BLOG/AI

Think of AI in your product as a layer, not a feature

Think of AI in your product as a layer, not a feature

When we built AI into the flow instead of adding it to the screen as a button, conversion doubled. The difference wasn't in the model; it was in the placement.

The limits of the button approach

Most products put AI in a separate place: a chat bubble, a sparkle icon, a "Create with AI" button. This approach is tempting because it's easy to measure, but adoption always stays low. That's because it asks users to leave their current flow and learn a new tool.

In the first version of the TRT Global app, we did exactly that, and weekly usage stalled at 6%. The people who tapped the button were happy with it, but few people tapped it.

Users don't want AI; they want their task done in fewer steps.

Placing it as a layer

In the second version, we stopped making AI a visible feature: ranking in content lists, rewritten headlines in search results, automatic section markers in long content. Nowhere does the user see the words "artificial intelligence."

What we measure changed too: instead of tracking how much the AI feature is used, we now track how much faster the flow itself has become.

Rules we followed when placing it

We set three rules so that the model could live inside the product:

  1. Reversibility. Everything the model produces can be undone with a single tap; users never get trapped at any point.
  2. Uncertainty is shown. If the model isn't sure, the result isn't hidden; a second option is offered. Openly offering a choice is better than quietly getting it wrong.
  3. A latency budget. No AI step adds more than 400 ms to the flow; anything that would go over runs in the background.

What we measured

The difference between the two versions:

%6→%2.1xIncrease in flow completion rate
-31%Time to find content
400 msLatency ceiling added to the flow

What we learned

When AI is a feature, it needs promotion; when it's a layer, it doesn't. The second gets talked about less but used more.

SEE THE PROJECT

The TRT Global mobile app project

Go to project

Related Articles

Software

The first 60 seconds in a mobile app

Jul 14, 2026 · 6 min
E-Commerce

Five quiet decisions that cut cart abandonment

Jun 19, 2026 · 5 min
Software

In scaling teams, it's decision debt that piles up, not code

May 8, 2026 · 7 min