How to run a listening session that gets you useful feedback

Play an unfinished track to friends and you usually get one of two answers: "sounds great" or a vague frown. Neither tells you what to change. A listening session is a small, planned way to turn those reactions into notes you can act on. It takes about an hour and needs nothing more than a decent playback system and three or four people.

Why casual feedback fails

Two problems spoil most informal feedback.

Your listeners know what they are hearing. If someone knows which version is "the new mix", that knowledge colours the answer. Audio testing has dealt with this for decades. The ABX test, introduced by Bell Labs researchers W. A. Munson and Mark B. Gardner in 1950, plays the listener A, then B, then an unknown X that is secretly one of them, and asks which it is. Run double-blind, it stops the person running the test from steering the result without meaning to. You don't need a lab, but you can borrow the idea: hide which version is which.

You have heard your own track too many times. The mere-exposure effect, best known from Robert Zajonc's work, describes people coming to like things simply because they are familiar. The same article notes that listeners tend to like a song more after a few plays, though many repetitions can reduce that preference. Either way, after fifty passes your own judgement of the track is no longer a fresh listener's. That is the reason to bring in other ears at all.

Before the session

1. Pick the question. Decide what you want to learn: is the vocal too quiet, does the drop land, does the intro drag? One to three questions per track is plenty.
2. Choose two references. Pick released tracks in the same lane, as described in our guide to using reference tracks properly.
3. Level-match everything. Bring your mix and the references to the same perceived loudness before anyone hears them. The record industry long believed listeners preferred louder CDs, "even though that may not have been true", and our view is that you should not let that belief sneak into your session: an un-matched comparison mostly tests loudness, not the mix.
4. Label versions neutrally. Export the files as "Version 1", "Version 2" and so on, and shuffle the order so nobody can guess which is yours or which is newest.

Running it

• Play the first pass with no talking. Ask listeners to write down the moment their attention drifted, with a rough timestamp. Drift points are the most honest data you will get.
• Then do the A/B round. Play short matched sections (about 20 to 30 seconds) of two versions, or your track against a reference, without saying which is which. Ask "which would you keep playing?" and "what is different?"
• Ask your prepared questions last, so they don't steer the first, open listen.
• Collect notes on paper or a shared form before discussion. Group talk tends to follow the loudest person in the room.

Questions that work

Good questions are specific and answerable by people who don't produce music:

• Where did you stop paying attention?
• Which part would you want to hear again?
• Could you hear every word of the vocal? Where not?
• Compared with the reference, does this feel thinner, fuller, or about the same?
• If this came on a playlist, would you skip it? When?

Avoid "do you like it?" and "is the mix good?". They invite politeness, not information.

After the session

Sort the notes into three piles: things several people mentioned, things one person mentioned, and taste. Fix the first pile. Consider the second. Mostly ignore the third unless it matches your own doubts. Then check the fixes against the common faults in our ten bedroom mix mistakes, and if most of your listeners were on headphones, read our notes on mixing on headphones before you change the low end.

What we think

The biggest win here is not the clever A/B round; it is deciding your questions before you press play. Producers who skip that step tend to hear what they hoped to hear in vague comments. Blind, level-matched comparisons are cheap to set up and remove the two biases that matter most, familiarity and loudness, so there is little excuse not to use them. And one honest drift timestamp from a non-musician is worth more than ten compliments from fellow producers who know which version you are proud of.