Teams treat intuitive as something to add near the end, once the layout is settled and a week is left. It arrives much earlier than that, out of three things agreeing: what you see first, what looks like it can be operated, and what happens when you touch it. Get those three to agree and nobody notices the interface at all. Let one of them drift and people slow down, then describe the result as confusing without being able to point at the cause.
What you see first.
Every screen makes a claim about what matters, whether or not anybody decided it. Put six things at the same size and the screen has claimed all six are equal, which is a claim almost no product can support.
On the Hotels.com work the navigation came down to four destinations from a cluttered mix of tabs and menus. Cutting to four was a decision rather than a tidy-up. It commits to there being four things people open a booking app to do, and invites anyone to prove otherwise. A shorter list is a stronger claim, and a stronger claim is easier to test.
What looks like it can be operated.
Affordance is the quieter of the three and it fails quietly too. Anything that looks tappable has to be tappable, and anything inert has to stop wearing the costume. Cards are the usual offender: a card with a border and a shadow looks like a button to everyone except the person who drew it.
It gets harder in a dark interface, where contrast is already working two jobs at once, separating the content from the background and marking what is live. Holding AA contrast across an entire dark palette sounds like an accessibility chore. In practice it is what stops the interactive parts from dissolving into the imagery behind them.
What happens when you touch it.
Feedback answers the question the person is actually asking, which is usually some version of what did that just do to my data. A toast in the corner rarely gets near it.
Contacts with AI is the clearest case of this I have worked on. The feature tidies a contact list automatically, which means it changes records people depend on without being asked. Every action shows the evidence it acted on, and every action can be undone. Not one merge in the shipped product destroys anything. The undo is not a safety net bolted on afterwards; it is the thing that makes the automation acceptable to the person whose list it is.
People do not ask whether an interface is intuitive. They ask what that just did, and how long it takes to find out.
When the three disagree.
Conversation Search replaced a scattering of per-channel searches with one box. The hierarchy says this is the place to look. The affordance is a text field, which nobody needs taught. The feedback is where it could have fallen apart: a result that arrives stripped of its tags, status and priority makes the person open it just to find out what it is, and one search has quietly become two. Carrying that context into the result list is the whole difference between a search that works and a search that technically works.
You cannot test this on yourself.
You built it, so you cannot see it. You know which word is a link and which region is a card, and that knowledge is not removable. The only reliable method is to watch someone who does not have it, give them a real task, and say nothing at all while they do it. Five people is plenty. The moments you are looking for are the pauses, not the complaints, because people will forgive an interface out loud while their hands tell you the truth.
Intuitive, in the end, is just the word people reach for when nothing made them stop.
