Skip to main content
All posts
AI writing for X

ChatGPT Prompts for Twitter That Sound Like You, Not Like ChatGPT

Copy-ready ChatGPT prompts for Twitter posts, hooks and replies to ideas, built around your own past posts, plus a before and after showing why generic prompts fail.

Daniel Smidstrup

Founder of ClimbX

Follow on X

I grew my X account from 0 to 10K followers in 137 days.

9 min read
A chat window feeding a stack of the writer’s own past post cards into a funnel that outputs one orange-highlighted draft; charcoal line art on cream.

The short version

The ChatGPT prompts for Twitter that work feed in your own past posts. Paste your last 20 posts and ask ChatGPT to describe your voice. Then give every request a word limit from your best posts, a named audience and a ban on hashtags and filler. Prompts that only name a topic give you the same polished, generic post everyone else gets.

Most prompt lists for Twitter are one line long: “write a viral tweet about X”. They fail for the same reason every time. ChatGPT has never read your posts, so it writes like the average of the internet. Below are 12 prompts I would really use. Each one is built on material only you have.

The limits in them come from my own account, where short posts win. That is not a rule for everyone. Across 669 other ClimbX users, longer posts did slightly better. So use the length that beats your own median. The prompts also aim at reach and replies. A post that teaches, brings in good leads or sells often needs more room. Judge those by saves, clicks and DMs, not reply count.

The two-minute setup before any prompt

Collect three things once and keep them in a note. Every prompt below uses them. (1) Your last 20 original posts, as plain text, one per line. Skip replies and reposts. (2) Your 5 best and 5 worst posts by impressions, with the number next to each. X shows impressions under every post. (3) A few lines of raw notes from this week: what you shipped, a number that changed, a conversation that surprised you.

Run prompt 1 first and save its answer as your voice notes. Paste those notes at the top of every later prompt. Or put them in ChatGPT’s custom instructions so you do not have to repeat them. The prompts work the same in other chatbots.

12 ChatGPT prompts for Twitter, copy-ready

Replace anything in square brackets. Each prompt asks for options instead of one answer. Picking from three is faster than fixing one.

1. Extract your voice from your last 20 posts

Use it when: you start a new chat, or your posts feel like they drifted. Run it once a month.

Here are my last 20 posts on X:

[paste posts]

Describe how I write so another writer could copy it. Cover:
- typical length in words, and the shortest and longest
- capitalization, punctuation, line breaks, emoji use
- how I usually open a post and how I end one
- words and phrases I repeat
- topics I come back to
- 3 things I never do

Be specific and quote my posts as evidence. Do not write any new posts yet.

2. Find what separates your best posts from your worst

Use it when: you have at least a month of posts and want to know which shape to write more of.

Below are my 5 best and 5 worst posts on X, with impressions.

BEST:
[post] - [impressions]
WORST:
[post] - [impressions]

List the differences in length, opening, structure, topic and ending.
Only name a difference if it shows up in at least 3 of the 5 best posts.
Then give me 3 rules for my next posts, each under 12 words.

3. Turn your week into post ideas

Use it when: you sit down to plan a few days of posts and the page is blank.

My voice notes: [paste]

Raw notes from my week: [paste]

Find 8 post ideas in these notes. For each, give:
- the specific detail it is built on (a number, a name, a moment)
- the shape: question, observation, milestone, or short story
- one sentence on why my audience would reply

Do not invent anything that is not in my notes.

4. Hook variations with a hard word limit

Use it when: you have the idea but the first line is weak.

My voice notes: [paste]

Idea: [one sentence]

Write 6 different first lines for this post. Each must be 12 words or fewer.
Use 6 different openings: a number, a named audience ("Founders,"),
a contradiction, a confession, a direct question, a before-and-after.
No hashtags, no colons after a label like "hot take".
Use emoji only the way my voice notes say I use them.

5. Turn a long thought into a short post

Use it when: you wrote a paragraph in your notes app and know it is too long for a post.

My voice notes: [paste]

Here is a long thought: [paste]

Find the single sharpest claim in it. Write it as 3 posts:
one under 10 words, one under 20 words, one under 30 words with line breaks.
Keep my words where you can. Do not add advice I did not give.

6. Generate answerable questions

Use it when: you want replies, not just likes.

My audience: [who, in one line]
Topic: [topic]

Write 8 questions my audience can answer from their own experience in under 10 seconds.
Rules:
- start 3 of them by naming the audience ("Founders,")
- make 3 of them ask for a one-word or one-number answer ("in 1 word")
- no question with a right answer you could look up
- each under 15 words

7. Rewrite shorter

Use it when: a draft works but drags.

