You have something to share: an event at your bakery, a new project at work, a local association looking for volunteers, a photo from a hike. But the blank text box is intimidating. What to say, how long, which tone, how to start? Twenty minutes later, the post still isn't written.
An AI assistant can help a lot here, as long as you stay the author. It can turn your rough notes into several drafts, shorten a post that's too long, or adapt it from one platform to another. This guide shows you how to write social media posts with AI, step by step, with ChatGPT, Claude, Gemini or any other assistant. If you've never used one, start with our guide on how to use ChatGPT as a beginner.
What the AI won't do for you
Let's be honest from the start, because a lot of promises float around on this topic:
- It won't make you go viral. No prompt guarantees reach, likes or followers. Platforms decide what they show, and what works depends on your audience, your timing and a fair amount of luck.
- It doesn't know your story. If you give it nothing, it will fill the gaps with generic phrases, or worse, invent details: a number, an anecdote, a customer who never existed.
- It doesn't know the facts of your situation. Dates, prices, addresses, names: if you don't give them, it may make them up with a confident tone (that's what we call a hallucination).
The AI is a writing assistant. The message, the facts and the responsibility for what you publish stay yours.
Step 1: write down your real message first
Before opening the assistant, answer three questions in a few words:
- What happened, or what do you want to say? "We're opening on Sundays from next month."
- Who is it for? "Our regular customers and people in the neighborhood."
- What should people do after reading? "Come by, or share with a friend."
Then list the facts the post must contain: dates, times, place, name of the event, link. These are the details you'll check at the end.
This takes two minutes and changes everything. A post built on a real message and real facts sounds like you. A post built on "write something about my bakery" sounds like everyone else.
Step 2: give the AI your voice
The biggest complaint about AI-written posts is that they all sound the same: enthusiastic, full of exclamation marks and buzzwords. The fix is to show the assistant how you write.
Find two or three posts or messages you wrote yourself and liked. Then:
Here are three short texts I wrote myself:
[text 1]
[text 2]
[text 3]
Describe my writing style in 5 bullet points: sentence length, tone, vocabulary, use of emojis, how I start and end.
Then use this style for everything you write for me in this conversation.
Read the description. If it's wrong ("you use lots of emojis" when you don't), correct it. If you post regularly, you can save the corrected description in your custom instructions so you don't have to repeat it.
You can also add a list of things you never want:
Never use: exclamation marks in every sentence, words like "game-changer" or "exciting journey",
more than two emojis, or hashtags in the middle of sentences.
Step 3: ask for several drafts
Here's a prompt you can copy and adapt. The bracketed parts are yours to fill in:
Write a post for [Instagram] for [our neighborhood bakery].
Message: [we're now open on Sundays, from 8 am to 1 pm, starting next month].
Audience: [regular customers and neighbors].
Goal: [people come by on a Sunday and tell a friend].
Facts to include exactly as written: [opening hours, the address: write ADDRESS where it goes, I'll add it].
Tone: [warm, simple, a bit of humor, no exaggeration].
Length: [5 lines maximum].
Give me 3 different versions: one short, one with a small story, one with a question for readers.
Don't invent any detail I didn't give you.
Two lines in this prompt do a lot of work:
- "Don't invent any detail I didn't give you" reduces the risk of a fake anecdote or a made-up number.
- Placeholders like ADDRESS keep sensitive or precise information out of the chat, and force you to add it yourself, correctly.
Then pick the version you prefer, or mix them: "take the opening of version 2 and the ending of version 3".
Step 4: edit it like it's yours (because it is)
The draft is a starting point. Read it out loud: does it sound like something you'd say? Common fixes:
- Too enthusiastic: "make it calmer, remove the exclamation marks".
- Too long: "cut it to three lines, keep the hours and the invitation".
- Too vague: replace generic phrases with a real detail only you know ("the first croissants come out at 7:30").
- A false note: if a sentence claims something that isn't true, or exaggerates, delete it. Even a small exaggeration can cost trust.
Don't hesitate to rewrite parts by hand. The best posts are often AI drafts with a human sentence or two added.
