The shape of a good recommendation

Most recommendation surfaces are optimized for the same thing: the next click. This is understandable — clicks are measurable, fast, and easy to attribute. But I've come to believe that the best recommendations optimize for something harder to measure: trust.

What trust means in a recommendation

When I trust a recommendation, I don't just act on it once. I return. I develop a prior about what the system will get right and what it won't. I start to understand its personality.

This is the difference between a friend's recommendation and a "you might also like" widget. Your friend has a model of you — not just your behavior, but your taste, your mood, your capacity for surprise. The widget has your history.

The Audible problem

At Audible, we're working on conversational discovery. The bet is that multi-turn conversation unlocks something that browse and keyword search can't: the ability to articulate what you don't yet know you want.

"Something like the last book I loved, but funnier. And shorter — I have a long flight."

That's not a query. It's a conversation. And a recommendation that responds to it well isn't just relevant — it's memorable. You come back.

What good looks like

The shape of a good recommendation is not just relevance. It's:

• Specificity: it knows why it's recommending this thing to you
• Honesty: it's willing to say "this isn't for everyone, but given what you said..."
• Restraint: one confident recommendation beats five mediocre ones

The widget gives you a grid. A good recommender gives you one thing and a reason.