Five sounds before
the keeper.

I pulled 5,223 Sample Vault searches from the last 120 days because I wanted to answer a pretty simple question: how many sounds do you audition before one feels right?

Only 258 searches gave me a complete trail from the search, through the auditions, to a drag or copy. Producers auditioned a median of five sounds and made the choice in 35 seconds.

Five sounds.

That was lower than I expected. One search in ten went through 23 sounds before anything left the browser though.

Most searches were short comparisons.

Forty-three percent ended after three or fewer auditions. Eleven percent ended without an audition at all. At the median, the first preview happened 3.5 seconds after searching.

So which producer are you? Do you know the sound from its filename and waveform, or do you run a 20-sound shootout every time?

I expected producers to move back and forth between a few finalists. They usually didn't.

Only 4% returned to an earlier auditioned sound. Producers mostly kept moving until something held up, then dragged that last sound out of the browser. Thirty-eight percent replayed at least one candidate before choosing.

Choosing without auditioning doesn't necessarily mean choosing blindly. You may know the filename, recognise the waveform, trust the folder, or have used the pack often enough that playing the file adds nothing.

Yes, quite a lot.

This measures where consideration starts. It can't tell me whether the first result became the final choice, because the current event doesn't attach rank to the later drag or copy.

Still, the cost of a weak first page is pretty blatant. If the first credible sound sits at position 20, you've probably started judging something else already.

Across roughly 3,950 text searches, 44% contained at least three words and 16% contained six or more.

Word count can't tell me whether those phrases describe a sound, name a pack, or contain a list of constraints though. I'd need a properly tested privacy-safe classifier before making that claim.

For now, the honest finding is smaller: a substantial part of sample search already looks like a phrase rather than one keyword.

I build Sample Vault, a desktop sample browser for searching the sounds already on your machine. These numbers come from its aggregated product analytics.

1. Search, audition, and outcome shared one generated search ID.
2. Outcome was a drag or copy within 30 minutes.
3. Each sound was counted once before the first outcome.
4. Internal and test accounts were excluded.

A drag or copy tells me that the sound left the browser. It doesn't tell me that the sound reached a finished track. Searches without a recorded outcome are excluded too, so this describes completed choices rather than every attempt.

Result-position tracking started on 17 August 2026. The analytics use the client opt-out, percentages are rounded, and no raw search text or filename appears here.

Sample browsers are usually compared by library size, tag count, or search speed. Those measurements stop at retrieval. You still need to compare the candidates and pick one.

The typical choice took five auditions and 35 seconds.

The harder problem is putting a plausible keeper near the top. Returning 200 technically relevant files isn't much help when you have to play 23 of them. Optimising everything around the median would ignore the searches where nothing feels right for much longer.

As for me, I expected more than five. Maybe that says more about how I choose kicks than it does about search quality though.

How many sounds do you usually audition before dragging one into your project? And when none of them fit, do you go back to an earlier one or start the search again?