Most lists of ChatGPT prompts for writing are padded with prompts nobody would run twice. This one is 35 prompts we use for blog posts, SEO articles, emails and social posts, each with a line on why it works and what to change for your case. Copy them, fill the brackets, and keep the ones that earn their place.
How to prompt ChatGPT for writing (5 rules)
The prompts below follow the same five rules. If you write your own, keep these and you will get better output than most templates.
- Give it the reader, not the topic. "Write about invoicing" produces a Wikipedia summary. "Write for a plumber who invoices from a van and hates admin" produces something a plumber would read.
- Paste an example of the voice. Three paragraphs of your own writing beats any adjective. "Conversational" means nothing to a model. Your last blog post does.
- State the format and the length. Number of sections, number of words, whether you want a table, whether you want a conclusion at all.
- Ban the tells up front. Tell it no em dashes, no "in today's fast-paced world", no rule-of-three lists, no closing summary. It obeys these more reliably than "sound human".
- Make it ask before it writes. Add "ask me up to three questions before you start" to any long-form prompt. The questions catch the gaps you forgot to fill.
Prompt engineering as a discipline is younger than most people think. Wikipedia's entry on prompt engineering dates the mainstream techniques, such as chain-of-thought prompting, to 2022. There is no secret syntax. There is only specificity.
Blog post prompts
1. The first draft that follows a real outline
Write a [word count]-word blog post titled "[title]" for [reader]. Use this outline exactly, one H2 per line: [paste outline]. Match the voice in this sample: [paste 2-3 paragraphs of your writing]. Keep the intro under 80 words and go straight to the first section. No em dashes, no closing summary section, no rhetorical questions.
Why it works: you own the structure and the voice, so the model only fills paragraphs. The output needs editing, but not restructuring.
2. The intro rewrite
Here is the intro of my post: [paste]. Rewrite it in under 70 words. First sentence states the problem in plain language. Second sentence says what the reader gets. Cut any sentence that could open any article on any topic.
Why it works: intros are where ChatGPT pads the most. Giving it a sentence-by-sentence brief removes the throat-clearing.
3. The concrete example generator
My section says: [paste paragraph]. Write one concrete example that shows this in practice for [reader]. Use real numbers where plausible and mark any number you invented with [CHECK] so I can replace it.
Why it works: the [CHECK] marker is the important bit. It stops made-up statistics from sliding into a published page.
4. The how-to step expander
Turn this step into instructions someone can follow without asking a question: "[step]". Write it as 3-6 imperative sentences. Name the exact menu, field or command where relevant. End with one sentence describing what the reader should see when it worked.
Why it works: how-to posts fail on vague steps. The "what you should see" line is what readers use to know they are on track.
5. The listicle entry
Write a 120-word entry for "[product or option]" in a list post titled "[title]". Structure: one sentence on what it is, one on who it suits, three bullets of specific reasons it earns the spot, one honest limitation. Facts I have verified: [paste]. Do not add features I did not list.
Why it works: the last sentence matters more than the rest. Left alone, the model invents features, integrations and prices.
6. The opinion piece with a position
Write a 900-word opinion post arguing that [position]. Reader: [reader]. Open with the claim in the first sentence. Give three reasons, each with an example from [industry]. Address the strongest counter-argument in its own section and concede what is true in it. No hedging phrases such as "it depends" or "there are many factors".
Why it works: opinion posts from ChatGPT default to both-sides mush. Forcing a first-sentence claim and a named counter-argument gives it a spine.
7. The comparison post on equal criteria
Compare [A] and [B] for [reader] on these criteria only: [list 4-5]. One section per criterion, same length, both options covered in each. End with a verdict that names which reader should pick which. Use only the facts I paste here: [facts].
Why it works: comparison posts drift when each option gets its own section. Criterion-first structure keeps it fair and scannable.
8. The FAQ block
Write 5 FAQ entries for a post about [topic]. Questions phrased the way people type them into Google. Answers 40-70 words, direct answer in the first sentence, no "great question" openers.
Why it works: the first-sentence answer is what AI search engines and featured snippets lift. Burying it in the third sentence wastes the section.
SEO writing prompts
9. Search intent check
The keyword is "[keyword]". Based on what people typing this most likely want, classify the intent (informational, commercial, transactional, navigational), name the format the top results probably use (list, how-to, comparison, definition), and suggest a title that matches that format. One paragraph.
