A good ChatGPT prompt for a LinkedIn post gives the model your raw material instead of a topic. Not "write a post about hiring" — but this:
Here are my rough notes from a hiring decision I made this week:
[paste your notes]
Write a LinkedIn post from these notes. Keep my phrasing where it's
already specific. Don't add statistics I didn't give you. Open with
the most concrete sentence in the notes, not a summary. Under 200 words.
That single change — raw material in, not a subject line — fixes most of what makes AI posts sound like AI.
What makes a LinkedIn prompt work
Five mechanics do most of the work. Every prompt below uses at least two.
- Feed it your material. Notes, a Slack message, a call transcript, a half-written paragraph. A topic produces the statistical average of LinkedIn. Your notes produce something only you could post.
- Name the reader, not the subject. "For engineers who've inherited someone else's codebase" beats "about technical debt."
- Constrain the shape. Word count, paragraph count, whether it ends on a question. Unconstrained models drift long and hedge.
- Ask for options, not an answer. Three angles gives you something to choose between. One draft gives you something to argue with.
- Ban the tells explicitly. "In today's fast-paced world," the "It's not just X, it's Y" construction, and openers that restate the title. Models produce these unless told not to.
Finding something to post about
1. Mine your week
Here's what I worked on this week: [paste]. Find the three things a
[your role] outside my company would find genuinely surprising. For each,
tell me why it's surprising to an outsider but ordinary to me.
2. The question you keep answering
I get asked "[question]" constantly by [audience]. Give me five different
angles for a post answering it — each taking a different position, not
five phrasings of the same one.
3. Mine a decision
We chose [X] over [Y] because [reason]. Turn this into three post angles:
one that teaches the decision framework, one that argues against the
obvious choice, one that admits what we gave up.
4. The thing you changed your mind about
I used to believe [X]. Now I think [Y]. What changed my mind was [Z].
Draft a post about the shift. Do not make me sound like I had a
revelation — make it sound like slow accumulation of evidence.
5. Turn a bookmark into a post
I saved this article: [paste or summarise]. I disagree with [specific part].
Write a post that engages with the actual argument, not a summary. Take a
position in the first two lines.
6. Audit your own expertise
I'm a [role] with [n] years in [domain]. List 15 things I probably know
that someone three years behind me does not. Be specific to the domain —
no generic career advice.
7. The boring thing that's actually interesting
Here's a routine part of my job: [describe]. Explain why it would be
interesting to someone who's never done it. What's counterintuitive
about how it actually works?
8. Mine a failure
This didn't work: [describe]. Draft a post about it. Rules: no redemption
arc, no "and that's when I learned," and do not turn the failure into a
humblebrag about resilience. End on what I'd do differently.
Turning a raw idea into a draft
9. The workhorse
Rough notes: [paste]. Write a LinkedIn post. Keep my phrasing wherever
it's already specific. Add nothing I didn't say — no invented statistics,
names, or outcomes. Open with the most concrete line in my notes.
Under 200 words. Short paragraphs.
10. Three angles
Idea: [describe]. Give me three drafts of the same idea:
1. Story-led — open mid-scene, no setup
2. Contrarian — lead with the position, defend it
3. Structured — a short numbered breakdown
Same facts in each. Don't tell me which is best.
11. Voice-matched draft
Here are three posts I've written: [paste]. Study the sentence length,
how I open, how formal I am, whether I use fragments. Now write a new
post about [topic] in that voice. Then list what you matched, so I can
tell you what you got wrong.
12. The specificity forcing function
Write a post about [topic]. Every claim must be attached to a concrete
detail — a number, a name, a date, a specific tool. If you can't attach
one, cut the claim rather than hedging it.
13. Explain it to one person
Write this post as if explaining [topic] to [specific person — a former
colleague, a junior hire, a friend outside the industry]. Keep their level
of context in mind. No jargon they wouldn't use.
14. The short one
Idea: [describe]. Write it in under 80 words. Every sentence must do work.
No preamble, no summary line at the end.
15. The teardown
Break down how [thing] actually works: [paste what you know]. Structure it
as 4-5 short steps. Assume the reader is smart but hasn't done this.
No "in conclusion" ending.
16. Argue the other side
Most people in [field] believe [common view]. I think that's wrong because
[reason]. Write a post making my case. Steelman the common view first in
one or two lines, then dismantle it. Don't strawman it.
17. The receipts post
Here's what happened and the numbers: [paste]. Write a post that leads
with the number, explains what produced it, and is honest about what
we can't attribute to our own actions.
18. Rewrite from scratch
Here's my draft: [paste]. Don't edit it. Write a completely different post
making the same point, using a structure I didn't use. Then tell me in one
line what yours does better and what mine does better.
