The first three are always bad, and then they are not

Recorded formats pay more than written ones and almost every earner who could offer them does not, for one reason: the first attempt is excruciating to listen back to. That is universal, it lasts about three recordings, and none of those three go to a buyer.

The practical content of this post is the setup and the discipline that gets you through the first week of it, because the discomfort is real but it is not the actual obstacle - the actual obstacle is not having decided anything before pressing record.

Decide the frame before you decide anything else

Anonymity is a setup decision, not an editing decision, and it is the one that is expensive to fix afterwards.

The workable options, in order of how many people choose them:

Audio only. No frame to manage, no background, no lighting. It is the cheapest recorded format to produce and it is chronically underpriced relative to what it delivers - the argument is in audio only as an underpriced format. Voice over the material. Screen or image with your commentary, no camera on you at all. Camera below the face, or from behind, or at an angle that shows hands and posture and nothing identifying. Full face. A different decision with different consequences, covered in choosing between anonymity and a name.

Whichever you pick, check three things in frame before every recording: what is behind you, what is reflective, and what is on screen if you are sharing one. Reflections in windows, glasses and screens are the single most common way an anonymous recording stops being anonymous, and they are invisible to you while recording because you are looking at yourself, not at the room.

What you actually need

Less than the internet will tell you.

A phone or laptop microphone in a small room with soft furnishings sounds better than a mid-priced microphone in an echoey one. Room beats gear at every budget up to the point where you are recording daily. The full list of what is genuinely required is in equipment that is actually required, and it is short.

Two things that do matter: record somewhere quiet at a predictable hour, and record at a consistent distance so your volume does not change between jobs.

One-take discipline

The rule that makes recorded work profitable is that you record once and send it.

Not because a second take would not be better, but because the second take is better by a margin the buyer cannot detect and costs you the entire margin on the job. A 12-minute recording re-recorded twice is 36 minutes of talking plus 12 minutes of listening back, on a fee that assumed 15.

One-take discipline requires a small amount of preparation that people skip because it feels like it is not the work.

Write four bullet points before you start. Not a script - a script produces the worst possible delivery, because reading aloud is a distinct skill and you do not have it yet. Four bullets is enough structure to prevent the spiral where you lose your thread at ninety seconds and start again.

Say the first sentence out loud once before recording. The opening is where nearly every restart happens, and it is the only part worth rehearsing.

Editing, minimally

Cut the start, cut the end, normalise the volume, export.

That is the whole edit and it should take under five minutes. Do not remove every pause, do not cut every "um", and do not add music. Buyers of this format are buying a person reacting, and a heavily edited recording sounds like a broadcast, which is the thing they did not order.

Leave in the moment where you think for two seconds before answering. It is evidence that you were thinking.

Getting past the discomfort

The listening-back reaction is not a signal about quality. Nearly everyone dislikes their recorded voice, and the standard explanation is physical. As Kimura and Yotsumoto put it in a 2018 PLOS ONE study, the voice you hear while speaking includes both bone and air conduction, while a recording includes only air conduction. Their experiment also found large individual differences in which filtered recording people judged closest to their own voice, so the gap is real but not identical for everyone. It is an artefact of hearing, not feedback.

The way through it is volume and a rule: record three practice pieces on generic material, listen to none of them properly, then record the fourth and send that one. Practice material and where to get it covers what to use so that nothing you practise on belongs to a buyer.

Time yourself while you do it. Recorded work has a pricing structure that written work does not, because every extra minute of output costs you a minute of production plus review - the arithmetic is in what each extra minute costs, and knowing it before you set a length prevents the classic error of offering "up to 20 minutes" at a written-work price.

The bits that belong to other people

Lighting and framing rules that make material assessable at all are a photography subject rather than a recording one, and what image conditions actually permit is the better reference when the problem is the submission rather than your setup.

If you are recording anything that involves stated dimensions, use a documented convention rather than narrating an estimate; the site that owns measurement gear and method is where that belongs.

And the privacy side of recorded material - what is stored, for how long, and what leaves your device - is worth understanding before you build a workflow around a cloud editor, with the account of how automated systems handle submitted material covering the general shape of it.

Platform rules on what a recorded deliverable may contain, and how files are transferred, are on Rate Cock's judges page, and they constrain the format choices above more than anything in this post does.