Why it works: the biggest SEO writing mistake is writing a how-to for a "best X" query. This prompt catches it before you write 2,000 words in the wrong format.
10. The outline from a SERP
Here are the H2s from the top three ranking pages for "[keyword]": [paste]. Build an outline that covers every topic they share, adds two sections they all miss for [reader], and drops anything that is filler. Return H2s and H3s only.
Why it works: you supply the real SERP, so the model is organising evidence rather than guessing what ranks.
11. Title tag options
Write 8 title tags under 60 characters for a page targeting "[keyword]". The keyword must appear near the start. Mix formats: number, how-to, question, plain. No clickbait, no year unless I say so. Show the character count next to each.
Why it works: the character count in the output saves a round trip and stops you shipping a title that gets truncated.
12. Meta description
Write 3 meta descriptions under 155 characters for a page titled "[title]". Each one states what the reader gets and includes "[keyword]" once. No "discover", "unlock" or "look at".
Why it works: the banned-word list is short but it removes most of the phrases that make a description read as generated.
13. Keyword placement without stuffing
Here is my draft: [paste]. The primary keyword is "[keyword]". Tell me where it appears now, then suggest at most three natural places to add it, quoting the sentence you would change. Do not rewrite the draft.
Why it works: asking for suggestions rather than a rewrite keeps your voice intact and stops the model from bolding or repeating the keyword everywhere.
14. Internal link anchors
I want to link from this paragraph: [paste] to a page titled "[target page title]". Suggest 3 anchor texts of 2-5 words that describe the target page and would read naturally in the paragraph. Show each one inside the rewritten sentence.
Why it works: anchors written from the target page's title are more accurate than anchors written from the source paragraph, which tend to describe the wrong page.
15. The quotable claim
Rewrite this paragraph so its main point is one self-contained sentence that could be quoted on its own: [paste]. The sentence must include the subject, the claim and the source or number. Keep the rest of the paragraph as support.
Why it works: AI Overviews and answer engines lift sentences, not paragraphs. A claim that stands alone gets cited. A claim spread over four sentences does not.
16. The content refresh
This post was published in [year] and ranks on page 2 for "[keyword]": [paste]. List every claim, number or tool mention that is likely out of date. For each one, tell me what to check. Then suggest one new section that the current top results cover and this post does not.
Why it works: refreshing a page that already ranks is faster than writing a new one. The prompt turns a vague "update this" into a checklist. If you want the full method, we wrote up the nine-step process for AI SEO writing that ranks with the refresh step included.
Editing and rewriting prompts
17. The AI-tell scrub
Edit this text and remove the following without changing the meaning: em dashes, sentences that start with "In today's", the phrase "not just X but Y", lists of exactly three adjectives, any sentence ending in a participle clause like "highlighting its importance", and any closing sentence that restates the paragraph. Return the edited text only. Text: [paste]
Why it works: naming the patterns beats asking it to "sound human". These six cover most of what makes a draft read as generated.
18. Cut by a third
Cut this to two thirds of its current length. Remove repetition and hedging first, then examples that make the same point as an earlier example. Do not remove numbers, names or steps. Text: [paste]
Why it works: the priority order stops it from cutting the substance and keeping the filler, which is what happens when you only say "make it shorter".
19. Vary the sentence length
This passage has sentences of similar length and rhythm. Rewrite it so sentence length varies between 4 and 30 words, keeping every fact. Do not add new content. Text: [paste]
Why it works: monotone rhythm is the tell readers notice without being able to name it. Explicit bounds fix it.
20. Plain-language pass
Rewrite for a reader with no background in [field]. Replace jargon with the plain word, and where there is no plain word, keep the term and add a five-word explanation in parentheses the first time. Reading level: someone who reads news, not papers. Text: [paste]
Why it works: "simplify" makes the model delete detail. Telling it how to handle each jargon term keeps the detail and fixes the reading level.
21. Fact-check list
List every factual claim in this text as a numbered list, one per line, with the exact quote. Mark each as: verifiable with a source, opinion, or unverifiable as written. Do not correct anything. Text: [paste]
Why it works: you cannot check what you cannot see. The list turns a wall of text into a checklist you can work through before publishing.
