Short answer: In Descript you edit a talking-head video by editing its transcript: delete a sentence and the video under it is cut. Work in three passes, content cut, then fillers and gaps, then retakes, and finish music and b-roll timing in a normal editor.
Descript turns the recording into a transcript, so you decide what stays by reading instead of scrubbing a timeline. That works well for footage that is a person talking, and it fights you for anything else. Below: when to cut in the transcript, the order of the passes, and which edits still belong in a timeline editor.
When editing the transcript is the right model
The test is simple: does your edit consist of choosing words? If you can describe the edit as "cut that sentence" or "drop the second take of the intro," the transcript is the fastest interface you have. If the edit is "hold the drone shot four more frames" or "cut on the snare hit," no transcript will help you.
Talking heads, interviews, lectures, webinars, podcasts, and course videos are word-shaped. The interesting decisions in those edits are almost always about content: this tangent dies, that retake is better, this answer is too long. Descript's edit-like-a-doc model makes those decisions the whole job. Select text, press Delete, and the media under those words is removed from the composition. Cut and paste a sentence and the media moves with it.
Two things make this more than a gimmick. First, it is non-destructive: deleted media is hidden, not gone, and you can right-click the edit boundary and restore it. Second, the transcript you edit is the one you ship, as captions or subtitles, so the editing pass doubles as a caption-accuracy pass.
Where the model breaks: music, montage, and anything cut to a beat. Descript has a timeline and scenes, but its native unit is the word, not the frame. Cutting a highlight reel to a music bed from a word processor is like landscaping with a scalpel. We covered Descript's fit for the clip-and-repurpose side in our long-video-to-shorts comparison; here we stay on the edit itself.
The three text actions that matter
Before the workflow, get these straight, because Descript gives you three ways to "remove" text and they do very different things to your media:
- Delete removes the words and the media under them. This is your actual cut.
- Ignore strikes through the text but keeps it visible in the script as a record of what you cut. Useful as a tentative cut you might reverse.
- Remove from transcript keeps the audio and video but hides the words from the script, captions, and exported transcripts. This is how you handle profanity in captions or off-mic chatter without touching footage.
And one adjacent mode people misuse: Correct mode fixes spelling, names, and speaker labels on screen without changing what was said. Fixing a transcript's spelling does not make the audio say anything different. If you have watched a caption say one thing while the mouth says another, that is why.
The pass order: cut, fillers, retakes
Editing a spoken-word video in one linear scrub is how people burn weekends. The transcript makes an ordered pass structure practical, and the order matters because each pass changes the text the next pass reads.
Pass one: the content cut. Read the transcript like an editor, not a proofreader. Delete whole sentences and paragraphs that do not earn their place: the false start of an answer, the repeated setup, the tangent. Do not fix grammar, do not touch fillers yet, just get the argument down to its final shape. Because deletions are reversible, you can be aggressive here, and you should be. On a talking head this pass takes minutes instead of hours.
Pass two: fillers and gaps. Once the surviving sentences are final, clean them. Descript detects filler words like "um" and "uh" automatically and offers bulk removal in a couple of clicks, with per-word escape hatches if one "uh" is doing actual work. Two caveats: bulk removal is English-only for now, and a fully de-ummed speaker can sound oddly robotic, so spot-check the tightest sections. Then run Shorten Word Gaps to compress long dead pauses, with a duration threshold you control rather than a blanket zap, which can make an interview sound assembled from confetti.
Pass three: retakes. Speakers rarely announce second takes cleanly, so search the transcript for phrases like "let me try that again." Select the weaker take's text and Delete it, then click the new edit boundary and use Regenerate: Descript generates a short transition between the real footage on either side of the cut, smoothing the jump cut so the speaker's head does not visibly snap between positions. This is the step that makes transcript editing survivable for video, as opposed to pure audio; every filler you removed and retake you cut left a jump cut before this existed.
What to finish elsewhere
Be honest about the border. Take these out of Descript:
- Music and beat-synced edits. The transcript has no concept of a beat. Any edit defined by rhythm belongs in a timeline editor.
- Frame-level b-roll timing. You can drop stock footage into Descript scenes, but getting a cutaway to land exactly as a word is spoken is a wordbar-and-fiddling exercise. A timeline does it natively.
- Heavy multi-cam interviews. If both speakers are on screen and the camera jumps are large, generated transitions struggle; real cutaway footage or a traditional multicam edit is the answer.
A real division of labor: Descript for everything the speaker said, Premiere, DaVinci Resolve, or CapCut for everything that plays behind, under, or between them. The audio-cleanup half, Descript's Studio Sound against Adobe Podcast and Auphonic, is covered in our AI podcast tools guide.
What it costs right now
Descript's pricing page, checked October 21, 2026, lists a free plan with text-based editing, three paid self-serve tiers priced per person per month, and a custom Enterprise plan. For publishing, you need a paid tier, because Hobbyist is the first one the page describes as watermark-free. Hobbyist is $16 a month billed annually ($24 monthly) with 10 media hours, 400 AI credits, and watermark-free 1080p export. Creator is $24 annually ($35 monthly) with 30 media hours, 800 credits, 4K export, and top-ups. Business is $50 annually ($65 monthly) with 40 media hours, 1,500 credits, and team features.
Prices are as published by Descript on October 21, 2026 and can change, so confirm on the official site before you subscribe. If you want the free plan's current limits, check the pricing page's plan comparison.
Verdict
Text-based editing is the right model whenever the edit is a word problem, which is true more often than people admit for spoken-word video. Descript executes it well in 2026: the delete-and-it-is-gone core, bulk filler removal, and Regenerate smoothing over jump cuts remove most of the drudgery from a talking-head edit. It is not a full replacement for a timeline editor, and pretending otherwise costs time. Use the transcript to decide what gets said; let the timeline handle what gets shown.
Next step: open one recent talking-head recording in Descript, do the content cut first, and only then run filler removal and Regenerate. If the edit you need is about rhythm or b-roll, skip the transcript and use a timeline. More on how Descript fits a wider workflow is in our content repurposing guide.