Step 5: adapt it to each platform
Each platform has its habits: a professional network calls for a different tone than a photo app, and some formats are very short. Instead of writing three posts, adapt one:
Here is my final post for [Instagram]:
[paste the post]
Adapt it for [LinkedIn] and for [a short post on another network].
Keep the same facts and my style. For each version, explain in one line what you changed and why.
Platform rules and features change often. If the assistant tells you something precise about a platform (a character limit, how an algorithm works, a new feature), don't take it as a fact: check it on the platform itself.
Step 6: check everything before you publish
This is the step that protects your reputation. Before posting, go through this list:
- Facts: every date, time, price, address, name and link matches your notes. Click the link.
- Claims: no "the best in town", "100% natural" or "proven" unless you can back it up. For anything factual, see our method to fact-check an AI answer.
- People: you have permission to mention or tag them, and to show them in a photo.
- Images: if you use an AI-generated image, make sure it doesn't mislead (a dish you don't sell, a place that doesn't look like that). Our guide on how to generate an image with AI covers how to review a visual.
- Rules: if you post for an employer or a client, follow their guidelines. Some platforms and organizations also have rules about labeling AI-generated content: check the ones that apply to you.
About honesty: should you say AI helped?
There's no single rule for every situation, but a simple principle works: don't let readers believe something false. Using an assistant to fix your spelling or tighten a sentence is like using a spell checker. Presenting an AI-generated image as a real photo, inventing a customer testimonial or posting an "experience" you never had is misleading, whatever tool you use.
If a post is largely written or generated by AI, or if your employer, your clients or the platform asks for it, say so. Being transparent costs little and protects the trust your readers give you.
Common pitfalls
- Starting from nothing: "write a post about my business" produces generic text. Start with your message and your facts.
- Publishing the first draft: read it, edit it, make it yours.
- Letting invented details slip through: a fake number or anecdote in a post is your responsibility, not the AI's.
- Chasing hacks: no prompt can promise reach or followers. Focus on saying something useful, clearly.
- Pasting private information: customer names, private messages, internal figures don't belong in a chatbot. See our guide on ChatGPT and personal data.
- The same post everywhere: adapt the tone and length to each platform.
Mini exercise: your first AI-assisted post in 15 minutes
- Your message (3 min): pick something real you could share. Write what happened, who it's for, what readers should do, and the facts to include.
- Your voice (3 min): paste two texts you wrote yourself and ask for a description of your style. Correct it if needed.
- Drafts (3 min): use the prompt from step 3 to get three versions.
- Edit (3 min): choose one, read it out loud, add one real detail and remove anything that sounds fake.
- Check (3 min): go through the checklist from step 6. Circle every fact and compare it with your notes.
What to notice: how much the post improves when you give a real message and your own style, and how many small inventions you catch in the check. That last reflex is the one to keep, whether you publish or not.
Going further
To understand why an assistant writes so smoothly and can still invent a detail with total confidence, play the free chapter "Light the Lantern" in your browser, no account needed: play the chapter. To practice writing clear requests, try the Lab, and to train your eye for confident mistakes, play a round of Hallucination Hunt.
For other everyday writing, see our guide on how to write an email with AI. And if a term trips you up, the glossary is there.
Frequently asked questions
Can ChatGPT write my social media posts?
It can write good drafts if you give it your real message, the facts to include, your audience and examples of your style. Read and edit every draft, check every fact, and only publish what you would be happy to sign.
How do I make AI posts sound less generic?
Show the assistant two or three texts you wrote yourself and ask it to describe your style, list the words and habits you don't want, and add at least one real detail only you know. Generic input gives generic posts.
Should I disclose that I used AI for a post?
Follow the rules of your employer, your clients and the platform first. Beyond that, the key is not to mislead: light help with wording is like a spell checker, but an AI image presented as a real photo or an invented testimonial is misleading. When in doubt, be transparent.
Can AI help me get more followers?
No tool can promise that. The AI can help you write clearer posts more quickly, and adapt them to each platform. Reach depends on your audience, the platform and many factors nobody fully controls.