I use AI every working day, and I build tools with it. So this isn’t a warning from someone standing outside the tent.
But I’ll say what I’m seeing. Inboxes are fuller. Feeds are fuller. Blogs that used to post monthly now post daily. And almost none of it is better than what was there before. A lot of it is the same article, lightly reworded, published by forty businesses in the same week.
The tools didn’t lower the standard. They removed the effort that used to stop people publishing things that weren’t worth publishing.
The one-prompt habit
The lax way to use AI looks like this. Open a chat window, type “write me five social posts about our winter sale”, paste the result, post it. It takes four minutes and it reads like it took four minutes.
What comes back is the average of everything ever written on the subject, because that is all the model had to go on. It doesn’t know your prices, your customers, what you said last month, what your competitors are claiming or which words your brand never uses. So it fills the space with pleasant, confident filler. Readers recognise that voice now, and they scroll past it.
What the good version looks like
The useful work I do with AI has three things the one-prompt habit doesn’t.
- Written rules. How the brand speaks, what it never says, which claims need proof, what a good answer contains. Written down once and used every time, so the output doesn’t depend on who typed the prompt or what mood they were in.
- Real data. The actual product feed, search terms, call notes, sales figures or page content. The model works on your facts and stops inventing its own.
- Approval points. A person checks the output before it reaches a customer or spends money. Not a glance. A real check against the rules.
That is the difference between a toy and a system. A toy gives you a different answer every time. A system gives you a dependable one and tells you when it isn’t sure.
Where it earns its money
In my experience the best uses are rarely the glamorous ones:
- Sorting and labelling thousands of search terms, reviews or enquiries, so a person can read the pattern in an hour.
- Checking product data and ad copy against a set of rules before it goes live.
- Drafting the first version of something a specialist will then properly rewrite.
- Joining up small jobs between systems that would otherwise be done by hand every week.
Boring, repeatable, checkable. That is where the hours come back. “Write my marketing for me” is where the quality goes.
Three questions before you automate anything
- Could I write down what good looks like? If not, the AI can’t hit it either.
- What facts does it need, and am I giving them to it? If it has to guess, it will, and it will sound certain.
- Who checks it, and against what? If the answer is nobody, don’t ship it.
What I’d do
Pick one repeated task that eats a few hours a week. Write the rules for it. Connect it to the real data. Put a person at the end to approve. Run it for a month and measure the time saved and the mistakes caught.
Then do the next one. That is slower than buying a tool that promises to run your marketing. It also works, and it leaves you with something your competitors can’t copy by typing the same prompt.