Rewrite this post at half its length: [paste]

Cut adjectives first, then any sentence a reader would not miss.
Keep the specific numbers and names. Keep my casing and line breaks.
Give 3 versions and tell me the word count of each.

8. Find the specific detail

Use it when: a draft is true but generic, and you cannot say why.

Here is a draft: [paste]

Do not rewrite it. Ask me 5 short questions that would pull out
the specific detail behind it: a number, a date, a name, what I saw,
what I did next. Wait for my answers before writing anything.

9. Audit a draft for AI tells

Use it when: before posting anything a model helped with.

Check this draft for anything that sounds machine-written: [paste]

Flag: emoji used as bullets, rhetorical "Here's the thing", lists of three
abstract nouns, "game-changer", "unlock", "dive in", closing "Thoughts?",
hashtags, and any claim with no example behind it.
Quote each problem and suggest a plainer replacement in my voice.

10. Turn your best reply into a post

Use it when: a reply you wrote on someone else's post got more attention than your own posts.

I wrote this reply on another post: [paste my reply]
The post I replied to was about: [one line]

Turn my reply into a standalone post that makes sense without the original.
Keep my wording. Under 25 words. Give 3 versions.

11. Find the counterpoint to your own take

Use it when: you are about to post an opinion and want to know how the replies will push back.

My take: [paste]

List the 3 strongest objections a smart person in my audience would raise.
Then tell me which one I should answer inside the post, in one line,
so the post holds up in the replies.

12. Outline a thread

Use it when: the idea needs more than one post. See the thread section below for the rules behind it.

My voice notes: [paste]
My raw material: [notes, numbers, examples]

Outline a thread. Do not write the posts.
- Post 1: one specific promise under 15 words
- Posts 2 to 6: one claim each, plus which example from my material proves it
- Last post: one question my audience can answer from experience
Mark any point where my material has no example, so I can add one or cut it.

Why do generic prompts make tweets sound like ChatGPT?

A prompt with no examples makes the model guess. Its guess is the most common social post: an emoji, a motivational claim, three vague tips and a “what do you think?” at the end.

That is why every prompt above starts from your own posts, your own week and your own numbers. Here is the same topic asked both ways. I wrote both outputs to show the pattern. They are not real posts.

Weak

Prompt: "Write a tweet about productivity for founders."

🚀 Productivity isn’t about doing more, it’s about doing what matters. Focus on high-impact tasks, cut distractions, and watch your startup soar. What’s your #1 tip? 👇 #StartupLife

Stronger

Prompt: 20 of my posts + "this week I shipped 3 features nobody asked for" + max 20 words, lowercase, one question.

shipped 3 features this week.

nobody asked for them.

founders, what did you ship that nobody asked for?

Made-up examples. Any account could post the generic one. The voice-fed one keeps the fact exactly as given and stays inside the limit at 18 words. It matches the writer’s lowercase short lines, names the audience and ends on a question a founder can answer from memory. Check every draft for that first part. A model that quietly changes your fact is worse than one that writes a dull line.

Why do the prompts carry hard limits?

A model adds length and labels by default. Those show up in my worst posts. It skips a named audience and an easy reply. Those show up in my best.

I tagged my top 500 and bottom 200 posts by trait. Then I compared how often each trait showed up in each group. A ratio of 4 means the trait was four times more common among the winners than the losers. Below 1 means it shows up more in the losers.

How much more common each trait is in my top 500 posts than my bottom 200
  • Community or connect call4.1x
  • Opens by naming the audience4x
  • Quick-reply hook3.3x
  • Question1.54x
  • "hot take:" label0.53x
  • 26 words or more0.5x
  • Mentions my own product0.4x
  • Opinion0.34x
Label study on @DanielSmidstrup, top 500 vs bottom 200 original posts. Win ratio = share of top posts with the trait divided by share of bottom posts with it.

The rules in the prompts come from that chart. Prompts 4 and 6 open with a named audience. That trait was 4.0 times more common among my winners. The word limits are there because posts of 26 words or more were half as common among them. Prompt 6 asks for one-word answers. Questions like “in 1 word” did well for me on a small set of posts. The result and its sample size are in how to get more replies on X.

Set your own word limit, not mine. In my 1,218 original posts from the last 180 days, posts of 1 to 8 words reached 1.29 times my median and posts of 26 to 45 words 0.80 times. Across 54,338 posts from 669 other ClimbX users in the same window the pattern flips: posts of 46 words or more reached 1.14 times each author’s median and posts of 1 to 8 words 0.92 times. There is no ideal length for everyone. Run prompt 2 on your own best and worst posts. Put the length that wins for you into prompts 4, 5 and 7.