Hooks and openers
19. Ten hooks, one post
Here's my post: [paste]. Write 10 alternative opening lines. Rules: no
questions, no statistics I didn't provide, none that start with "I", and
none that restate the topic. Under 12 words each.
20. Start mid-scene
Rewrite my opening so it starts in the middle of the action, with no setup:
[paste first paragraph]. The reader should be slightly disoriented and have
to keep reading to orient.
21. The specific detail hook
Find the single most specific detail in this post and make it the first
line: [paste]. Cut whatever currently opens it.
22. Kill the throat-clearing
Here's my opener: [paste]. Delete every sentence before the point actually
starts. Show me what's left. If that's now the whole opener, say so.
23. Hook against the grain
Post: [paste]. Write 5 openers that contradict what the reader expects
from the topic — but that the post actually delivers on. No clickbait
I can't back up.
24. Number-led
Post: [paste]. Write 5 openers built around a concrete number from the
post. If there's no usable number in it, tell me instead of inventing one.
25. The one-line test
Compress this entire post into one sentence someone would stop scrolling
for: [paste]. Then tell me whether the post delivers on that sentence.
26. Fix a weak close
Here's my ending: [paste]. It's flat. Write 5 alternatives. No "what do
you think?", no call to comment, no summary of what I just said.
Editing a draft you already have
27. De-AI it
Edit this so it doesn't read as AI-generated: [paste]. Specifically: cut
"in today's...", any "It's not just X, it's Y" construction, adverbs
carrying weight the verb should carry, and any sentence that could
open any post on any topic.
28. Cut 40%
Cut this by 40% without losing a single concrete detail: [paste]. Cut
qualifiers, throat-clearing, and restatement first. Show me what you cut
and why.
29. Find the hedges
List every hedge in this post — "I think", "arguably", "somewhat",
"can be": [paste]. For each, tell me whether it's protecting me from a
claim I can't support, or just weakening one I can.
30. Line-level tightening
Rewrite each sentence in this post to be shorter, without merging or
cutting any: [paste]. Same number of sentences, same order, less weight.
31. The credibility check
Read this post as a skeptical [role in your field]: [paste]. What would
they push back on? What claim is doing more work than the evidence
supports? Be specific, not general.
32. Formatting for the feed
Reformat for LinkedIn: [paste]. Short paragraphs, no paragraph over three
lines, line breaks where a reader would pause. Don't change the words.
33. Remove the LinkedIn voice
This sounds like a LinkedIn post: [paste]. Rewrite it so it sounds like
me explaining something to a colleague. Keep the substance, lose the
performance.
34. Second-opinion pass
Here's my post: [paste]. Don't improve it. Tell me the strongest reason
not to publish it, and whether that reason is fixable.
Repurposing what you already have
35. Long doc to posts
Here's [an article / talk / internal doc]: [paste]. Pull out 5 separate
post ideas. Each must stand alone without the others — no series, no
"part 1 of 5". Give each a one-line summary, not a draft.
36. Talk to post
Here's a transcript of a talk I gave: [paste]. Find the three moments
where I said something the audience reacted to, and draft a post from
the strongest one. Keep my actual phrasing.
37. Comment to post
I wrote this comment on someone else's post: [paste]. Expand it into a
standalone post that doesn't depend on the original for context.
38. Newsletter to LinkedIn
Here's a section from my newsletter: [paste]. Rewrite for LinkedIn — it
has to work for someone who's never read anything else I've written.
Cut the internal references.
39. Q&A mining
Here are questions I answered this month: [paste]. Group them into themes.
For each theme, tell me which single question would make the best post
and why.
40. Slide deck to post
Here's the outline of a deck I presented: [paste]. Write the post version.
A deck can be a list of assertions; a post can't — connect them into an
argument.
41. Update an old post
I posted this a year ago: [paste]. What's changed since, based on what I've
told you? Draft the follow-up. Be explicit about what I got wrong.
42. One idea, three audiences
Idea: [describe]. Write three versions — one for [audience A], one for
[audience B], one for [audience C]. Change the examples and the stakes,
not just the vocabulary.
Series and staying consistent
43. Month of posts from one theme
My theme for the next month is [topic]. Give me 12 post ideas that build
on each other without requiring readers to have seen the previous ones.
One line each.
44. The capture prompt
Here's a raw thought I had today: [paste]. Don't draft anything. Tell me
whether there's a post in it, and if so what's missing before it's
ready to write.
45. Batch first lines
Here are 10 post ideas: [paste]. Write only the opening line for each.