22. Voice match
Here are three paragraphs I wrote: [paste]. Here is a draft written by someone else: [paste]. Rewrite the draft so it reads like I wrote it. Match my sentence length, my contractions, how often I use "you", and how I open paragraphs. Keep every fact.
Why it works: the model is good at pattern matching when you give it the pattern. It is bad at guessing what "my voice" means.
Email and landing page prompts
23. Cold email, one ask
Write a cold email under 90 words to [role] at [company type]. One specific observation about their business in the first line: [observation]. One sentence on what we do: [what you do]. One ask: [ask]. No "I hope this finds you well", no bullet points, no calendar link in the body.
Why it works: the constraints are the email. Length, one observation, one ask. Everything the model would add on its own is what gets cold emails ignored.
24. Onboarding sequence
Write a 4-email onboarding sequence for a new user of [product]. Email 1: the one action that gets them value fastest. Email 2: the most common mistake and how to avoid it. Email 3: a feature they will not find on their own. Email 4: an honest check-in with a reply-to question. Each under 120 words, subject lines under 45 characters.
Why it works: each email has one job. Sequences written without a per-email brief blur into four versions of the same welcome message.
25. Landing page hero
Write 5 hero headline and subheadline pairs for [product] aimed at [reader]. Headline under 8 words, states the outcome. Subheadline under 20 words, states how, with one concrete detail. No "smooth", "effortless", "boost" or "unlock".
Why it works: hero copy from ChatGPT defaults to the banned list. Removing four words removes most of the generic output.
26. Objection section
List the 5 objections a [reader] would have before buying [product] at [price]. For each, write a 40-word answer that concedes what is true in the objection before answering it. Do not use the phrase "we understand".
Why it works: an objection answer that concedes nothing reads as sales copy. The concession is what makes it credible.
27. Pricing page FAQ
Write 6 pricing FAQs for [product]. Plans: [paste plans and prices]. Cover: what happens at the end of the trial, cancellation, what is counted against limits, refunds, upgrades mid-cycle, and who the higher plan is for. Answers under 50 words, plain language, no "simply".
Why it works: pricing questions are predictable. Listing them stops the model from inventing softer questions nobody asks.
Social media prompts
28. LinkedIn post from a blog post
Here is a blog post: [paste]. Write a LinkedIn post under 200 words that makes ONE point from it, not a summary. Open with the point in the first line, no "I've been thinking about". Give one example. End with a question a reader could answer in a comment. No hashtags, no emojis.
Why it works: summaries get no engagement. One point with one example does. The comment question is the part that gets the post shown to more people.
29. X thread
Turn this into a thread of 6 posts, each under 240 characters. Post 1 states the result or claim. Posts 2-5 each give one step or reason. Post 6 links to [URL] with a plain sentence about what is there. No numbering like "1/6", no emojis.
Why it works: a thread is a list with a hook. Giving it the slot-by-slot structure produces a thread you can post without rewriting.
30. Reddit answer
Someone on r/[subreddit] asked: "[paste question]". Write a reply that answers the question directly in the first two sentences, then gives specifics from experience. Mention [your product] only if it is the direct answer, and if so, disclose that you work on it. No marketing language.
Why it works: the disclosure rule and the "only if it is the direct answer" rule keep you inside most subreddit rules. A pitch dressed as an answer gets removed.
31. Video description
Write a YouTube description for a video titled "[title]". First two lines: what the viewer learns, with "[keyword]" once. Then 5 timestamps I will fill in: [00:00] format. Then one line linking to [URL]. Under 150 words total.
Why it works: the first two lines are what shows before "more". Everything else is structure you can fill in.
Research and outline prompts
32. Topic cluster
I want to own the topic "[topic]" for [reader]. Propose one pillar page and 8 supporting articles. For each supporting article give a working title, the question it answers, and which section of the pillar it should link to. Skip anything that is a synonym of another entry.
Why it works: random one-off posts do not build authority. A cluster does, and the "which section it links to" column is the internal linking plan for free.
33. Questions people ask
List 20 questions a [reader] would ask about [topic] before, during and after [action]. Group them by stage. Phrase them the way someone would type them, not the way an expert would.
Why it works: the stage grouping maps to intent. Before-questions become explainers, during-questions become how-tos, after-questions become troubleshooting posts.
