Descript: our review
Descript edits a video or podcast by editing its transcript: deleting a word in the text deletes the matching moment in the picture and the sound. Useful for a meeting or an interview, provided you accept that uploaded content also trains its models.
What we take away from it
What counts in its favor
- Cutting a hesitation or a botched sentence means selecting text, not hunting for the right frame on a timeline: a typical thirty-minute edit shrinks to a few minutes of proofreading.
- Transcript and subtitles come out of the same motion as the edit, with no separate software and no export to line back up.
- A free plan exists to test the tool on a small volume before committing to a subscription.
What counts against it
- The contract allows Descript to use uploaded content (Input) and generated content (Output) to train and improve its models, with an opt-out option — but it isn't turned on by default.
- For the voice-clone feature (AI Speaker), the terms state plainly that employees, vendors, and outside contractors — including Amazon Mechanical Turk workers — may listen to audio samples for quality control.
- An uploaded meeting almost always contains the voice of participants who never accepted Descript's terms; the privacy policy protects the project as confidential information, but says nothing specific about these non-user third parties.
A real time saver for everyday editing, best kept to recordings that hold no sensitive information unless you've turned on the training opt-out yourself.
The facts
- First paid tier
- $16/month — Hobbyist, billed annually
- Data hosting
- Data in the United States · source
- Categories
- Video editing,Subtitles,Audio production
- Data verified on
- 30 July 2026
What happens to your data
- Is your data used to train their models?
- Subprocessors
- The terms name vendors and contractors, including workers on the Amazon Mechanical Turk platform, for listening to audio samples tied to the voice-clone feature (AI Speaker).
- Retention
- Voiceprints and facial-geometry data are kept for a maximum of three years after the last account login; other personal data is kept for as long as Descript deems necessary.
Descript in detail
Descript turns video editing on its head: instead of hunting for the right moment on a timeline, you read the transcript and delete the words to cut. The cut follows automatically in the picture and the sound. The tool works as well for a podcast as for a training video shot in a single take, hesitations included.
What it actually does
You upload a recording — an interview, a meeting, raw footage — and Descript produces an editable transcript in a few minutes. Deleting a sentence in this text deletes the matching section in the video. Subtitles and a first edit come out of the same motion, with no extra step. A firm or a marketing department filming a thirty-minute talk can bring it down to a clean edit by working on the text, rather than hunting for the right frames one by one.
What happens to an uploaded recording
This is the point that matters most for an organization that films client meetings or interviews: what happens to the file once it's sent? Descript's terms place each project in a protected category, described as confidential information. But the same contract provides that uploaded content and the produced result can be used to train and improve Descript's models, with an opt-out option that exists but isn't turned on by default. For the voice-clone feature specifically, the text is even more direct: employees, vendors, and outside contractors — specifically workers on the Amazon Mechanical Turk platform — may listen to audio samples to check the quality of the clone produced. No anonymization makes this listening harmless when the sample contains a recognizable voice.
Why it interests us
Transcript-based editing fits well with what we set up in training engagements: a single recording, a fast edit, ready-to-use subtitles. But a meeting almost always contains the voice of people who never read Descript's terms — a client at a legal or accounting firm, for example. Caution is warranted before uploading this kind of recording without having checked the account's training settings.
Where it falls short
Two distinct reservations. The first touches model training: the opt-out option exists, but it requires a deliberate action, it isn't the default setting. The second touches voice cloning: the stated use of outside contractors to listen to samples is a rare practice to disclose this clearly, and it deserves to be known before uploading a sensitive voice there.
Our recommendation
Descript is worth the subscription for everyday editing with no sensitive data. For a client meeting or a confidential interview, first turn off data training in the account settings, or use a tool that structurally has nothing to train on.
Who it's for, and who it isn't
Suited for
- Communications or training department that regularly edits videos from interviews or internal recordings
- Freelancer or small team producing a podcast who wants transcript, subtitles, and editing out of the same tool
Not a fit
- Firm or department that records client meetings containing personal data or privileged information, without having turned off data training beforehand
- Any organization that requires no human contractor ever listen to an audio clip, even anonymized: the voice-clone feature explicitly rules this out
Frequently asked questions
- What happens to a recorded meeting uploaded to Descript?
- The file is treated as a "Project," which Descript says it protects as confidential information. But the content (Input) and the edited result (Output) can also be used to train Descript's models, unless you opt out in the account settings. For a meeting with personal data about third parties, turn off this setting before uploading the file.
- Do humans really listen to my recordings?
- For the voice-clone feature (AI Speaker), yes: the terms state that employees, vendors, and contractors — including Amazon Mechanical Turk workers — may listen to audio samples to check the quality of the clone produced. This isn't systematic for a regular edit without voice cloning, but it's written plainly into the contract for this specific feature.
- Do you need consent from the person whose voice is cloned?
- Yes. The terms require you to have obtained consent from each person whose voice is used to train a clone, through a recorded consent statement. Uploading a third party's voice without their agreement is expressly forbidden by the contract.
- How much does Descript cost for a small team?
- The Hobbyist tier starts at $16/month billed yearly ($24/month billed monthly), for ten hours of media a month. The Creator tier, at $24/month billed yearly, adds 4K export and more AI credits.
- What's the difference with Riverside or CapCut?
- Riverside focuses on high-quality remote recording, without transcript-based editing. CapCut targets fast, visual editing built for short formats. Descript remains the only one of the three built around the principle "fix the text, fix the video," which changes how you edit, not just which tools you use.
Alternatives to Descript
Video editing · Subtitles
VEED
All-in-one platform to edit a video, generate its subtitles automatically and translate them.
Video editing
CapCut
Online and mobile video editor with automatic captions, background removal and AI effects.
Audio production
Riverside
Records remote podcasts and interviews in separate high-quality tracks, with AI-assisted editing.
Video editing
Wondershare Filmora
Consumer video editing software with AI features: automatic reframing, noise reduction, captions.
Our sources
- Pricing pageaccessed on 30 July 2026
- Terms of useaccessed on 30 July 2026
- Privacy policyaccessed on 30 July 2026
Prices and hosting terms change often. We recheck them periodically, but the publisher's page remains the only up-to-date reference.