Descript

What you get

Descript isn't a recorder with editing bolted on. It's a media app that treats your video as a document. The 2026 release expanded what the AI can do without changing the core model.

• Transcript-first editing. Delete the text, delete the video.
• Screen recording, plus remote multi-track sessions for podcast-style guests
• Underlord AI co-editor with a Claude Sonnet 4.5 model picker (2026)
• AI cleanup: Studio Sound, filler word and silence removal, retake regeneration
• Eye Contact (gaze correction), Green Screen, semi-automatic editing
• AI Speech with custom voice clones for patching narration
• Translation and dubbing in 30+ languages (Business tier)
• Public API in open beta as of 2026 for programmatic project, media, and Underlord actions
• Publish web pages with share links and embeddable players
• Shared project spaces with comments and timestamped feedback

For software orgs, this means sprint demos, onboarding modules, and release explainers can all come out of the same workflow without a dedicated video team.

Who it's for

Training teams under pressure from rapid product change. Every release ages your enablement content, so speed beats production polish.

1. Customer onboarding (record once, update fast)
2. Internal enablement across Product, CS, Sales, Support
3. Documentation videos that pair with written docs
4. Release notes as narrated walkthroughs
5. Podcast and long-form content teams (the original Descript audience)
6. Course creators producing weekly modules

The unspoken benefit: a PM, trainer, or solutions engineer can produce something genuinely good without waiting on a media team.

The AI features

Wide and uneven. The 2026 release added a model picker so you can choose Claude Sonnet 4.5 for Underlord, and dramatically expanded what Underlord can do.

Underlord (AI co-editor)

The 2026 Underlord can:

• Remove filler words and silence across a full recording in one step
• Identify bad takes flagged for review
• Suggest B-roll placements based on transcript content
• Generate show notes and chapter markers
• Create short social clips by flagging shareable moments in longer recordings
• Reorganize scenes
• Extend footage by pulling from existing frames to create new segments

The model picker lets you swap between Claude Sonnet 4.5 and other models for different reasoning styles. Useful if you've found your default has a tic you don't like.

Studio Sound

Makes a noisy office recording sound like a podcast booth. Burns AI credits. Use it intentionally.

AI Speech and voice clones

Custom voice training works. Useful for patching a sentence in narration without re-recording. Privacy note: voice samples are processed on Google Cloud, with samples shared with Amazon Mechanical Turk for quality assessment.

Eye Contact and Green Screen

Eye Contact corrects gaze toward camera. Green Screen handles compositing without a real green screen. Both work well enough to ship.

Translation and dubbing (Business)

Translate the transcript and re-voice the audio in 30+ languages. Useful for global training content.

Captions and chapters

Auto generated. Match the new social clip extractor for short-form output across LinkedIn, X, TikTok.

You still need editorial review for terminology accuracy, compliance wording, and product-name consistency. Don't ship Underlord's first cut on a billing flow.

Collaboration

Built around shared project spaces, not isolated local files. That matters when training ownership spans Product, CS, Support, and Marketing.

• Trainers and SMEs iterate in the same project
• Timestamped comments for QA loops
• Link-first review so stakeholders don't download anything
• Reuse the project structure when the UI changes
• Brand Studio access on Business tier for consistent assets

For a team building a reusable training library, the collab layer matters more than any single AI feature.

Security

The 2026 security posture is solid for content teams handling sensitive material.

• SOC 2 Type II compliance (via CPA audit)
• AES-256 encryption at rest
• TLS 1.2 in transit
• GDPR and CCPA aligned, Privacy by Design framework
• MFA required for employees on sensitive systems
• All AI features opt-in
• Annual security reviews and vulnerability scans
• Data Protection Officer reports directly to the CEO

Third-party processors disclosed:

• Rev and OpenAI Whisper (transcription)
• AWS and Google Cloud (storage)
• Stripe (payments)
• Zendesk (support)
• Hedra (avatar generation)
• Segment, Amplitude, Google Analytics (analytics)

Note: ISO 27001 is not claimed. SSO/SAML/SCIM not on the public security page (likely Enterprise tier). Voice cloning samples are processed via Mechanical Turk for quality assessment, which procurement should review.

If your recordings include staging environments or customer data, validate the third-party processor list and AI opt-out controls in procurement.

