AI is very good at producing social media posts and very bad at producing good ones. Both things are true, and the gap between them is why so many accounts got blander in the last two years.
Why generated posts underperform
Social platforms reward one thing above all: a reason to stop scrolling. That reason is almost always specificity — a real number, a real mistake, a real opinion someone might disagree with.
Generated posts optimise for the opposite. They produce the average of everything ever written on the topic, which is by definition the least surprising version of it. Add three emoji and a "What do you think? 👇" and you have something nobody has any reason to read.
The model can write like everyone. It cannot write like you, unless you give it something only you have.
There is also a saturation problem. If a tool produces the most probable phrasing, and thousands of accounts use the same tools on the same topics, the output converges. Anything that sounds like everyone else is invisible.
The parts it genuinely helps with
Reformatting, not creating
You wrote a blog post, gave a talk, had a good conversation with a customer. Turning that into six posts across three platforms is mechanical work, and it is where the actual hours go.
The substance came from you. It is changing the shape.
Getting past the blank post
Ask for ten angles on a thing you want to say. Hate nine. The tenth reminds you what you actually meant. You then write it yourself.
Variations for testing
Five versions of one hook, so you can see what lands. Genuinely useful and genuinely tedious to do by hand.
The unglamorous accessibility work
Alt text for images. Captions for video. Descriptions that make your content usable by people who cannot see or hear it. This is work that should always be done and often is not, and AI removes every excuse.
Replying to the obvious
The same four questions in your comments every week. Draft answers, edited by you, sent by you.
How to keep your voice
Feed it yours. Paste in your ten best-performing posts and say "in this style". Examples work far better than adjectives like "casual" or "punchy".
Give it the specific thing. Not "write a post about productivity" but "write a post about how I lost a client because I over-promised a deadline in March". The second has something in it. The first has nothing.
Cut the AI seasoning. Delete "In today's fast-paced world", "Let's dive in", "Game-changer", "It's not just X, it's Y", and any sentence that could appear on any account in your industry.
Write the last line yourself. The final sentence carries the opinion, and the opinion is the thing that gets a response.
Read it aloud. If you would not say it, do not post it. This one test catches most of it.
What not to hand over
- Anything time-sensitive or newsworthy. It does not know what happened this morning and will confidently imply it does.
- Community management. People can tell, and being answered by a machine in a space you asked them to join is a fast way to lose them.
- Your actual opinions. They are the product. Outsourcing them leaves you with a feed that says nothing.
- Anything with a number in it that you have not checked yourself.
The honest summary
Use it for the mechanical half — reformatting, variations, alt text, first drafts of repeat replies — and keep the half that requires having something to say.
Accounts that do the opposite get more efficient at producing content nobody reads, which is not a win.
If you want the general framing, what AI can actually do for you covers where these tools are strong, and how to write a prompt that actually works is most of the difference between bland output and usable output.