Should I use AI to write my social media posts?

We build these tools, so treat this with appropriate suspicion. The honest answer is that AI is very good at one part of the job, actively bad at another, and the difference is easy to state — which is more than most people selling it will tell you.

What it is genuinely good at

The blank page. Most small businesses do not stop posting because they cannot write; they stop because starting from nothing, at nine at night, after a full day, is miserable. A draft to react to removes almost all of that friction, and reacting is far easier than creating.

It is also good at adaptation. Taking one idea and reshaping it for a different platform, length or format is mechanical work that nobody enjoys and nobody notices you doing well. Handing that over costs you nothing.

And it is good at the things you already know but never write down: the answers you give on the phone every week, the objections you handle in person, the details customers always ask about. Those make excellent posts and they never get written because they feel too obvious to you.

What it is bad at, and will stay bad at

Knowing anything about your business. It does not know that your Tuesday trade is quiet, that one supplier let you down last month, or that your customers care more about parking than provenance. Given nothing specific, it produces the average of everything ever written about your industry — which is exactly the flat, weightless register people have learned to scroll past.

Having an opinion. Generated copy defaults to pleasant and non-committal. Actual persuasion involves saying something someone could disagree with. That has to come from you.

Judging whether a post is any good. It cannot tell you it has produced something bland, because bland is its centre of gravity. You remain the editor. Any workflow that removes you from that step will publish embarrassing things eventually.

The thing that decides which you get

Almost the entire difference between useful AI output and obvious slop is how much specific material it has been given. Not prompt cleverness — specifics.

Given "write a post about my cafe", any model produces the same beige paragraph. Given your actual prices, the fact you only do the sourdough on Thursdays, the words you refuse to use, your service area, and the three objections customers actually raise, it produces something only your business could have said.

This is why the setup step matters more than the writing step, and why skipping it is the real cause of "AI sounds like AI". The complaint is usually accurate and usually self-inflicted.

How to tell if it is working

A test that costs nothing: read the draft and ask whether a competitor could publish it with the name swapped. If yes, it is not finished — and no amount of regeneration fixes that, because the missing ingredient is information, not attempts. Add the specific that is absent and try once more.

A second test: would you say this sentence to a customer standing in front of you? Generated copy drifts towards a register nobody uses out loud. Reading it aloud catches that instantly.

Where it genuinely should not be used

• Anything with a legal or safety claim. Regulated trades, medical claims, financial promises. If getting it wrong has consequences, write it yourself.
• Replies to an unhappy customer. A person who feels ignored can tell, and an automated apology reads worse than a slow human one.
• Anything asserting a fact you have not checked. Fluent and wrong is the failure mode.
• Your founding story, if you want anyone to believe it.

The bottom line

Use it to beat the blank page and to adapt one idea across platforms. Do not use it to decide what your business has to say. The businesses getting good results are not the ones with better prompts — they are the ones who spent twenty minutes telling the tool what makes them different, and who still read everything before it goes out. If the blank page is the problem, our caption templates solve it without any AI at all.