, your experience or your opinion. So it produces the average of everything that’s already been written on the topic. The result is fine, but interchangeable.
In other words, it’s not that the AI is “bad”: it’s the raw material that’s missing. The vaguer your request, the flatter the text. The more content you bring that’s uniquely yours, the more distinctive the text becomes.
So in my view, the question to ask isn’t “which AI writes best?” but “what do I have to say that nobody else can?” AI then helps you shape that content.
What AI does well (and what it doesn’t)
Before you start, it helps to know where AI saves you time and where it wastes it.
| Task | Is AI useful? | My advice |
|---|---|---|
| Finding angles and outline ideas | Yes | Ask for several options and pick one |
| Suggesting titles and subheadings | Yes | Keep the one that sounds like you |
| Rewording a clumsy passage | Yes | Specify the tone you want |
| Finding counterarguments | Yes | Great for strengthening a piece |
| Writing an FAQ | Yes | Check every answer |
| Providing facts, figures, dates | With caution | Always verify at the source |
| Telling your story | No | It would make it up |
| Having an opinion for you | No | That’s your added value |
The 6-step method
Here’s the method I use. Each step has its own prompt, to adapt to your topic.
- Gather your personal material.
- Define the reader, the goal and the tone.
- Build a detailed outline with the AI.
- Write section by section, feeding in your examples.
- Check the facts yourself.
- Rewrite the final version.
Step 1: Gather your personal material
This is the step almost everyone skips, and yet it’s the one that makes the difference. Before you open ChatGPT, take ten minutes to jot down:
- what you’ve experienced on the topic (a success, a failure, a surprise);
- the concrete examples you know;
- your opinion, even a strong one;
- the questions your readers or customers actually ask you;
- the misconceptions you want to correct.
If you’re short on ideas, the AI can interview you. It’s one of my favorite uses:
I want to write an article about [topic] for [reader]. Ask me 10 questions, one at a time, to draw out my personal experience, my concrete examples and my opinions on this topic. Write nothing but the questions, and wait for my answer before asking the next one.
Answer freely, out loud if you prefer dictating. By the end, you’ll have material the AI could never have made up.
Step 2: Define the reader, the goal and the tone
An article is written for someone. Describe that reader, what they already know and what they should take away. Add specific tone guidelines: that’s what keeps you out of “brochure” style.
Here's the context for my article. Reader: [description]. What they already know: [level]. What they should remember or do after reading: [goal]. Tone: [direct, warm, a bit of humor], short sentences, speak directly to the reader, no superlatives, no phrases like "in today's ever-changing world" or "let's dive in." Confirm you understand in one sentence, without writing anything else.
To avoid repeating these instructions every time, you can save them once and for all with ChatGPT’s custom instructions.
Step 3: Build a detailed outline
Don’t ask for the article right away. Ask for an outline first, discuss it, change it. A detailed outline is the best protection against filler.
Based on my notes below, suggest 3 different outlines for an article of about [number] words. For each outline: the title, the subheadings, and under each subheading, the main idea and the personal example to use (taken from my notes). If a section has no example in my notes, flag it instead of making one up.
--- [your notes] ---
Pick an outline, then refine it: cut a useless section, merge two, move your best example to the top. You’re the one who decides the structure.
Step 4: Write section by section
Write one section at a time rather than the whole article in one go. You stay in control, and each section stays tight.
Write only the "[subheading]" section, in about [number] words. Use my example: [your example]. Stick to my information, and don't add any numbers, studies or quotes. If you're missing information, write [TO FILL IN] instead.
The [TO FILL IN] instruction is really useful: it stops the AI from plugging a gap with a plausible invention.
If a text is too polished, ask for a more human version:
Rewrite this paragraph so it sounds like someone talking to a friend: shorter sentences, a concrete example, a clear opinion. Cut unnecessary adjectives.
Step 5: Check the facts
Everything the AI states must be checked before publishing: dates, numbers, names, prices, quotes, legal rules. An AI can be wrong with total confidence. I recommend asking it for the list of things to check:
Reread this text and list every verifiable factual claim (numbers, dates, names, quotes, rules). For each one, say whether it came from my notes or whether you added it. Don't correct anything.
Then check each item against a reliable source: official website, text of the law, original document. To move faster through a long source document, my guide to summarizing a text with ChatGPT shows how to ask for the exact passages.
Step 6: Rewrite the final version
This is the step that turns an AI-assisted draft into your article. Read it out loud. Anywhere you wouldn’t say the sentence that way, rewrite it. Cut repetition, empty transitions and end-of-paragraph wrap-ups that restate what was just said.
AI can also play the role of a demanding editor:
Act as a tough editor-in-chief. In this text, point out: generic sentences that could appear in any article, passages without a concrete example, repetitions and clichés. Don't rewrite anything, just list the problems with the sentence in question.
More useful prompts to enrich an article
Once the draft is written, AI is still a great partner for taking it further.
Find an original angle
Here's my topic: [topic]. List 10 different angles to cover it, from the most classic to the most unexpected. For each one, say in one sentence what sets it apart from the usual articles.
Strengthen your argument
Here's my thesis: [thesis]. Give me the 5 most serious objections a skeptical reader could raise, and for each one, the most honest possible response.
