An AI button on everything is an AI button on nothing.

You’ve seen the screen.

A sparkle icon next to the title. Another on the chart. One on each KPI tile. One beside the column header. Maybe one on the logo, just in case. 😂

Each one opens the same chat.

Nobody knows what any of them will do.

It usually starts with good intentions. The team wants people to use the AI. So they make it visible. Everywhere.

But people learn fast. When every element has an AI button, the button stops meaning anything. People stop seeing it. And the few places where AI would really help get lost in the noise.

I ran into this while building ClarisTXM, my independent concept for running BPO work. So I wrote a rule, and then checked every screen against it.

The rule has two parts.

One assistant that knows the page. It sits in the same place on every screen. When you ask it something, it already knows which page you’re on and what’s showing. So a question asked from there is never a question about nothing.

One button on each thing a person acts on. A row in a work list. A case. An error. A recommendation. An order line. The header of the record you’ve opened. That button opens the same assistant, already asking about that one thing.

And that’s it.

No button on a chart. No button on a KPI figure, a legend, a column header or a count. No button on decoration.

If you can’t say what the button would ask, it doesn’t belong there.

That one test does most of the work.

A button on an overdue account can ask: “Why is this account on my list today, and what should I try first?” Useful.

A button on a bar chart asks… what, exactly? “Tell me about this chart”? The chart is already telling you. 🤔

When I checked the ClarisTXM screens, I listed every AI entry point: what it was attached to, what it would ask, and why it should stay. Nine things came out. None of them was sitting on a heading, a count or a chart by then, which told me the rule had already been shaping the screens.

Here’s how I’d run the same check on any app.

1. 1List every AI entry point. Every sparkle, every chat box, every “Ask AI”. Screenshots help.
2. 2Name the object each one is attached to. If the answer is “the page” or “the chart”, mark it.
3. 3Write the question it would ask. One sentence. If you can’t, it goes.
4. 4Keep one assistant that knows the page. Same place, every screen.
5. 5Label what comes back. Anything a model wrote says so, every time.

This isn’t about using less AI. It’s about putting it where the work is.

A person working a list of exceptions doesn’t want a chat window that knows nothing. They want to point at the exception and ask about it.

Fewer buttons. Better questions. And people actually press them.