Your Microphone Is Part of Your Score
Your answer exists only as the audio file it produced. Everything that is assessed is assessed from that file, which makes the thirty seconds you spend checking your microphone the highest return thirty seconds available.
The recording is the answer
This is the part of speaking preparation that receives no attention at all, and it is the only part that can reduce a competent answer to nothing. Whatever you said, what is assessed is what was captured. If the microphone was muted, the answer does not exist. If the room was loud, the answer is harder to make out. If you sat too far back, the quiet halves of your sentences are gone.
On this site the assessment pipeline makes that dependency explicit. Each clip is transcribed automatically, and a set of measurements is taken from the audio itself: how many words you produced, over how many seconds, your speech rate, your filler count and density, your longest pause, and a confidence figure from the transcription. Those measurements are then assessed against the four criteria. A recording that captures badly produces a poor transcript and unreliable measurements, and the band that follows from them is a band for your microphone rather than for your English.
The same dependency exists on the official test, for the obvious reason that nothing can be assessed from audio that did not record. This is why the test gives you a device check before anything counts, and why skipping it is the single most expensive shortcut in the section.
The silent recording
A muted microphone does not produce an error. It produces a perfectly valid audio file containing fifteen minutes of nothing, and there is no way to score that. The candidate usually has no idea until the results arrive, because from the inside a silent recording feels exactly like a successful one.
This happens more often than it should for a mundane reason: browsers and operating systems pick an input device on your behalf, and the one they pick is frequently not the one near your mouth. A laptop with a headset plugged in may still be listening through the lid microphone, or through a monitor's built-in one, or through a webcam on the far side of the desk.
The check takes ten seconds and has one requirement: you must see or hear evidence that sound is arriving, not merely that permission was granted. A level meter that moves when you speak is evidence. A green permission indicator is not. If you have a headset, plug it in before you open the test rather than during it, because changing devices mid-session does not always move the recording.
Distance, placement and clipping
There is a narrow band of microphone placement that works, and most people sit outside it in one direction or the other.
Too far away, which is the common case with a laptop's built-in microphone, and the room is recorded along with you. Your consonants soften, the ends of sentences drop below the noise floor, and a transcription system resolves your words with lower confidence, which is reported as a clarity signal. Nothing about your pronunciation changed. The microphone simply did not hear you properly.
Too close, or too loud, and the signal clips. Clipping is distortion rather than volume: the waveform is cut off flat at the top, and the affected syllables are unrecoverable. Plosive sounds, the p and b at the start of words, are the usual culprits, and a headset microphone sitting directly in front of your lips will produce them on every sentence.
The placement that works is a hand's width from your mouth and slightly off to one side, so your breath passes the microphone rather than hitting it. For a laptop, that means sitting forward, not leaning back in the chair. It is worth rehearsing in the same position you will actually sit in, because a practice run done leaning over the desk does not tell you anything about a test taken slumped in a chair.
The room
Background noise does not just sit underneath your voice, it competes with it. Continuous noise is the most damaging kind: a fan, an air conditioner, traffic through an open window, a fridge compressor. Your ear filters these out within a minute of sitting down, which is exactly why you will not notice them and the recording will.
Intermittent noise is less damaging overall but worse when it lands. A door closing over the first three words of your Task 5 argument removes the part of the answer where you named your choice.
Hard, empty rooms add reverberation, which smears each syllable into the next and is the single worst thing for intelligibility. A room with a bed, curtains and a rug in it is dramatically better than a kitchen or a stairwell, and this costs nothing to arrange. If you are recording practice answers at home, record them in the softest room you have and keep doing it in the same one, so that your recordings are comparable week to week.
Check that a level meter moves when you speak, not just that the browser granted permission. A muted or wrongly selected input device records a perfectly valid file full of silence, and there is no way to score silence.
What good audio does to your measurements
The measurements taken from your audio are not opinions, and improving the capture improves several of them at once without changing your English at all.
Word count and speech rate both depend on words being detected. If the ends of your sentences fall below the noise floor, your word count drops and your speech rate is computed lower than it really was, which pushes you towards the slow-and-effortful profile that caps Listenability at 7.
Longest pause is measured as the gap between one detected word and the next. Words that were spoken but not captured turn into silence, so a clean four second stretch of quiet speech can be recorded as a four second pause, which triggers a ceiling you did not earn.
Transcription confidence is the most direct one. Low confidence indicates that the words were hard to resolve, and it is reported as clarity rather than as accent, but a poor recording produces exactly the same signal as unclear speech does. There is no way to separate them from the file, which is why the fix belongs before the test rather than after it.
A ninety second pre-test routine
Do this in the same order every time you practise, so that on test day it is a habit rather than a decision.
Plug in your headset, if you use one, before opening the test. Close everything that makes a sound: messaging applications, mail notifications, anything with a background tab playing audio. Shut the window and turn off the fan, even if the room is warm, because fifteen minutes of warmth is cheaper than fifteen minutes of fan.
Then check the level. Speak a full sentence at the volume you intend to use during the test, not a quiet testing, testing, and watch the meter move. If it barely registers, you are too far away. If it pins to the top, back off or turn the input gain down.
Finally, sit in the position you will hold for the next quarter of an hour, and check the meter again from there. Most microphone problems are posture problems. People check their audio sitting forward with intent, and then relax into the chair the moment the test begins, at which point they are eighteen inches further from the microphone than when they tested it.
Test your audio sitting in the position you will actually hold for fifteen minutes. Most capture problems are posture problems: people lean in to test, then sit back the moment the first task starts.
Practise it
Five questions on capture, placement and what a poor recording does to measurements you cannot see.