August 15, 2026

Over the last month we've been helping a few of our clients' staff get set up with AI properly. Not the "here's a login, good luck" version. The version where it actually saves time.
And the same thing keeps coming up. Everyone's first week with AI looks identical. You type something in, you get something generic back, you fix it by hand, and you quietly conclude the whole thing is overrated.
The problem usually isn't the tool. It's that you're briefing it like a stranger. Because it is one. Every single time.
An AI tool knows nothing about you until you tell it. Not your business, not your patients or customers, not how you like to sound. So the first fix is the brief: who you are, who it's for, what you want back and what good looks like.
"Write a recall letter" gets you something that sounds like it was written by an insurance company.
"I'm the practice manager at a GP clinic. Write a friendly reminder for patients overdue for their annual care plan review — warm, plain English, no medical jargon, under 100 words" gets you something you'd actually send.
Same tool. The difference is entirely in what you gave it.
Here's the bit nobody tells you: doing that properly takes a minute or two, and you'll be doing the same job again on Thursday. So you shortcut it, the output gets generic again, and you're back where you started.
Typing the same introduction every morning is the thing that kills AI habits in small businesses. Not the technology. The repetition.
The fix is a feature most people have never clicked on. ChatGPT calls them Projects. Claude calls them Projects too. Copilot has something similar. Whatever the name, the idea is the same: a workspace where your instructions are saved once and apply to everything inside it.
Ten minutes of setup looks like this:
Now every conversation inside that project starts already knowing the job. You type "reminder for the ones who missed their review in July, keep it gentle" and it answers like someone who's worked at the front desk for a year.
The temptation is one giant project with everything in it. Resist it. An AI that's been briefed on your patient letters, your job ads, your social posts and your complaint handling all at once starts blending them, and you end up with a recall letter that's weirdly promotional and a complaint reply with a call to action.
One project, one job, one voice. If a task doesn't fit any project, that's a sign it needs its own — or that it's a job the AI shouldn't be doing at all.
Which brings us to the rule from our July post that still stands, and it matters double in a practice: the project knows about your business, but it never gets your patients or clients. Names, Medicare numbers, clinical notes, results, financials, ID documents — none of it goes in, project or no project. The template can be smart; the person in it stays out. If you missed that post, it's on our blog and it's the other half of this one.
Once the project's set up, the day-to-day habit is small: ask for one change at a time, push back when it misses, and keep the good outputs — they become tomorrow's examples.
That's it. That's the whole skill. A one-off setup, a decent brief and the discipline to keep each project in its lane. The staff we set up in August aren't AI experts. They just stopped introducing themselves every morning.
And if you'd rather someone set the lanes up for you — the projects, the instructions, the fences around what it can and can't touch — that's a conversation we're having a lot lately.