Show notes
I have lost count of the conversations that start the same way at the moment. Someone with a budget and a deadline leans in and says the business needs AI. Where, exactly? Everywhere. In the product, on the site, in the onboarding flow.
It's not that they've no idea what they want. They usually arrive with something in mind. The trouble is it's a half-baked solution to a problem that may or may not exist.
Push back and you're the difficult one, the blocker, the person who doesn't get it. Go along with it and you build a chatbot nobody asked for. Neither ending is much fun.
So I stopped arguing. Now I do something that works far better. I send the whole thing to the users.
Don't win the argument. Sidestep it.
When someone hands me a shaky AI idea, I don't tell them it's shaky. That's a quick way to make an enemy and lose. Instead I say something like this.
"That's a really interesting idea. I think there's something in it. Let me go away and test it with a few users so we build it in the right way."
Often that's enough. You sound keen, not obstructive, and the stakeholder feels heard. You've quietly moved the decision out of a meeting room, where the loudest voice wins, and handed it to the only people whose opinion really counts.
But sometimes they won't budge. They're so sure they've got it right that testing feels like a waste of time. When that happens, I fall back on one of two tactics.
The first is to ask questions. I throw a lot of very specific ones at them about how the thing should work. What happens in this case? What about that one? Before long they start to struggle, and that's the moment to step in. It will be quicker to ask a few users than to guess our way through all of this.
The second is to talk about risk. If we build this without testing, there's a real chance we go down the wrong path and waste the budget. So I ask whether they're happy to own that risk. In my experience, nobody ever is. The moment they hesitate, you've got your user research.
If the idea is hollow, the users will tell you, and you get to be just as surprised as your client. No bruised egos. Just evidence. And if the idea is solid, even better. You now know it's worth building.
But validation isn't a thumbs up or a thumbs down. The real prize is what you learn in those conversations. The same questions that tell you whether to build also tell you how to build.
The questions worth exploring
When you sit down with users, you're not just asking "would you use this?" You're working out the shape of the thing. Three questions matter most.
How much control do they want?
Ask people how much say they want over what the AI does on their behalf.
Some will want to set it and forget it. They'd happily never see it. For them, the best answer is an invisible solution. No interface, no buttons, no chat window. The AI gets on with the work in the background and the problem quietly goes away.
Others will want their hands on the wheel. They don't trust a black box making choices for them, and fair enough. The moment people want control, your invisible solution becomes a visible one, and you've got an interface to design. You only know which camp they're in because you asked.
How do they want to see it?
If it does need to be visible, the default everyone reaches for is text. That's usually the client or stakeholder talking, not the user, and it's rarely a deliberate choice. They land on text because it's familiar, not because it gets the point across best. This is exactly where asking the user opens things up.
Ask users what they're actually trying to understand. Often a chart, a simple dashboard, or a quick visual does in a glance what a paragraph fumbles. AI is getting genuinely good at generating that sort of thing on the fly, so there's no reason to settle for a wall of prose when a graph would land faster.
How do they want to interact with it?
There's one last thing to ask, which is how people want to interact with it. Most will expect a conversation. Type a question, wait, read the answer, type again.
For plenty of tasks that's slower and more irritating than the alternatives. A few form fields can beat a back-and-forth with a bot that keeps asking you to clarify. A dashboard can feel like the most natural way in the world to poke at data. Ask users how they'd rather do it, and many will tell you the chat box was never the point.
Let the users build your case
Notice what's happened here. You haven't had a single argument about whether AI belongs in the product. You've turned a turf war into a research question, and come back with answers nobody can wave away.
Maybe the project dies because users don't care. Maybe it lives, but as a quiet background helper rather than the chatbot your client pictured. Either way, the decision was made by the people who'll live with it, not by whoever was most confident in the room.
That's a far stronger place to design from. And it's a much easier life than being the one who's forever saying no.