LinkedIn content prompts: 9 templates to post without sounding like AI
An effective prompt for LinkedIn content does not say "write a post about AI": it gives the tool your real experience, a clear stance, your audience and the phrasing you refuse to use. The AI structures and tightens; the idea, the example and the opinion stay yours. That is also what keeps you out of the mass of generic posts LinkedIn now recommends less and less.
SME owners, marketing managers, startup founders: many already use ChatGPT, Claude or Gemini for their posts, and many end up with the same result. Clean, well-paced posts that trigger nothing. This article sets out a three-step method (a context sheet you fill in once, nine prompts for the formats that work in B2B, a review pass to get your own voice back) along with the data limits to respect. For the general principles, the prompt engineering guide for SMEs remains the starting point of this series.
Why your AI-written LinkedIn posts all sound the same
Because a model without context writes the average of what it has read, and the average LinkedIn post is an opinion with no evidence. An LLM knows nothing about your last project, the question a client asked you on Tuesday or the figure that surprised you in your reporting. If you do not give it those elements, it replaces them with generalities.
Anyone who scrolls a feed knows the signs:
- a "big reveal" hook ("Nobody talks about this, but...") followed by an idea everyone already agrees with;
- the systematic "it's not A, it's B" contrast;
- lists of three adjectives, one-line sentences stacked to fake a rhythm;
- an open question at the end ("What do you think?") that asks nothing specific;
- no name, no figure, no place, no date.
Switching tools does not fix this. Our ChatGPT vs Claude vs Gemini comparison shows all three write correct prose; style differences exist, but they weigh little against the quality of the material you provide. The lever is the prompt, and above all what you put in it.
What LinkedIn does with content that reads like AI
Since spring 2026, LinkedIn says it removes from its recommendations posts that repeat existing ideas without adding anything, as well as comments produced in bulk by automation tools. Those posts stay visible to your followers and connections, but they are no longer pushed beyond them. Among the patterns the platform mentions is precisely the binary phrasing "the problem isn't A, the problem is B".
A second change, described by LinkedIn's engineering team in November 2026: the feed now relies on language models to match the topic of a post with the professional profile of its author (role, industry, skills, experience). A plant manager writing about industrial maintenance is more likely to be shown to readers interested in that subject than with a post on "5 leadership lessons" unrelated to their background.
Two practical consequences for your prompts:
- Originality becomes a distribution criterion, not just a matter of style. A post that brings a field observation, an internal figure or a mistake you made gets through; a rewording of a fashionable idea does not.
- Topical consistency matters. Writing about your area of expertise, in the vocabulary of your trade, helps your visibility more than chasing every trending topic.
These mechanisms change quickly and LinkedIn does not publish its full criteria. Keep the direction in mind rather than fixed recipes.
Before the first prompt: the context sheet
A context sheet of about fifteen lines, filled in once, improves your posts more than any prompt wording. It gives the model what it cannot guess. Put it in your tool's custom instructions or in a dedicated project, as explained in our article on system prompt vs user prompt.
| Section | What you write | Example |
|---|---|---|
| Who is speaking | Role, company, years in the trade | "Managing partner of a 14-person accounting firm in Lyon, 20 years in the profession." |
| For whom | Target reader, their level, their concerns | "Owners of small service businesses, not comfortable with numbers, who want to plan their cash flow." |
| Topics | 3 or 4 themes where you have legitimacy | "Cash flow, e-invoicing, hiring in accounting firms." |
| Positions | What you think and others dispute | "An annual forecast is useless without a monthly review." |
| Material | Where your examples come from | "Client questions, mistakes seen in file reviews, anonymised internal figures." |
| Tone | How you write, with one or two real posts as examples | "Direct, short sentences, some humour, never unexplained accounting jargon." |
| Banned | Phrasing and formats you refuse | "No emojis, no 'Nobody talks about this' hook, no 'it's not A, it's B', no generic closing question." |
The two real example posts are the most valuable line of the sheet. A model imitates a style it is shown far better than a style it is told about: that is the principle of few-shot prompting. If you have not posted anything yet, use two emails or internal messages you wrote yourself.
The "Positions" line is often left empty. Without it, the model produces consensual text. For a general method to build an instruction, see our article on the anatomy of a good prompt. The reader profile is worth formalising as a persona.