I'll pick which ones to develop.
46. Content audit
Here are my last 15 posts: [paste]. What am I repeating? What do I keep
circling without ever landing on? What haven't I said that I clearly
believe?
47. The pillar check
I want to be known for [expertise]. Here's what I've posted recently:
[paste]. How much of it actually supports that? Be blunt.
48. Idea to backlog
Here are rough notes from the past two weeks: [paste]. Sort them into:
ready to write, needs one more detail, not a post. For the middle group,
tell me exactly which detail is missing.
49. Break the format
My last 10 posts all used this structure: [describe]. Give me 5 structures
I haven't used that fit my material — not generic formats, ones that suit
what I actually write about.
50. The consistency prompt
I can write for 30 minutes a week. Based on these notes: [paste], what's
the realistic posting cadence, and which single idea should I do first?
Don't tell me to post daily.
Three prompts, and what they actually return
The gap between a good prompt and a publishable post is usually one edit. Here's what that looks like.
Prompt 9 (the workhorse), given real notes about a failed migration:
We moved off the managed queue in March. Took 6 weeks, not the 2 we scoped. The thing nobody tells you is the migration isn't the hard part — it's that you're running both systems at once and every bug could be from either.
That's usable almost as-is. It kept the specific numbers and the actual insight. What still needs fixing: it opens with "We moved off" — chronology, not tension. Prompt 21 pulls the specific detail forward, and you get "Six weeks, not the two we scoped."
Prompt 10 (three angles), on the same material, returned a story version that invented a Thursday-night incident that never happened. Models fill narrative gaps when you ask for story structure. That's why prompt 9 says add nothing I didn't say — and why you still read the output rather than trusting it.
Prompt 27 (de-AI it) reliably catches the constructions but not the rhythm. Every sentence comes back the same length. The tell moves rather than disappearing — you'll still want prompt 30 after it.
Where prompts stop working
Everything above works. The limit isn't quality, it's that prompts are stateless.
Every new chat is a cold start. The model that learned your voice in yesterday's conversation doesn't know you today. So you re-paste your samples, re-explain your constraints, re-correct the same three habits. At two posts a month that's fine. At two posts a week it becomes the work.
The best free workaround is a custom GPT with your samples in the system prompt and your style rules in the instructions. It's genuinely good, and if you post occasionally it may be all you need.
It has two limits worth knowing: you maintain it by hand, and it doesn't learn from the edits you make. So the corrections you apply every single time stay corrections, rather than becoming defaults.
That's the problem ThoughtFuel is built around. Writing Styles are trained on your samples once and persist across every draft, and AI memory tracks the themes you keep returning to so suggestions sharpen instead of resetting. Same idea as a custom GPT, without the maintenance — and it improves as you use it.
For the deeper version of this argument — why AI output sounds like AI in the first place, and the manual craft techniques that fix it — see how to write LinkedIn posts that don't sound like AI.
Frequently asked questions
Can ChatGPT write good LinkedIn posts?
It can write good drafts from good input. It cannot write a good post from a topic alone, because it has nothing to work with except the average of every LinkedIn post it has seen — which is exactly what generic AI output sounds like. Give it your notes, your numbers, and your phrasing, and the output improves immediately.
Will LinkedIn penalise AI-written posts?
There is no evidence LinkedIn algorithmically penalises AI-assisted writing, and no announced policy against it. What does hurt is generic writing — posts that could have been written by anyone get no engagement signal, and engagement is what the feed responds to. The risk is being boring, not being detected.
How do I make ChatGPT sound like me?
Paste three to five of your own posts and ask it to study sentence length, how you open, and how formal you are — prompt 11 above does this. The catch is that it forgets between sessions, so you repeat it every time. A saved voice profile solves that permanently.
What's the best prompt for a LinkedIn hook?
Prompt 21 — take the single most specific detail already in your post and make it the first line. It beats asking for a hook from scratch, because the detail is real and yours. Generated hooks tend to over-promise relative to the post.
Should I use a custom GPT for LinkedIn posts?
If you post occasionally, yes — it's the best free option. Put your writing samples in the system prompt and your style rules in the instructions. The limits are that you maintain it manually and it doesn't learn from your edits, so the same corrections recur.
Is ChatGPT enough, or do I need a dedicated tool?
ChatGPT is enough if your bottleneck is drafting and you post a few times a month. A dedicated tool earns its place when the bottleneck is consistency — deciding what to post about, keeping a voice steady across dozens of posts, and shipping without a separate formatting and scheduling step.
Working on the post itself? Our free LinkedIn post formatter adds bold, italic and bullets with a live preview, and the headline generator handles your profile headline.