Articles that say something, in a structure that earns the reader’s next paragraph. Below are 7 copy-ready prompts. Fill in the [BRACKETS], copy, and paste into ChatGPT, Claude, Gemini or any capable assistant.
The generic AI article exists because the generic AI prompt asks for an article about a topic. A topic is not an argument, so you get a survey of the topic — which is what everyone else got too.
The 7 prompts
Outline an article that has an actual argument
Plan a piece before writing it, structured around a point rather than a topic.
Outline an article. TOPIC: [TOPIC] THE POINT I WANT TO MAKE: [YOUR ARGUMENT - not the topic, the claim] READER: [WHO THEY ARE AND WHAT THEY ALREADY KNOW] WHAT THEY SHOULD DO OR THINK DIFFERENTLY AFTER READING: [OUTCOME] LENGTH: [WORD COUNT] WHAT I KNOW THAT OTHERS DO NOT: [YOUR EXPERIENCE, DATA, OR ANGLE] Produce: 1. THE THESIS in one sentence, phrased so that a reasonable person could disagree with it. If nobody could disagree, it is a topic, not a thesis - tell me and propose three sharper ones. 2. THE OUTLINE - sections with: the heading, the claim that section makes, the evidence or example supporting it, and the word budget. Each section must advance the argument. If a section does not, cut it. 3. THE OPENING - three options: a specific scene or example, a counterintuitive claim, or a question the reader has already asked themselves. No 'in today's fast-paced world'. 4. WHAT TO CUT - the obvious sections a generic article would include that add nothing here. Naming them stops you writing them. 5. THE OBJECTION - the strongest counter-argument a smart reader will have. Where in the structure do you address it? An article that ignores the obvious objection reads as naive. 6. WHAT ONLY YOU CAN WRITE - given what I told you I know, the parts of this article that could not be written by someone without that knowledge. This is the whole value of the piece; make sure the outline gives it room. Do not produce a listicle unless the content is genuinely a list.
What you get: A thesis that can be disagreed with, a section-by-section outline with word budgets, and the parts only you could write.
Tip: Point 1's disagreement test kills most article ideas on contact, which saves you writing them. 'Why content marketing matters' fails; 'most content marketing is worth less than one good landing page' passes.
Write an opening that earns the second paragraph
Fix the first hundred words, which is where most readers leave.
Write the opening for this piece. WHAT THE ARTICLE ARGUES: [THESIS] READER: [WHO AND WHAT THEY CARE ABOUT] HOW THEY ARRIVED: [search result / newsletter / social link / sent by a colleague] MY CURRENT OPENING: [PASTE, or 'none yet'] TONE: [DESCRIPTION OR EXAMPLE] Write five different openings, each under 90 words, each a different approach: 1. THE SPECIFIC SCENE - a concrete moment, with detail. No abstraction. 2. THE CONTRADICTION - something the reader believes, and why it is wrong 3. THE COST - what the reader is losing right now by not knowing this 4. THE DIRECT ANSWER - state the conclusion immediately, then explain. Best for search arrivals. 5. THE QUESTION THEY ALREADY HAVE - phrased in their words, not yours For each: one line on who it works for and what it risks. Rules: - No 'in today's world', 'in an era of', 'we all know that', 'imagine if', 'have you ever' - No defining a term the reader already knows - No explaining what the article will cover. Just start. - The first sentence must contain something specific: a number, a name, a moment, or a claim Then: given how the reader arrived, which of the five you recommend and why. Search arrivals want the answer; newsletter readers will tolerate a scene.
What you get: Five distinct openings under 90 words each, with a recommendation matched to how the reader arrived.
Tip: Matching the opening to the traffic source is the part most writers skip. Someone arriving from a search wants the answer in sentence one, not a scene.
Cut a draft down without losing anything
Reduce word count by a third and make it better.
