How to record course videos: audio first, then batches, then ship.

The only part of a lesson a student cannot work around.

A soft, dim picture is something people tolerate while they listen. Audio that echoes, clips or fades means they cannot follow the sentence — and in a course they are following sentences for hours rather than minutes, usually while trying to do the thing you are describing.

Most of the improvement available here is distance rather than equipment. A microphone a hand-span from your mouth collects your voice and a little of the room; the same microphone across the desk collects the room and a little of your voice. This is why wired earbuds beat a laptop, and why a headset beats a better microphone that is badly placed.

The item that is specific to courses, and the one people skip, is writing the setup down. A webinar is recorded once. A course is recorded across weeks, in whatever room was free, at whatever level the software defaulted to that day — and the student hears every join, because they play the lessons in a row. Mark where the microphone sits, note the input level, and record every lesson from that same position, including the one you add in November.

Then run the two-minute test before lesson one exists rather than after eleven do. Record yourself teaching a real part of a real lesson, play it back on headphones, and listen rather than watch. It finds the echo you stopped noticing, the fan under the desk, the clipping when you get enthusiastic and the room that sounds like a corridor — all of which are cheap now and expensive once they are baked into a module.

The frame you pick in lesson one is the frame you are stuck with.

Screen recordings are where a course reads as either produced or improvised, and almost all of that is decided before you press record. None of it is about quality settings. It is about deciding once.

Fix the capture size and the aspect ratio for the whole course. A course whose lessons change shape halfway through gets letterboxed differently in every player it lands in, which is the specific kind of small wrongness students cannot name but do notice. Write the numbers down next to the microphone position.

Record from a profile that has nothing personal in it. A separate user account or a second browser profile costs five minutes and removes an entire category of problem: the bookmarks bar, the unread badges, the autocomplete that offers a client’s name in a search box, the calendar notification that arrives on screen carrying somebody else’s meeting title. A notification that lands mid-demonstration is not an edit — it is a re-record.

Then make the interface legible at the source. Scale the text up in the application or the operating system rather than zooming in afterwards, because zooming a recorded frame softens exactly the thing the student is trying to read. Check it at the size a phone will show it: if you cannot read the label on a phone, neither can somebody following along on one.

Framed once, then left alone for the whole course.

The talking-head pieces are usually the shortest lessons and the most rewatched: module intros, the transitions, the part where you explain why any of this matters. They are also the only place students see a person, which is most of why they finish.

Eye level does two things: it puts you in the conversational position people are used to, and it removes the angle that makes everyone look tired. The fix costs nothing — books, a box, a stand — and it is the largest single improvement available to a webcam you already own.

Consistency matters more here than quality does, because these lessons get played in sequence. Frame yourself once with a little room above your head and your eyes about a third of the way down the picture, then mark the tripod, the chair and the light and put everything back there next time. One controllable light in front of you is what makes an evening recording match a midday one — a window is free and excellent, and it is also the reason module four looks like a different course.

Treat the background and what you wear as course-wide decisions rather than daily ones. Every choice you make in this group is a constraint on the version of you who has to re-record one lesson in six months, and the difference between an afternoon and a re-shoot is whether that person can reproduce this frame.

Batch the setup, not the subject.

The expensive part of recording is not talking. It is changing setup, finding your voice again, and deciding what this lesson actually is. Batching removes two of those three from most of the day.

Group lessons by the setup they share rather than by where they sit in the curriculum. All the screencasts in one block, all the talking-head pieces in another. Alternating costs a physical change — camera, light, microphone position — and a mental one, and the mental one is worse: teaching to a camera and narrating a demonstration are different performances, and switching between them every twenty minutes is how a day produces two lessons.

Work from an outline, not a script, with one exception. An outline — the promise, three to five beats, the one thing the student does at the end — keeps you conversational and keeps the lesson to a single idea. Write out exact words only where they are load-bearing: definitions, numbers, disclaimers, and the ordered steps of a process where getting them out of order breaks somebody’s afternoon.

