Agreed. One of the biggest mistakes I see startups making when talking to users is using surveys at all.

The theory seems to be "Oh, we'll get more information that way." Or, from the more honest ones, "This was less scary than trying to talk with people one on one." But surveys are mainly useful for confirming or denying particular generalizations from real conversations.

Surveys throw out all sorts of data. Every raised eyebrow. Every excited look. Every long pause where they try to figure out what they hell you are talking about. Every gesture, every tone of voice, every tangent. You really need that data when you're first starting out. Early on, the problem isn't the things you know you don't know. It's the things that you're pretty sure about that are false.