What users actually say

Public sentiment is split clearly: people who use it for content production love the transcript model. Heavy users who hit the new credit limits complain loudly.

Trustpilot reviewer signal (capturing patterns across 2025-2026):

"Very nice, it did exactly what I asked it to do (joining two videos together/gluing them)..." Trustpilot reviewer, Dec 16, 2025

"It 'got the job done'... But the glitches, and errors on uploading, and exporting issues, is a pain..." Trustpilot reviewer, Jan 23, 2026

"Now, I can only create 3-4 videos a MONTH without being prompted to 'top up.'" Trustpilot reviewer, Jan 7, 2026 (referencing the Sept 2025 pricing change)

"If your plan is to scale your academy large and globally... Descript is a really powerful companion to have." Gabriel Caudill, Senior Enablement and Digital Customer Success Manager, Algolia

What people like

• Editing model that non-experts actually get
• Underlord 2026 with Claude Sonnet 4.5 picker is genuinely useful
• Social clip extraction has replaced a separate clipping tool for many teams
• Translation and dubbing in 30+ languages on Business
• SOC 2 Type II, AES-256, TLS 1.2, MFA on employee access
• Public API in beta opens up programmatic workflows

What people don't like

• September 2025 pricing change to media minutes + AI credits was unpopular
• Studio Sound and Underlord can drain Creator's 800-credit allocation faster than expected
• Some users hit export and sync instability on heavy workloads
• Voice clone privacy posture (Mechanical Turk for quality review) gives some procurement teams pause
• No ISO 27001 yet
• Complex for teams who only need a basic recorder

How we'd use it

Design for updates, not perfection

Use modular scripts and short scenes so you can replace only changed sections when the UI shifts.

Watch your credit burn rate

The 2026 credit system rewards intentional use. Reserve Studio Sound and full-recording Underlord passes for the videos that need them. Don't apply Studio Sound to every internal Loom-style send.

Lock down terminology

Maintain a shared glossary (feature names, role labels, compliance language). Validate AI-generated text against it before export.

Use the model picker

Test Claude Sonnet 4.5 vs default for your Underlord workflows. The output style differs. Pick the one that matches your brand voice.

Use the transcript as your QA pass

Before export, review the transcript and captions for product naming, keyboard shortcuts, role-based instructions.

Build role variants from one base

Reuse a base project. Create short role-adapted versions for admins, end users, and support. Use the translation feature for global versions on Business.

Track what people actually watch

Pair publish analytics with support ticket themes and onboarding drop-off. Retire low-value videos quickly.

Isolate sensitive content

Use scrubbed environments. Avoid real customer data in demos. Set strict sharing defaults for internal versus external content.

Pricing

Descript moved to a two-meter system in September 2025: media minutes and AI credits, replacing the previous unlimited transcription model.

• Free: $0/mo. 60 min media/mo, 100 one-time AI credits, 5GB storage, 720p export
• Hobbyist: $16/mo annual, $24/mo monthly. 10 hr media/mo, 400 credits/mo, 100GB storage, 1080p, Studio Sound, custom voice clones
• Creator: $24/mo annual, $35/mo monthly. 30+5hr media, 800+500 credits, 1TB storage, 4K export, scales to 3 team members
• Business: $50/mo annual, $65/mo monthly. 40+10hr media, 1500+1000 credits, 2TB storage, dubbing in 30+ languages, custom avatars, Brand Studio, scales to 5 members
• Enterprise: custom. Advanced security, SSO/SCIM, custom media minutes and credits

Annual saves up to 35% vs monthly. The credit burn rate is the thing to model: Studio Sound on every video plus heavy Underlord use can exhaust Creator's 800 credits faster than the listed hour budget suggests.

Descript vs the alternatives

Pick Descript when:

• Transcript-first editing matches how you think
• Your team produces content weekly (podcasts, courses, product education)
• You need translation and dubbing for global audiences
• The Underlord 2026 features (B-roll suggestions, social clip extraction) are in your workflow
• Public API access for programmatic workflows matters

Pick something else when:

• Quick async messages (Loom)
• Advanced timeline editing with motion graphics (Camtasia)
• Polished Mac product demos (Screen Studio)
• Auto generated step-by-step documentation (Clueso, Guidde)
• Light usage where credit metering feels punitive

Descript is less a recording utility, more a compact media system. That's an advantage or overhead depending on your volume.