Cut this draft. Target: [TARGET WORD COUNT] from [CURRENT WORD COUNT]. DRAFT: """ [PASTE] """ WHAT MUST SURVIVE: [THE ESSENTIAL POINTS] Work in this order: 1. WHOLE SECTIONS - anything that does not advance the argument. Cutting a paragraph beats trimming ten sentences. Name what you cut and why. 2. THE RUN-UP - openings of sections that clear their throat before saying something. Usually the first one or two sentences of each section can go. 3. RESTATEMENT - the same point made twice in different words. Keep the better one. 4. HEDGING - 'I think', 'it seems', 'arguably', 'somewhat', 'quite', 'really', 'very', 'in order to', 'the fact that'. Cut unless the uncertainty is real and load-bearing. 5. EXPLAINING THE OBVIOUS - things this specific reader already knows. 6. SENTENCE-LEVEL - tighten what remains. Output: - THE CUT VERSION at the target length - WHAT WENT - a list of cut sections with one line each on why - WHAT I PROTECTED - anything I said must survive that was hard to keep, and how you kept it - WHAT GOT BETTER - the places where cutting improved the argument rather than just shortening it - WHAT I WOULD CUT NEXT if you need to go further, and what it would cost Do not cut specificity. Examples, numbers and names are what make writing worth reading; cut the connective tissue around them instead.
What you get: A cut version at target length, a log of what went, and a note on where cutting improved the argument.
Tip: The 'do not cut specificity' rule is what stops this producing bland summaries. The examples are the article; the transitions are not.
Turn your own raw notes into a draft in your voice
Get from scattered thoughts to a first draft that still sounds like you.
Turn my notes into a draft. Preserve my voice - do not smooth it into generic prose. MY NOTES (messy, unordered): """ [PASTE] """ A SAMPLE OF MY WRITING, so you can match the voice: """ [PASTE 200-400 WORDS YOU HAVE WRITTEN] """ TARGET LENGTH: [WORDS] READER: [WHO] THE POINT: [WHAT IT SHOULD ARGUE] Before writing, tell me: - THE VOICE ANALYSIS - from my sample: sentence length and variation, how formal, whether I use contractions, first or second person, how I use humour if at all, my typical paragraph length, and two or three habits that are distinctively mine. Then write the draft, following that analysis. Rules: - Use my own phrasings from the notes wherever they work. My rough phrasing is usually more alive than a polished replacement. - Where my notes are just a fragment, expand it rather than replacing it. - Do not add transitions I would not write. Do not add a conclusion that restates the article. - Mark [THIN] anywhere the notes do not give enough to write the section properly. Do not invent content to fill it. - Mark [CHECK] on any factual claim derived from my notes that I should verify. - No em-dash-heavy rhythm unless my sample uses it. No tricolons unless my sample uses them. After the draft, list: the three places where you were least confident of matching my voice.
What you get: A draft in your analysed voice, with thin sections and unverified claims flagged rather than filled in.
Tip: Making it state the voice analysis before writing is what makes this work. Without it, the model defaults to its own register regardless of the sample.
Write headlines and test them against each other
Generate title options and know which to use where.
Write headline options. WHAT THE PIECE ARGUES: [THESIS] READER: [WHO] WHERE IT WILL APPEAR: [blog / newsletter subject / social post / search result] PRIMARY KEYWORD IF SEO MATTERS: [KEYWORD, or 'not relevant'] THE MOST INTERESTING THING IN THE PIECE: [THE BEST BIT] Write 12 headlines across these approaches, three each: A. THE SPECIFIC CLAIM - states the conclusion. Works in search and for busy readers. B. THE UNEXPECTED - something the reader does not expect to be true C. THE USEFUL - what the reader gets, concretely D. THE CURIOSITY GAP - but honest. It must be answered by the piece, and the answer must be worth it. For each, note the character count. Then assess: - SEARCH: which work if someone is searching for this. Keyword placement, length under 60 characters for full display. - SOCIAL: which would be clicked in a feed. - EMAIL: which work as a subject line - shorter, more personal, no clickbait because it damages sender reputation. - THE HONEST TEST: for each, does the piece actually deliver what the headline promises? Mark any that overclaim. An overclaiming headline gets the click and loses the reader. Finally: your top pick for each destination, and the one you would avoid despite it being the most clickable, and why. No 'the ultimate guide', no 'you won't believe', no 'here's why', no numbers unless the piece is genuinely a list of that many things.
What you get: Twelve headlines in four styles with character counts, assessed per channel and checked for overclaiming.
Tip: The honest test is the one that protects you. The most clickable headline is often the one the article cannot back up, and that trains readers not to click next time.
Add specificity to writing that reads as generic
Fix a draft that says nothing wrong and nothing memorable.