Suggest headlines
Suggest 10 headlines for this article, with no exclamation points, no superlatives and no clickbait. Vary the formats: question, concrete promise, number, benefit.
Write an FAQ
Based on this article, list the 6 questions a beginner reader would still have. Answer each in 2 or 3 sentences, using only the information in the article.
Which AI should you use for writing?
All three major assistants work for this method, and all have a free version.
- ChatGPT: very versatile, with a free version that’s widely accessible for text. The Plus plan costs $20 a month. The minimum age is 13, with parental consent under 18.
- Claude: to my taste, often the most natural in style, which makes it a good pick for writing. Free version with limited capacity, Pro plan at $20 a month or $200 a year. Claude.ai is for users 18 and older. See my Claude AI guide.
- Gemini: handy if you already work in Google Docs or Gmail. Free version with limits that depend on usage; Google AI plans start at $9.99 a month.
If you’re torn between the first two, I compared ChatGPT vs. Claude in detail.
On rights, all three companies’ terms point in the same direction: OpenAI says generated content belongs to the user to the extent permitted by law, Anthropic says you own your generated content under its terms, and Google doesn’t claim ownership of original generated content. That doesn’t exempt you from respecting other people’s rights.
Mistakes and limits to know about
Publishing the raw output
This is mistake number one. An unedited text almost always contains approximations, empty phrases and sometimes factual errors. Never present raw output as automatically accurate.
Letting the AI invent experiences
If you ask it to “add an example,” the AI will make up a believable anecdote. Presented as something that really happened, that’s misleading to your reader. Give your real examples, or clearly label a case as fictional.
Accidental plagiarism
An AI can reproduce phrasing close to existing texts, especially on very common topics. If you rely on a source, cite it. And never ask it to “rewrite” someone else’s article so you can publish it as your own: that’s plagiarism, even reworded.
School and workplace rules
In high school or college, AI use is often regulated, or even banned for certain assignments. Check with your teacher or professor before using it, and don’t turn in a generated text as your own work. At work, check your company’s policies, especially if you’re pasting confidential documents.
Watermarking of Claude text
On August 14, 2026, Anthropic announced a text watermark for content produced by its Claude models released after August 2, 2026. According to the company, it’s invisible to readers, adds no hidden characters and doesn’t identify the user; your rights to the text are unchanged. For me, it’s one more reason to genuinely rewrite: the goal isn’t to “hide” the AI, but to publish a text that’s truly yours in substance.
Example: from a vague request to a personal piece
Let’s take Megan, who runs a small travel blog and wants to write about Rockport, Massachusetts.
Her first request was: “Write an article about Rockport.” The result: a decent text about the harbor, the painters and the colorful fishing shacks, identical to what you find everywhere.
Her second attempt followed the method. She first wrote down her memories: the time of day she found the harbor almost empty, the café where she took shelter from the rain, the mistake she made by coming on a Saturday in August. She defined her reader (a couple planning a weekend getaway), her tone (friendly, a little wry) and a five-part outline. The AI wrote each section from her notes, flagging the [TO FILL IN] spots. Megan checked opening hours and practical information on official websites, then rewrote the whole thing reading it out loud.
The second article is about the same town, but it doesn’t look like any other. That’s exactly what you’re after.
My pre-publishing checklist
- I’ve included at least two real or concrete examples.
- The text includes my opinion, clearly presented as such.
- Every number, date, name and price has been checked against a reliable source.
- There’s no [TO FILL IN] left.
- I’ve cut the clichés and empty transitions.
- I’ve read the text out loud and rewritten whatever didn’t sound like me.
- The sources I used are cited.
- I’ve followed my school’s or company’s rules.
FAQ
Can you write an entire article with AI?
Technically yes, but the result will be generic if you don’t bring your own material. I recommend using it for the outline, variations and proofreading, and keeping control of the substance and the final version.
Is free ChatGPT enough to write an article?
For writing text, the free version is now widely accessible and enough for most uses. Paid plans offer higher limits. All the details are in my free ChatGPT guide.
Does Google penalize AI-written content?
I can’t give you a hard-and-fast rule. My take: a search engine, like a reader, is looking for useful, original content. A generic text is unlikely to stand out, whether it was written by an AI or a human.
How do you make AI writing sound more natural?
Bring your examples, set a specific tone, ban clichés, then rewrite the final version yourself while reading it out loud. Human rewriting is what makes the biggest difference.
Can text written with Claude be detected?
On August 14, 2026, Anthropic announced a text watermark for its models released after August 2, 2026, invisible to readers, with no hidden characters and no user identification. Your rights to the text don’t change.
Who owns an article written with AI?
Under the terms of OpenAI, Anthropic and Google, you own the generated content (to the extent permitted by law, for OpenAI) or the company doesn’t claim ownership of it. You remain responsible for what you publish, including respecting other people’s rights.
The bottom line
To write an article with AI without ending up with generic content: bring your personal material, set the reader and the tone, build a detailed outline, write section by section, check every fact and rewrite the final version. The AI structures, suggests and proofreads; you bring the experience, the opinion and the responsibility for the text.
To go further, check out my comparison on which AI to choose.