Integrations

Partly native, partly export driven, with a new public API as of 2026.

Public API (open beta, 2026)

• Programmatic access to projects
• Underlord AI actions via API
• Media imports

Export and publishing

• Share links and embed for LMS, help centers, wikis
• Local exports to docs hubs and course platforms
• Transcript and caption assets for accessibility and SEO
• Social clip exports sized for LinkedIn, X, TikTok

Third party processors (disclosed)

• Storage: AWS, Google Cloud
• Transcription: Rev, OpenAI Whisper
• Avatars: Hedra
• Payments: Stripe
• Support: Zendesk
• Identity: Stytch

The dependable pattern: produce in Descript, share for review, distribute final assets to docs, LMS, help center, and in-app education.

Use cases we've seen work

Podcast and longform content

The original Descript use case. Multi-track sessions, Studio Sound, Underlord cleanup, social clip extraction for promotion.

Course and academy content

Modular lessons with consistent narration via custom voice clones. Translation and dubbing for global launches.

Release-day enablement

Master recording for a feature launch, then output customer walkthrough, internal support briefing, sales objection clip, and short "what changed" updates from the same source.

Support deflection

Top recurring tickets converted into short explainers linked from support macros and help center answers.

Persona-based onboarding

Role-specific variants (admin vs end user) reduce cognitive overload and improve activation speed.

QA and bug repro

Record the issue path. Annotate via transcript edits. Engineering gets reproducible, timestamped context.

Social clip extraction (new in 2026)

Underlord identifies shareable moments in longer videos and generates platform-sized clips. Saves a dedicated clipping tool.

ROI

Returns come from throughput and cycle time, not raw media quality.

What pays off:

• Faster content refresh as the product changes
• Less dependence on specialized editors for routine training
• Better accessibility via captions and transcripts
• Global expansion via translation and dubbing on Business
• Reusable content blocks across future productions
• Social clip extraction replaces a separate clipping tool

What can hurt you:

• Overconsuming credits with reflex Studio Sound and Underlord usage
• Inefficient workflows that trigger rework
• Org-wide adoption without content standards
• Hitting the 2026 credit limits unexpectedly on Creator tier

KPIs to model:

• Hours per finished asset
• Translated language coverage if you ship globally
• Credit burn rate per asset
• Watch completion on shipped training
• Social clip CTR if used for distribution

Is it right for you?

In 2026, Descript fits if you:

• Produce video content weekly
• Want transcript-first editing
• Need AI-assisted cleanup and repurposing
• Care about collaborative review
• Will use the translation and dubbing on Business
• Want the new social clip extractor

Wrong fit if you:

• Only need occasional no-edit screen captures
• Are budget sensitive to the 2026 credit metering
• Need a timeline editor with motion graphics
• Have strict ISO 27001 procurement requirements (not currently claimed)

Descript rewards teams that build operating discipline around templates, review, and credit consumption.

The honest take

Descript in 2026 is more capable than it was a year ago: Underlord with Claude Sonnet 4.5, B-roll suggestions, social clip generation, scene reorganization, and a public API. The 2025 pricing change to media minutes + AI credits was the most controversial product decision in years. If your team is producing weekly, the math still works. If you're light usage, the new credit limits are a real consideration.

For training teams that value speed and iteration, Descript is a high-leverage choice. Build governance and credit-burn discipline into the workflow and the ROI lands.

Sources reviewed

• https://www.descript.com/ (verified June 2026)
• https://www.descript.com/pricing
• https://www.descript.com/security
• https://www.descript.com/security/ (third-party processors and AI training disclosure)
• https://descript.canny.io/changelog
• https://www.letscompareai.com/post/descript-underlord-update-faster-ai-video-editing-for-creators-in-2026
• https://sonix.ai/resources/descript-pricing/
• https://sonix.ai/resources/descript-review-pricing/
• https://www.trustpilot.com/review/descript.com

Use with Knolbase

Descript is your production engine. Record screen walkthroughs, role-specific variants, and multilingual training. Then upload into Knolbase. We map them into persona-based learning paths, sequence them by activation milestone, and connect training outcomes to product adoption signals.