9 prompts for your LinkedIn content
Copy each template as is, after your context sheet, then fill in the brackets with true information. An empty bracket is an invitation to invent.
1. Turn a field observation into a post
Here is a real situation from this week: [describe in 3 to 5 lines the context, what happened, what you concluded]. Write a LinkedIn post of 150 to 220 words that opens directly on the concrete fact, then draws one single lesson my target reader can apply. Do not add any detail I did not give you. End with a specific question tied to the situation, not a general one.
2. Defend a reasoned position
My position: [one sentence]. What it rests on: [2 or 3 facts, examples or figures from my business]. The most common objection: [objection]. Write a post that sets out the position, answers the objection honestly and states in which case I could be wrong. 200 words maximum. No preachy tone, no binary contrast formula.
3. Share an internal figure and what it reveals
Data point: [figure, period, scope, internal source]. What surprised me: [explanation]. Suggest 3 different post angles based on this figure, one sentence each, stating for each which reader it interests most. I will choose, then you will write the post on the chosen angle.
This two-step prompt stops the model from picking the most obvious interpretation for you.
4. Tell a mistake and what it changed
Mistake made: [what happened, real cost in time or money if you can say it]. What we do differently since: [concrete change]. Write a sober post, without self-flagellation or a heavy closing moral. 180 words maximum.
5. Answer a client question
Question a client asked this week: "[question as worded]". My usual answer: [a few lines, even rough]. Write a post that uses the question as the hook and gives the answer in 3 concrete steps. No unexplained jargon.
6. Summarise an event or a read, with an opinion
I attended [event] or read [document]. Raw notes: [paste notes]. Extract the 2 ideas most useful to my reader and add my opinion on each: [your opinion]. Flag in brackets any statement in my notes you cannot clearly attribute.
7. Plan a carousel
Topic: [topic]. Reader: [reminder of the reader]. Single takeaway: [sentence]. Suggest a carousel outline of 7 to 9 slides: a title of 8 words maximum and 25 words of text maximum per slide. Slide 2 states the problem with a concrete example; the last slide gives one action to take today.
8. Turn a blog article into three posts
Here is an article published on our website: [paste text]. Suggest 3 independent LinkedIn posts, each built on a different idea from the article and readable without having read it. For each, give the hook and the source paragraph. Do not reuse any sentence word for word.
It is the simplest way to feed a content marketing effort without multiplying writing hours.
9. Prepare useful comments, not automated ones
Here is a post by [role, industry]: [paste post]. My experience on this topic: [2 lines]. Suggest 2 possible comments of 40 to 70 words that bring something missing from the post (example, nuance, figure). No congratulation formula, no rewording of the post.
This last prompt is for preparing, not for posting in bulk. Comments posted en masse by automated tools are exactly what LinkedIn says it filters out.
The same principles apply to your other sales writing; our sales email prompts apply this logic to prospecting.
Getting your voice back: the review pass
The AI's first draft is a draft; the published post needs a two-step review, first by the tool, then by you. Start by asking the model to critique its text against explicit criteria, then rewrite yourself whatever does not sound like you.
A critique prompt that works:
Read this post as a [persona] reader who sees 200 posts a day. For each of these criteria, say whether the post meets it, quoting the sentence at issue: 1) the hook contains a concrete fact; 2) the post gives information the reader will not find elsewhere; 3) no phrasing from the banned list in my context sheet; 4) no statement I did not provide; 5) the ending asks for something specific. Then suggest a corrected version of the failing passages only.
Having the model reason criterion by criterion, rather than asking straight for "a better version", gives more targeted corrections. It is chain-of-thought prompting applied to review.
Criterion 4 is not a detail. A model short of material readily adds a "plausible" figure or an untraceable study. On LinkedIn, that figure goes out under your name. The mechanism is described in our article on AI hallucinations and in the glossary entry on hallucination.
Last check, the simplest one: read the post aloud. If a sentence would never come out of your mouth in a client meeting, rewrite it.
What should stay out of the prompt
A LinkedIn post is public, but the prompt that prepares it may contain information that is not: client names, unpublished figures, employee situations. That is where the risk lies, more than in the publication itself.
