Descript's core idea is that you edit video and audio by editing a transcript. Delete a sentence in the text and it disappears from the recording. Remove every filler word with one command. Rearrange sections by moving paragraphs.
For talking head video, podcasts and interviews it is a genuinely better way to work, and anyone who has spent hours dragging clips on a timeline understands the appeal within about ninety seconds. The help centre explains this well. The organisation by task, edit, create, repurpose, manage, matches how people search when they are stuck, which beats organising by feature. Articles are short, accurate and get to the point.
If you want to know how to remove filler words, split a project, add captions or export at a particular setting, you find it quickly. As a support resource this is above average and I have rarely failed to find an answer. The workflow coverage is better than the usual button reference. Articles walk through the sequence of getting a recording in, cleaning it, editing it and getting it out, which helps someone new to the tool build a mental model rather than a list of features.
The captions and transcription material is solid, which matters because that is the foundation everything else sits on. Now the AI features, which is where I have real reservations. Studio Sound, which cleans up bad audio, Eye Contact, which adjusts your gaze to the camera, Overdub, which clones your voice so you can correct a mistake by typing, translation with lip sync, and Underlord, the assistant that performs editing tasks on request. Every one of these is documented at the level of how to turn it on.
Almost none of them are documented at the level of when you should. Take Studio Sound. It works well and it is aggressive. Applied to already decent audio it can produce something that sounds processed and slightly artificial, and knowing when to use it lightly or not at all is a judgement the documentation does not help you make.
Eye Contact is the same. Used on a slightly wandering gaze it is invisible and useful. Used heavily it produces the unsettling stare that viewers notice without being able to say why. The article tells you where the toggle is.
Overdub is the one that concerns me. Cloning a voice so that you can fix a flubbed line by typing the correction is a legitimately useful feature and I have used it. The technology also allows generating speech in a voice, and the guidance around consent is a short note rather than a serious treatment. Descript does require verification to train a voice model, which is more than some competitors, and the documentation moves past the ethics quickly.
Anyone learning this tool encounters how to clone a voice well before they encounter any real discussion of when that is acceptable. That ordering is wrong. Translation with lip sync deserves the same scrutiny. Producing a version of yourself apparently speaking a language you do not speak has obvious utility and obvious potential for misuse, and the article treats it as a repurposing feature.
There is no teaching on craft. Nothing on pacing, on when a cut should land, on how long to hold a shot, on structuring a piece so it holds attention, on when silence is better than a cut. Descript makes editing mechanically easy, which means the bottleneck moves entirely to judgement, and the documentation does not address judgement at all. Someone who learns every feature here will produce technically clean video with no sense of timing.
The product ships fast and the articles lag it. Interface changes, features get renamed, Underlord's capabilities expand, and the help centre catches up afterwards. Expect some mismatch between what you read and what you see. Cost catches people.
The free plan is enough to learn on and transcription minutes disappear quickly under real use, and the features most people want sit on the paid tiers. Worth understanding before you build a workflow around it. My three point five is for clear task based support content that gets you productive quickly on a genuinely well designed tool, marked down because the AI features are documented mechanically rather than thoughtfully, because consent guidance around voice cloning is nowhere near proportionate to the capability, and because there is nothing here about the editing craft that actually determines whether your output is worth watching.