Add two rules to every prompt, whatever your numbers say. Ban the “hot take:” style label. It sat at 0.53 in the chart. And tell the model not to mention your product unless the post is about it. That trait sat at 0.40. For a longer list of what makes AI text sound like AI, see what separates good AI content from filler.

ChatGPT prompts for Twitter threads

Use ChatGPT for the outline of a thread, then write the posts yourself. In my experience a fully generated thread has the same rhythm in every post. That makes it easy to stop reading. I have no data on that. It is just what I have noticed.

Here is how to keep your voice: (1) paste your material: notes, numbers and examples; (2) run prompt 12 for an outline only; (3) fill the gaps by adding an example or cutting the point; (4) write each post yourself, one per point; (5) run prompt 7 to shorten each post. The model does the structure and trimming. You write the sentences. That is where your voice is.

The outline step catches the most common thread problem: a point with no example behind it. Prompt 12 asks the model to mark those. For each one, find a real example or cut the point. A five-post thread with five real examples beats a ten-post thread padded out to look thorough.

Where does prompting stop being enough?

Prompting stops being enough when you need data the chat does not have: how your posts did, what is working right now for accounts like yours, and the context you would otherwise paste in every single day.

What you needChatGPT with these promptsWhat it costs you
Your voiceGood, once you paste 20 postsRe-pasting voice notes into each new chat
Your performance dataOnly what you copy in by handExporting impressions every week
What is working in your nicheNone. It cannot see current X postsScrolling for outliers yourself
Learning over timeStarts fresh unless you save notesKeeping a running doc of what worked

If you post a few times a week, the prompts above are enough. They cost nothing beyond the chatbot. If you post daily, the re-pasting becomes the job. ClimbX fixes that. Your past posts, how they did and what works for accounts a step ahead of you are already loaded. So the drafts start from your material with no copy-paste. You still edit every draft before it goes out. I wrote about how voice-trained drafting works in AI tweet writer in your voice, and about using AI for a personal brand without losing the person in how to use AI for personal branding. If you are worried about reach, the rules and data are in does X penalize AI generated posts.

How these numbers were measured

All numbers were queried on October 3, 2026. My own data is my 1,218 original posts on @DanielSmidstrup from the last 180 days, replies excluded, posts younger than 3 days left out so their impressions had settled. Lift is a post’s impressions divided by my own median. Each group shows the median lift. The trait chart is a separate label study. It compares my top 500 and bottom 200 original posts. Its ratios count how often a trait shows up, not how far those posts reached. The cross-user numbers are 54,338 original posts from 669 ClimbX users with at least 20 posts in the same 180-day window, each post divided by its own author’s median. My account is excluded from them, and no individual user’s post is shown.

Common questions

What are the best ChatGPT prompts for Twitter?

The best ChatGPT prompts for Twitter include your own past posts as examples, a word limit set from your own best posts, a named audience, a ban on hashtags and filler, and emoji only where you already use them. A prompt that only names a topic returns the same generic post everyone else gets.

Can ChatGPT write tweets in my voice?

Closer than most people expect. Paste 20 or more of your own posts and ask it to describe your voice before it writes anything. Without examples it falls back to a polished, emoji-heavy default. Expect to edit every draft, and add the one detail only you know.

What is a good ChatGPT prompt for Twitter threads?

Ask for an outline before any writing: a first post under 15 words that makes a specific promise, 4 to 6 numbered points, each one a single claim with a real example from notes you paste in, and a last post that asks one answerable question. Then write each post yourself from the outline.

How long should a tweet written with ChatGPT be?

Set the limit from your own posts, not a universal rule. In my 1,218 original posts from the last 180 days, posts of 1 to 8 words reached 1.29 times my median and posts of 26 to 45 words 0.80 times. Across 54,338 posts from 669 other ClimbX users in the same window, posts of 46 words or more did slightly better, at 1.14 times. Check which side your audience is on.

Will X penalize tweets written with ChatGPT?

X’s published rules do not penalize text for being written with AI. They restrict spam, duplicate posting and unsolicited automated replies. The real risk is a post that sounds generic. People scroll past it, and you see weak reach.

Drafts that start from your own posts.

ClimbX reads how you write and what has worked for accounts slightly ahead of you, then hands you ready posts that sound like you on a good day. You edit every one before it goes out.

Read next

Allow optional cookies for ads, analytics and referrals? Privacy details