The five AI questions small teams actually ask
Not the questions from conference stages. The five that come up ten minutes into every session I run with people who just want to get their work done properly.
I run AI sessions for small teams, usually a handful of people around a table, often with one person who is quietly worried about the whole thing. Whatever the industry, the same five questions come up. They are good questions, and none of them are about model architecture.
1. Is it going to make things up?
Yes, sometimes, and it will do it in exactly the same confident tone it uses for the true answers. That is the bit that catches people. My rule in the studio is simple: anything factual, a price, a date, a legal detail, a name, gets checked against a source I trust before it leaves the building. Use it to draft, not to know.
2. Can I paste client information into it?
Assume not, unless you have specifically checked the tool, the account type you are on, and where the data actually goes. Consumer accounts and business accounts often behave very differently on training and retention. This is the single most common way I see a small team create a problem for itself, and it is entirely avoidable, write a one page rule about what can and cannot be pasted, and stick it somewhere people will see it.
3. Who owns what comes out?
This is genuinely unsettled in places, and it varies by country and by tool. For me that is reason enough to keep AI output away from anything you need to protect long term. I will happily use it to explore directions or speed up admin, but a client's logo gets drawn properly, by hand, by me. If you might need to defend ownership of a mark later, you want a clean human trail behind it.
4. Is it going to take our jobs?
It changes parts of them. In my own week it has taken out a lot of typing, first drafts and fiddly research, and none of the judgement. The person who understands the client, the audience and the constraints still makes every decision that matters. If your job is entirely producing generic text at volume, that is worth thinking about honestly. Most jobs are not that.
5. When should we just not use it?
Anything confidential, anything with legal consequences, anything where being wrong is expensive, and anything where the whole point is that a person did it. A condolence note, a hard conversation with a customer, a sensitive apology, do those yourself. People can tell.
One caveat: I am a designer, not a lawyer or a data protection specialist. What I offer is a practical, plain English way in. For a formal position on data protection or contracts, get someone qualified alongside it.