Almost nothing, if you make it for the right reason.
New players treat a wrong guess as damage. They wince, they apologise to themselves, they feel the streak sliding away. I want to make the case that a wrong guess is the cheapest diagnostic tool in the game, provided it was made for a reason.
The cost of a wrong guess is one attempt. If you have six, and the first two are intentionally cheap, the real cost is close to zero. What you buy for that attempt is a data point: this era is wrong, this artist is wrong, this production style is wrong. That is exactly the kind of information a short clip cannot give you directly.
The condition is that the guess must be specific and defensible. A random title gives you nothing. A guess that says I believe this is late-seventies disco gives you a clean yes or no on a hypothesis that covers hundreds of songs. When the answer is no, you have eliminated a hypothesis rather than wasted a line.
I keep a separate page in my notebook called the elimination list. It has entries like not seventies, not female vocal, not a cover. Each entry was bought with a wrong guess, and each one still helps. That page is the reason my third guess accuracy is now much better than my first: by guess three I have usually eliminated enough to reason instead of guess.
There is a limit to this logic. A wrong guess on the final attempt costs you the round, so the last attempt should always go to your best defensible answer, not to a hypothesis you are testing for fun. Cheap early, careful late.
You can test the principle at click here on a lane you do not usually play, where even your informed guesses will be wrong often enough to be useful.