Descript Review: Edit Video by Deleting Text — But Watch the Meters

Descript's founding idea is still its best one: transcribe the recording, then edit the media by editing the text. Delete a rambling sentence in the transcript and the video cut happens for you. For podcasters and talking-head creators — people whose raw footage is mostly speech — that single inversion replaces most of what a timeline editor is for. Around that core, Descript has grown into an all-in-one shop: recording, remote sessions, screen capture, AI voice tools, captions, and an agentic AI editor called Underlord. Based on its published capabilities and a deep pool of user reports, the core workflow still delivers. The honest complication is what happened to the pricing in late 2025 — which is where this review spends real time, because it is where users report the most pain.

What Descript actually does

You record into Descript (locally, via screen capture, or remotely with guests through Rooms — the rebranded SquadCast), or import existing footage. Everything is transcribed — 25 languages are supported on every tier — and from then on the transcript is the timeline. Deleting words cuts the media. One click removes filler words ("ums," "uhs," "likes"). Studio Sound cleans up room echo and noise. AI Speech — the current home of the feature formerly called Overdub — can regenerate a flubbed word from typed text in your cloned voice, matching mouth movement on camera. Captions, templates, and per-platform export handle the packaging pass. It is a genuinely complete pipeline for speech-driven content: many creators go from raw recording to published episode without leaving it.

Who it's for — and who it isn't

The tool is built for people whose content is talk: podcasters, interview shows, course creators, tutorial and screen-recording channels, and talking-head YouTubers. For that work, editing text instead of waveforms is a step-change in speed. It is not a replacement for a full NLE if your work is visually driven — multi-cam music videos, heavy motion graphics, color-critical grading. Editors doing that work treat Descript as a fast first pass (the "radio edit") and finish elsewhere; Premiere Pro has even adopted its own text-based editing for exactly this reason. If your footage is mostly B-roll and music, the transcript trick has nothing to grab onto.

Pricing: you're buying two meters now

In late 2025 Descript replaced its old transcription-hour plans with a system that meters two separate things. Media minutes count everything you upload or transcribe — and every file counts, so a one-hour remote interview with one guest plus a B-roll reel draws three files' worth of minutes from the same pool. AI credits are consumed by AI features: Studio Sound, filler-word removal, Underlord requests, and the rest. Top-up packs for both are sold separately (check current top-up pricing in-app — it is not published on the main pricing page). Here is the current tier structure:

An Enterprise tier (custom pricing, SSO/SCIM, dedicated support) sits above Business, which also adds multitrack transcription, custom avatars from a photo, Brand Studio, and translation/dubbing in 30+ languages. Two honest flags on the numbers above: the advertised $16/month Hobbyist price requires paying a year up front — month-to-month is $24 — and the Free plan's 100 AI credits are a one-time grant, not a monthly allowance. Prices and limits reflect descript.com/pricing as of August 2026; this structure has changed before, so confirm the current page before committing.

Underlord: the AI co-editor is the real differentiator

Underlord started as a menu of one-click AI actions and has evolved into an agentic editor: type a request like "remove filler words, tighten the pacing, and make three social clips" and it executes the steps against your project — jump cuts, captions, animated text, even building an edit from an idea. Full access starts at the Creator tier; Hobbyist gets a limited version. Early hands-on reports are consistent about the shape of it: impressive at multi-step grunt work, weaker at taste — B-roll selection and pacing choices still benefit from a human pass. Treat it like a fast junior editor whose work you review, not an editor you hand the final cut to. And because every request burns credits, it is also the feature most responsible for the credit-consumption complaints above.

The rest of the toolkit

AI Speech (voice cloning plus stock voices) is the quiet workhorse: the Regenerate feature fixes a flubbed word or phrase from typed text, cloning your voice and matching mouth movement on video — a retake without a reshoot. Studio Sound remains one of the better one-click noise/echo cleanups in any editor. Rooms gives you SquadCast-grade remote recording inside the editor, capped by plan (2/5/15/25 hours per month) — enough for a weekly show on Creator, tight for daily recording. Built-in screen recording makes it a strong fit for tutorial and software channels. None of these alone justifies the subscription; together, they are why "record, edit, fix, caption, publish" can genuinely happen in one app.

The honest limitations

The complaint record is worth taking seriously, because it is consistent across platforms. Capterra reviewers — who still rate Descript 4.6/5 across 183 reviews — report recordings stopping mid-session without warning, crashes with lost edits, and in one January 2026 report, projects edited months earlier reopening as raw recordings with cuts and effects missing. Support and refund handling draws sharp criticism in recent reviews, including promised refunds not issued. The free tier is effectively a watermarked demo rather than a usable plan. And the post-overhaul pricing model is the thread running through all of it: new and light users generally love the product, while long-term heavy users are the ones reporting bill shock, credit burn, and reliability erosion. A pragmatic mitigation for the stability risk: keep source recordings backed up outside Descript, and export finished cuts promptly.

How it compares

Riverside beats Rooms for remote capture (separate local tracks that survive bad connections) — record there, edit in Descript is a common stack. CapCut is the free general editor, but has no true text-based editing and its captions are no longer free. Podcastle is a lighter, browser-based, audio-first alternative with a more usable free tier. Adobe Podcast covers free audio cleanup, and Premiere Pro now has text-based editing built in for editors already in Adobe's world. Camtasia is the screen-recording specialist, and for pulling shorts out of long footage, dedicated clipping tools like Vizard or Opus Clip go deeper than Descript's clip features. Descript's edge over all of them is the combination: no other tool puts transcript editing, remote recording, voice repair, and an agentic AI editor in one project.

Pros and cons