34. Interview prep
I am interviewing [role] about [topic] for an article aimed at [reader]. Write 10 questions ordered from easy to hard. Avoid questions that can be answered yes or no. Include two that ask for a specific story or number.
Why it works: the story and number questions produce the quotes that make an article worth reading. The rest is warm-up.
35. Brief for a writer
Write a content brief for an article targeting "[keyword]". Include: reader, intent, the format, the working title, an H2 outline, 3 questions the article must answer, 3 sources to cite (name the type of source, do not invent URLs), the internal pages to link to: [paste], and a list of phrases to avoid. Under 400 words.
Why it works: a brief is where quality is decided. This is also the prompt to use if you hand writing to a tool or a freelancer, and it is the shape of brief we use when Distribb writes an article for a project. For the rules the article itself should follow, see our list of 14 content writing rules for SEO that still hold.
What ChatGPT still gets wrong
Prompts fix a lot. They do not fix everything, and it helps to know where the model fails no matter how you ask.
- Numbers. It invents statistics, prices and dates with full confidence. Prompt 21 exists because of this. Treat every number as unverified until you have the source open.
- Products and features. Ask for a list of tools and it will include ones that shut down, ones that never existed, and features that were never built. Prompt 5 tells it not to add features you did not list. Check every product name in a browser before publishing.
- Recency. The training data has a cutoff, and browsing does not always fire. Anything that changed in the last year is a candidate for being wrong.
- Sameness. Ten posts written from ten prompts on one account converge on one voice. The fix is prompt 22, run against your own writing, every time.
- Distribution. A prompt gives you a draft. It does not publish it, link it to your other pages, find the keyword it should target, or earn a single backlink to it. The draft is maybe a fifth of the work of getting a page to rank.
From prompt to published page
The prompts above cover the writing. Distribb covers the rest of the loop. It runs keyword research against your Search Console data so you write for queries you can win, writes and publishes the article to WordPress, Webflow, Shopify, Framer, Ghost and the other common platforms, adds internal links from your existing pages, and includes your site in a backlink exchange of real businesses so each article picks up links instead of sitting alone. It runs on autopilot, and you can still bring your own prompts and voice samples into the project instructions so the output reads like you.
If you have been pasting ChatGPT drafts into a CMS by hand, the part you are missing is Distribb's backlink exchange, which is the piece that decides whether a good article ranks or just exists.
FAQ
What is the best ChatGPT prompt for writing a blog post?
The best prompt gives ChatGPT the reader, an outline, a sample of your voice and a word count, then asks it to write one section at a time. Prompt 1 above is that prompt. A single-line request like "write a blog post about X" produces a generic summary you will end up rewriting.
Can ChatGPT write SEO content that ranks?
It can write the draft of a page that ranks, if you supply the keyword, the intent, the structure of the pages already ranking and the facts. It cannot do keyword research from live data, check what the top results say, or build the links a page needs. Those steps still decide the ranking.
How do I make ChatGPT writing sound less like AI?
Paste three paragraphs of your own writing and ask it to match sentence length, contractions and paragraph openers. Then run an edit pass that bans specific patterns: em dashes, "in today's", "not just X but Y", triple adjectives and participle endings. Naming the patterns works. Asking it to "sound human" does not.
Should I use ChatGPT or a dedicated AI writing tool for SEO?
ChatGPT is better for one-off drafts, edits and brainstorming, because you control every prompt. A dedicated tool is better when you need keyword data, publishing to your CMS, internal linking and backlinks handled together, because those are the steps ChatGPT does not do. Many teams use both.
How long should a ChatGPT prompt for writing be?
As long as the brief needs to be, usually 60 to 200 words. Short prompts produce generic output because the model fills the gaps with its defaults. The prompts above are long because each constraint removes a default you would otherwise edit out by hand.
Does ChatGPT make up facts when writing?
Yes, and it does so in the same confident tone it uses for real facts. Statistics, prices, product features and quotes are the most common inventions. Ask it to mark anything it is unsure of, extract every claim as a list, and verify each one against a source before you publish.
Conclusion
Pick five prompts from this list that match what you write most, save them with your voice sample filled in, and use them for a week. Drop the ones you keep rewriting. The remaining ones are your writing system. When the drafts are good and the ranking still is not there, the missing pieces are keyword data, links and publishing, and that is what Distribb is for.