Your staff are already using ChatGPT. Now what?

Nobody rolled it out

In a lot of small businesses, nobody decided to adopt AI. Someone tried ChatGPT on a tricky email, it helped, and now half the team uses it for something: replies to customers, quotes, job adverts, social posts. You might be one of them.

That's not a bad place to be. Your team has already done the hardest part of adopting any new tool, which is finding where it actually helps. The question is what happens next.

The real risk is inconsistency

The obvious worry is what people paste into it. That matters, and it deserves its own conversation about which tools and settings you're comfortable with. This piece is about a quieter problem.

Every person is giving the AI their own version of your business. One person's refund rule is out of date. Another tells it you offer next-day delivery, because you did last year. A third doesn't mention the discount for regulars at all. The AI believes each of them, so customers get different answers depending on who replied. That's why two teams get different results from the same AI, happening inside one small team.

Why banning it doesn't work

People use it because it saves them time, and a ban doesn't make that time come back. Usually it just moves the AI out of sight, onto personal phones and accounts, where you can't see what it's being told or help make it better.

You also lose what your team has learned. The person who's been using it for quotes for six months knows things about where it helps and where it goes wrong that nobody else does.

Why a policy isn't enough

An AI policy says what people can and can't do with it. That's worth having. But it governs behaviour, not information, so it doesn't make a single answer more accurate. Nothing in a policy tells the AI that custom orders can't be refunded, or which customers get the regulars' discount.

The accuracy problem is about what the AI knows, and the fix has to be too.

What to put in place instead

1. Find out what people actually use it for. Ask openly, without judgement. You'll learn where it's saving real time.
2. Collect what they keep telling it. The background people type before a request is the paragraph everyone types, and it's a list of what your business never wrote down.
3. Agree one version of the facts. Where people's versions differ, decide which is true, and give each fact an owner who updates it when things change.
4. Do one workflow properly. Shared instructions, the agreed facts where the tool can read them, and someone checking the output for a few weeks before you trust it.

Then move on to the next workflow.

Treat it as a head start

Most businesses trying to adopt AI are guessing where it will help. Yours already knows, because your team found out by using it. The job now is to give them one agreed version of the business to work from, instead of one each.

If you'd like help doing that for the workflow that matters most, that's where the audit starts.