Don’t build what users ask for

The gap between what users say and how they actually behave.

If you jump in and give users exactly what they say they want, you might end up disappointing them badly. Listening literally to user requests can mislead you. Not because they’re lying, but because they often don’t actually know.

But why do users ask for the wrong thing?

First of all, people often tell you what they think they should care about, not what they actually want.

Forget about products and features for a second. Have you ever had a friend who describes their “type” in a very specific way, and then keeps getting attracted to someone totally different? They weren’t hiding what they like. They really believed they should like a certain type, but reality proved otherwise.

Coming back to products, the same thing happens. When Netflix first used star ratings (users giving shows 1–5 stars), their assumption was: “If users rate movies highly, we’ll retain them more.” But the data didn’t support it. Over time they found that high star ratings didn’t lead to higher retention.

In response, they shifted to a “percentage match” model, focusing on what users will actually enjoy and stick with.

This shows that what users say they want (high rated content) can differ from what they actually need (content they’ll watch and come back to).

Secondly, if your users already have a comfortable experience with your product, they can confuse their needs with their comfort, and they’ll resist change. That’s why it’s usually less biased to run A/B tests on new users when you’re introducing something that dramatically changes the existing experience.

Remember when Instagram introduced the algorithmic feed and everyone freaked out? People swore they wanted the chronological feed back and had zero interest in algorithmic suggestions. But Meta ran a large scale experiment during late 2020 where they applied a purely chronological feed to users on Facebook and Instagram. The result: engagement dropped significantly, and many users shifted their time to TikTok and YouTube. So even though users said they preferred the chronological feed, their behavior showed they actually engaged more with algorithmic curation.

Third, user requests often reflect existing paradigms, not underlying needs. “If I had asked people what they wanted, they would have said faster horses.” Even though there’s no evidence Henry Ford actually said this, it captures the point perfectly.

Users imagine incremental changes, not radically new solutions. Before the iPhone, many mobile users said they wanted “better physical keyboards.” Their mental model was still BlackBerry. They couldn’t picture a world where typing on glass would feel natural. If you asked early Netflix users, they probably would have said they wanted “better DVDs,” not streaming.

People anchor to what they already know. Listen to what they say, but build for what they do.