Yesterday it took me five drafts to get one post out. The writing was fine. Nothing in it was wrong. It was also interchangeable with a thousand other posts about AI, and I knew it before I finished reading it back.
The fifth one shipped. What changed was not the writing. I named the person I disagreed with.
That is the real problem with AI-drafted content, and it is not the one most service business owners think they have. Your posts do not sound wrong. They sound borrowed.
The short version
- AI writes toward the middle of everything it has read. The middle is where every other business is already standing.
- Clean, correct and forgettable is the normal output, not a sign you prompted it badly.
- Three checks catch a borrowed post before you publish: is it standing on something real, does it take a side, and could ten other people have posted it.
- The repair is almost never more editing. It is putting something in that only you could have put in.
Why does AI-written content all sound the same?
An AI model has read an enormous amount of business writing and learned what the average piece looks like. Ask it for a post about AI and small business, and it gives you the average of everything it has seen on that subject.
The average is not wrong. That's what makes this hard to spot. Every sentence is defensible, the structure is sensible, nothing is inaccurate. You are reading competent work with nobody behind it.
There is a second pull working on you at the same time. Models are trained to be agreeable, so they avoid taking a side that might annoy someone. Left alone, they hedge. A hedged opinion reads as balance, and balance reads as nothing.
Put the two together and you get the standard output. Tidy and correct, with nothing in it to push against. Then you post it, nobody replies, and you assume you need to post more often. That's the cost, and it's a slow one. Months of work that was never going to land, and no obvious fault to point at. This is the same trap as most of the advice aimed at small businesses using AI: technically fine, built for nobody in particular.
Check one: is it standing on something real
Read your draft and find the thing it rests on. A job you did. A conversation you had. A number you can produce. A date. Something a person could check.
If you can't point at it, the post is standing on air. A post with no anchor reads safe no matter how you word it, because there is nothing in it that could only have come from your week.
This is the check that fails most often, and it fails early. It usually fails at the idea, not at the writing. If you sat down with nothing that happened and asked AI for a post anyway, no amount of rewriting puts an event back in.
The repair is not a better prompt. It's going and getting the material: what you fixed on Tuesday, what a client said that surprised you, what you got wrong last month and changed.
Check two: does it take a side
Find the sentence in your draft that somebody could disagree with. Not disagree in a shouting way. Disagree in a "no, I think the opposite" way.
If there isn't one, you have written a summary. Summaries are read by nobody because there is nothing to do with them. Agreement is silence. A reader who nods and scrolls has given you exactly as much as a reader who never saw it.
Taking a side does not mean picking a fight. It means being specific about what you think is a bad idea, and saying so plainly. "Most owners buy the tool before they know the job" is a position. "There are many considerations when choosing AI tools" is the middle of the road with a hedge on top.
One warning worth having. Hit ideas, choices, habits and bad advice. Never the reader's intelligence or worth. The point is to be worth arguing with, not to be unpleasant.
Check three: could ten other people have posted it
Cover your name and your logo. Read it back. Could ten other businesses in your line of work have published that exact post?
If yes, you have written the average. The parts that survive this check are always the specific ones: the real numbers, the tool you open every day, the client situation, the thing you tried that did not work.
This is where most people edit in the wrong direction. Told a draft feels generic, they smooth it further, cut the odd detail, and make it worse. Generic is not fixed by polish. It's fixed by putting something odd and true back in.
Running the three checks on your own drafts
The checks take about two minutes and they run in this order, because each one can send you back further than the last.
- Anchor. Name the real thing this post stands on. No anchor, do not write the post. Go and find one.
- Side. Underline the sentence someone could argue with. If there isn't one, the post is a summary. Decide what you think and say it.
- Cover the name. Read it as a stranger. If ten others could have posted it, add back the specific detail you took out for being too small.
Two things worth knowing before you run them. First, the checks are quick to apply and slow to satisfy, and that is correct. Most of the work happens before drafting, in noticing what is worth writing about. Second, asking AI to check its own draft against these three will get you agreement unless you make it hostile, which is a separate discipline I have written about in getting AI to find real problems in your work.
You can also run the checks backwards on someone else's content. Read three posts in your feed and ask which ones have an anchor. It's a fast way to see how much of the feed is average, and how low the bar for standing out really is.
Common questions
Does this mean I should stop drafting with AI? No. It means the AI draft is the first pass, not the finished thing. Use it to get the shape down quickly, then run the three checks and put yourself back into it. The problem is publishing the first pass.
What if nothing interesting happened this week? Something did, it just didn't feel like content at the time. The job that took twice as long, or the question a client asked you twice in the same month. Most service business owners have plenty of anchors and no habit of catching them.
Won't taking a side lose me clients? It will lose you the wrong ones sooner, which isn't the same thing. The people who agree with a flat position tend to be the people you want to work with, and they can't find you while you sound like everyone else.
Isn't this just about writing better? Not really. Better writing just makes a borrowed post nicer to read. It stays borrowed. The three checks are about what goes in, not how it is worded, which is why they work even if writing is not your strength. The same order applies to most AI work: fix what feeds the tool before you polish what comes out.
Where this leaves you
The shelf is full of good, polished AI content. Every piece on it is saying something valid in the same voice, and all of it is getting scrolled past by the people it was written for.
Take the last thing you posted and run the three checks on it. If it fails all three, that is not a writing problem, it's a material problem, and it's fixable this week. Finding where the useful material in a business is going missing before it ever reaches a draft is the mapping work I do in an AI Workflow Architecture. If you want a hand with yours, book a Clarity Session and we will go through it together.
