Nearly every owner I speak to tells the same story. Someone in the business tried ChatGPT (or Claude, or Copilot) last year. It was impressive for an afternoon. Then everyone quietly went back to doing things by hand.
The usual conclusion is “it's not really for businesses like ours” or “the team isn't technical enough”. Both are wrong, and the real reason is more useful, because it's fixable.
What a blank chat box actually asks of you
When you open a fresh chat and type “write a reply to this customer”, the tool knows nothing. Not your prices, your tone, your opening hours, the fact that you never say “reach out”, or that the customer has already been told twice that the part is on back-order.
So it gives you something generic. Polite, plausible, and wrong in six small ways. You spend four minutes fixing it, decide you could have written it in three, and you're right. That's the experience most people had, and it's a completely rational reason to stop.
The mistake isn't using AI. It's using it with no context, every time, and expecting it to read your mind.
What the people who kept using it did differently
Everyone I've met who uses these tools daily did one thing the rest didn't: they front-loaded the context once, so they never had to explain the business again.
In practice that means five things.
- A written brief. Who the business is, who the customers are, how you sound, what you never say, what you never promise. Two or three hundred words. It goes in as custom instructions and it's read every single time.
- Separate workspaces for separate kinds of work. Claude calls them Projects. One for customer replies, one for quotes, one for marketing, one for the internal stuff. Each has its own instructions and its own documents.
- The actual documents loaded. Price list. Terms. FAQs. Three past quotes you were happy with. This is what stops it inventing prices.
- Saved prompts for the jobs you do every week. Not clever prompts. Boring, specific ones: “quote from my notes”, “chase this invoice, reminder three”. Kept somewhere the team already looks.
- One real task per person, done properly. Not a demo. Their own email, their own quote, with someone standing next to them the first time.
Do those five and the tool stops being an effort. It already knows the business, so using it is faster than not using it. That's the only reason any habit sticks.
Why it doesn't happen on its own
Because it's a project, and small businesses don't have a spare Tuesday. The owner is the only person who knows the business well enough to write the brief, and the owner is the busiest person in the building. So it sits on the list next to “update the website” for eighteen months.
There's also a quieter reason: nobody wants to be the one who spent a day on AI and had nothing to show for it. So nobody starts.
How to do it in a day yourself
If you want to do this without help, here's the order that works.
- Morning, first hour: write the brief. Use the structure above. Paste in three real emails you were proud of and say “sound like this”.
- Second hour: create four projects, paste the brief into each, upload the relevant documents. Test each with a real example until the tone is right.
- After lunch: get everyone to do one real task from their own week. Watch. Fix the brief when it gets something wrong.
- Last hour: write down the ten prompts people actually used, name them in plain English, and put them in a shared doc someone owns.
That's a genuine day. Most owners I know won't find it, which is fine, because it's also exactly what I do for £500 and you get a written roadmap at the end. Either way, the point stands: the tool was never the problem. The setup was.
If you want to see what the finished setup looks like before deciding, there's a sixty-second simulated version here.