Agree the restart rule before the first take. When you fumble a sentence, do not stop the recording: stop talking, hold still for three seconds, and start the sentence again from the top. The silence is visible in the waveform, the take log says where it is, and the edit is a straight cut. Stopping and restarting the recording instead produces a folder of fragments, a different level on each, and an afternoon of reassembly.

Cut the waste. Leave the person.

Editing a course is not filmmaking. There are three cuts worth making and a long list of temptations that cost weeks and change nothing a student experiences.

The three: top and tail, so the lesson starts on the first useful word and ends on the last one; the restarts, which the take log has already located for you; and any stretch where you genuinely lost the thread and talked in circles. That is the whole edit for most lessons, and it is fifteen minutes rather than an evening.

Then do one loudness pass across every lesson in the module. Levels drifting between lessons is the most-noticed technical fault after echo, because it is the one that makes a student physically reach for something: the volume that was right for lesson four is wrong for lesson five, twice per module, all the way through a course they paid for.

And then stop. Leave the pause where you were thinking, the word you corrected, the aside that tells a student you have taught this before and know where they get stuck. Courses edited to remove every breath are exhausting to watch and, in teaching, read as somebody performing expertise rather than having it. The pauses are also where the student takes notes and where they hit pause themselves.

A file you cannot find is a lesson you will record twice.

The last group is the boring one, and it decides whether updating module three next year is an afternoon or a re-shoot. All four items cost minutes now and days later.

Set the filename convention before the first export and make it sort correctly on its own: course, module number, lesson number, version, date, with numbers padded and the date written year-first. Something like bizfound-m02-l04-v2-2026-07-29 reads the same to you, to an editor, to a captioner and to a folder listing. Boring and consistent beats clever, because the value is entirely in being able to find lesson four in eighteen months.

Keep exactly one version marked as published. When a lesson is re-recorded, the old export leaves the working folder entirely rather than acquiring another word at the end of its name. Every course that has ever shipped the wrong file shipped it from a folder containing two plausible candidates.

Store the masters and the project files in two places before anything is uploaded — a published lesson is not a backup, because it has been compressed and stripped of everything you would need to edit it again. And produce the transcript or caption file now, while your terminology is fresh in your head, rather than in a batch six months later when you have forgotten how you spelled things.

One recording day, laid out.

Below is a working shape for a day: two screencast blocks, one talking-head block, two real breaks, and a filing session that is part of the day rather than something you will do later. The times are illustrative; the shape is the point.

The two shaded rows are the ones that get deleted when the day runs late, and they are the two that pay. Skipping the setup block puts a level problem into every lesson recorded before lunch. Skipping the filing block turns a morning of good takes into a folder nobody can edit, which is the same as not having recorded them.

Three tiers, described rather than priced.

Read a row left to right and stop at the point where the next column stops describing a problem you actually have. Most people recording their own course belong in the first two columns on every row.

The rule for moving right on any row: upgrade only the thing that is currently the weakest link, and only after watching a lesson back and being able to name what is wrong with it. Buying a better camera while the microphone is still across the desk produces a sharper picture of somebody who is tiring to listen to for six hours.

The last column is honest about itself. It buys repeatability rather than quality — the ability to reproduce the same shot in a year, to hand the setup to somebody else, or to re-record one lesson without matching it by eye. That is worth paying for when a course is a product you maintain, and it is a hobby before then, which is fine as long as it is called one.

What is worth recording again, and what nobody but you will notice.

Recording ends when you decide it ends. Courses die in the gap between “recorded” and “good enough”, and the gap is almost always about things students never mention.

One test settles it: does this stop a student following, or does it only bother the person who recorded it? You have watched the lesson eleven times. They will watch it once, at 1.25 speed, while trying to do the thing you are describing. Those are entirely different experiences, and only one of them is the product.

The practical version is to publish the module, watch what happens, and re-record with evidence. The lesson worth fixing is the one students stop at — and you cannot know which one that is from the editing chair, because the reason they stop is rarely the reason you would guess. It might be a step that no longer exists in the interface, a diagram nobody can read on a phone, or a lesson that quietly needs to be two.

Which is the honest limit of this page. Recording well removes the reasons a student gives up on your production. It cannot tell you where they gave up, and that is a question about where the lessons live rather than about how they were made.