This draft is too generic. Diagnose it and tell me what to add. DRAFT: """ [PASTE] """ WHAT I ACTUALLY KNOW ABOUT THIS: [YOUR EXPERIENCE, EXAMPLES, DATA, STORIES] READER: [WHO] Produce: 1. THE GENERIC SENTENCES - quote every sentence that could appear in any article on this topic by any author. These are the ones to fix. 2. THE ABSTRACTION LADDER - for each, what the concrete version would be. 'Businesses struggle with communication' → what business, what communication, what happened. Where I have given you the specific, write the replacement. Where I have not, write [NEED: the specific thing required]. 3. THE MISSING PROOF - claims stated without evidence. For each: what would make it credible - a number, an example, a named case, a quote, or a description of a specific instance. 4. WHAT YOU KNOW THAT IS NOT IN THE DRAFT - from what I told you, the specifics I have that did not make it in. This is usually where the article's value is hiding. 5. THE ONE PARAGRAPH TO REWRITE - if I only fix one, which, and the rewritten version. 6. THE STRUCTURAL PROBLEM - if the draft is generic because it is trying to cover too much, say so. Breadth is the most common cause of vagueness, and no amount of sentence editing fixes it. Propose the narrower piece. Do not rewrite the whole draft. Tell me what is missing, because I am the one who has the specifics.
What you get: Quoted generic sentences, their concrete replacements or a [NEED] marker, and a diagnosis of whether breadth is the real problem.
Tip: Point 6 is the honest answer more often than writers want. An article covering five things is vague about all of them; the fix is to write about one.
Update an old article that has gone stale
Refresh existing content rather than writing something new.
Help me update this article. ARTICLE: """ [PASTE] """ ORIGINALLY PUBLISHED: [DATE] WHAT HAS CHANGED SINCE: [WHAT YOU KNOW] HOW IT IS PERFORMING: [TRAFFIC, RANKINGS, ANY DATA] WHY I AM UPDATING IT: [it's ranking and stale / it stopped ranking / it's wrong now] Produce: 1. WHAT IS NOW WRONG - factually incorrect, referring to things that no longer exist, or describing a state of affairs that has changed. Quote each. 2. WHAT IS DATED WITHOUT BEING WRONG - references, examples, tools, prices, screenshots, and phrasings that mark this as old even though the substance holds. 3. WHAT IS MISSING - developments since publication that a reader would expect to see covered. Mark anything you are not certain about as [VERIFY]. 4. WHAT STILL HOLDS - the parts that are as true as they were. Protect these; the instinct when updating is to rewrite everything, and that usually loses the good bits. 5. STRUCTURAL CHANGES - what should be reordered. Often the once-novel part is now background and should be compressed, while a formerly minor point is now the main event. 6. THE UPDATE PLAN - ranked by (impact / effort). Distinguish: must fix (wrong), should fix (dated), and nice to have. 7. DATE HANDLING - whether to update the published date, add an update note, or both. If the piece ranks, warn about what changes risk: substantially rewriting a ranking page can cost the ranking. Say which of your recommended changes carry that risk. 8. YEAR IN THE TITLE - if the title contains a year, say whether to update it or remove the year entirely. Removing it is usually better for evergreen content. Do not rewrite from scratch. This piece has history; preserve what works.
What you get: A separation of wrong from merely dated, protected good parts, and a ranked update plan with ranking-risk warnings.
Tip: Point 4 exists because updating usually makes articles worse. The instinct to rewrite everything destroys the specific bits that earned the traffic.
Where AI actually helps here
- Structuring material you already have into something with a spine
- Writing the connective sections between points you care about
- Producing the four openings you would have agonised over so you can pick one
Where it falls down
- Having a point of view. The only opinions in the output are the ones you supplied
- Current facts. Anything time-sensitive needs checking against a source, every time
- Originality of example. It reaches for the same three case studies everyone reaches for
The mistake almost everyone makes: Asking for an article instead of an argument
Change “write an article about X” to “write an article arguing that X, for readers who currently believe Y”. The second produces something with a direction. If you cannot state the argument, that is the work — and no prompt will do it for you.
Free tool: Prompt Builder
Runs in your browser. No sign-up, nothing uploaded.
Questions people ask
Does Google penalise AI-written content?
Google’s stated position is that it rewards helpful content regardless of how it was produced, and acts against content mass-produced primarily to rank. In practice that means the volume-and-thinness pattern is the risk, not the tool.
How do I make AI writing sound like me?
Give it three samples of your own writing and ask it to describe the patterns it sees — sentence length, how you open, what you never do. Keep that description and reuse it as a style block. It works better than adjectives like “conversational”.