In practice:
- No problem: your ideas, your reading notes, aggregated and anonymised figures, content already published.
- With a properly configured business account: internal meeting notes, unpublished business figures you plan to round or anonymise.
- Avoid: a client's name without their consent, anything that identifies an employee, health or personal situation data, information covered by a confidentiality commitment.
Check that reuse of conversations for training is turned off in your tool, and write these rules into your AI policy. Our article on AI and data confidentiality details the settings.
On the regulatory side, Article 50 of Regulation (EU) 2024/1689 (the AI Act), applicable since 2 August 2026, requires disclosure in specific cases, notably deepfakes and text published to inform the public on matters of public interest without human review or editorial responsibility. A company post reviewed and published by an identified person does not fall into that second case. A realistic AI-generated image or video showing a real person or place, however, may fall into the first. The details are in our articles on the AI Act and customer information and on workplace rules for consumer generative AI. This reminder does not replace advice from a lawyer.
Organising production without losing authenticity
AI saves time on formatting, not on gathering material: that is where your routine needs to focus. An executive who posts once a week needs a steady source of observations, not a topic generator.
A simple setup for an SME:
- A material notebook. A shared note where you add, through the week, client questions, striking figures, situations that made you react. Five minutes a day is enough.
- A weekly 45-minute writing session. Pick an entry from the notebook, use the matching prompt, review, rewrite by hand.
- A library of validated prompts. When a prompt gives you a post you are happy with, keep it with its context sheet. ChatGPT, Claude and Gemini all let you store these instructions in projects or custom assistants; our pages on ChatGPT use cases for SMEs, Claude for SMEs and Gemini for SMEs give examples. In a Microsoft environment, see our Microsoft 365 Copilot beginner's guide.
- Honest measurement. Look at comments from people in your target audience and the contacts that follow, rather than the number of reactions. A post seen by 800 people in the right industry beats a viral post outside your market.
If you publish from a large body of internal documents (product sheets, studies, reports), a document retrieval approach can take over from plain copy-paste; see RAG vs prompting.
Frequently asked questions
What is the best prompt to write a LinkedIn post?
There is no universal prompt. The most effective combines a context sheet (who you are, who you write for, your positions, your banned phrasing), a real fact to tell, and length and tone constraints. Without a real fact, even the best prompt produces generic text.
Does LinkedIn penalise posts written with AI?
LinkedIn says it does not target the tool but content that adds nothing new or repeats formulaic writing patterns. Such posts remain visible to your network but are recommended less beyond it. A post prepared with AI from your own experience, then rewritten by you, has no reason to be treated differently from a handwritten one.
ChatGPT, Claude or Gemini: which one for LinkedIn?
The one you already use with a business account. Claude is often seen as more natural in tone, ChatGPT as more versatile, while Gemini and Copilot integrate with your office documents. The context sheet and the review matter more than the choice of model.
How long does it take to write a post with AI?
Allow 20 to 30 minutes once the context sheet is in place: a few minutes to choose the topic, five for the prompt and first draft, the rest for review and rewriting. The gain over a blank page is real, but it disappears if you skip the review.
Can you publish LinkedIn posts automatically with AI?
Technically yes, with an automation tool. We advise against it for an executive's personal account: publishing without review multiplies factual errors, and posts or comments produced in bulk are exactly what LinkedIn is trying to filter out.
Do you have to say a post was written with AI help?
For a text post reviewed and published under your responsibility, no general obligation requires it today. Disclosure is still required for some content, in particular realistic AI-generated images or videos, under Article 50 of the AI Act.
How GrowthPerf can help
The prompts that work on LinkedIn are those built on your own experience, and that material cannot be delegated. GrowthPerf, a Qualiopi-certified training organisation, helps executives and marketing teams in SMEs, startups and nonprofits build their context sheet, their prompt library and their review routine, on their own topics.
The operational AI training includes writing workshops applied to your real content. For a course focused on prompting techniques, see our page on prompt engineering training. In France, these courses can be funded by your OPCO depending on your entitlements.
To go further on method, return to the complete prompt engineering guide for SMEs.
Already posting with AI but your posts bring no leads? Book a free 30-minute audit: we review your latest posts and prompts, and suggest a LinkedIn prompt pack tailored to your business.