Descript Virtual Assistant: episode edits done in the transcript, not the timeline

Nobody searches “descript virtual assistant” in week one of owning Descript. You search it around episode fifteen, when the tool has kept its promise, editing really is as easy as editing a doc, and you have discovered the catch: easy is not the same as done. Every episode still needs ninety minutes of transcript surgery, a filler pass, Studio Sound, clips, captions, show notes and an upload, and the only person doing it is the person whose face and voice are on the thing.

The daily rhythm a VA runs in your Descript

Most Descript work runs on an episode cycle rather than a daily one, so picture the week after a recording lands, whether that is a Riverside export, a Zoom file or a session captured in Descript itself.

Day one is ingest and correction. The file goes into the right project in your workspace, Descript transcribes it, and your VA does the pass the software cannot: fixing speaker labels, correcting the names Descript mangled (Australian suburbs, client names, clinical and technical jargon all get creative treatment from the transcription engine), and marking the obvious cut candidates. This correction pass matters more than it sounds, because everything downstream, captions, clips, show notes, chapters, inherits the transcript. A sloppy transcript becomes a typo burned into a caption on LinkedIn.

Then the edit itself, done the Descript way. False starts, second takes and the eight-minute tangent about the flight to Brisbane get deleted as text, and the audio and video follow. Word gaps get shortened with Descript’s gap-shortening tool so the pacing tightens naturally instead of every pause being amputated. The Remove filler words pass runs next, and this is where a trained VA earns the rate: Descript will happily detect every um and you-know, but accepting all of them makes a relaxed speaker sound like a hostage reading a statement. Your VA reviews each detection and keeps the fillers that carry rhythm. Studio Sound goes on last, applied per voice track rather than across the whole timeline, because it is superb on a voice in a boomy home office and rough on anything with music underneath.

Anything ambiguous, a section that might be gold or might be waffle, a guest comment that could read badly clipped, gets a timestamped note back to you rather than a unilateral cut. The mechanics are the VA’s job. The taste calls come home to you, at least until your VA has absorbed enough of your judgement to predict it.

Day two is the multiplication step, which is honestly where a Descript VA pays for themselves. From the finished edit, 3 to 5 clip-worthy moments get pulled straight from the transcript, dropped into your saved clip templates with the branded frame and caption style already built, and resized for vertical, square and widescreen. Because the captions come from the corrected transcript, they are right, not the auto-caption roulette you get making clips natively in each platform. Show notes, chapter markers and a shortlist of episode titles come off the same transcript.

Then the hand-off: MP3 exported and loaded to your podcast host, video pushed to YouTube through Descript’s direct publish where you use it, SRT file exported for anywhere that wants captions as a file, thumbnails prepped, and the project filed so the workspace does not become an archaeology site. If a word needs fixing in your voice, the AI-voice correction gets drafted and flagged, never shipped without your sign-off.

Across a week, that is one full episode cycle plus template upkeep, usually inside 5 to 8 of the VA’s hours, which is exactly why Descript work tends to live inside a broader role rather than standing alone.

The honest bit

Descript deserves its reputation, and it will still let you down in five specific ways that no VA can edit around.

The transcription is very good and not perfect. On clean Australian audio you will see the odd wrong homophone, a garbled place name, a client’s surname reinvented. Around 95 percent accurate means one word in twenty needs a human eye, and if nobody gives it that eye, the errors ship in your captions.

Deleting text in a video creates a jump cut, and until mid 2026 that was the end of the story. Descript’s new Smooth Jump Cuts feature can now bridge a cut with AI-generated frames, and it is impressive when it lands, but it is synthesised footage of your face, so every smoothed cut gets a human review before it ships. Where it does not hold up, the older options still apply: cover cuts with B-roll or scene changes, switch angles with Automatic Multicam if you record more than one camera, or accept the YouTube-native jump-cut look. A VA can manage all of these, but the trade-off is yours to pick, not theirs to hide.

Remove filler words is a detector, not an editor. Run it on auto and your most natural speaker comes out sounding clipped and anxious. The tool finds; a human chooses. Same with Studio Sound: it is close to magic on a single voice in a bad room and it can smear music, crowd noise and anything layered. There is judgement at every switch.

The AI voice fixes words, not answers. If you said “March” and meant “May”, Descript’s voice tools solve that cleanly. If you gave a rambling, wrong answer to question four, no voice model rescues it; you re-record.

And the one that matters most: Descript has made the mechanics of cutting nearly free, which means the entire remaining value of an editor is knowing what is boring. That is a skill, not a feature. It is the actual thing you are hiring, and it takes a few supervised episodes before a VA holds your standard for it.

What stays with you

Descript is not a regulated tool, but the boundary still needs drawing, and for some of you it is regulated by proxy.

Editorial voice stays yours. What gets cut, what leads the episode, which guest moment becomes a clip and whether that clip is fair to the guest stripped of its context, those are your calls, made on your VA’s flagged shortlist rather than discovered after publishing. Your AI voice is the hardest line: corrections are drafted, timestamped and approved by you before any export, every time, no exceptions for deadline pressure.

If your podcast touches health, money or law, one more layer applies. Plenty of Descript enquiries come from practitioners, and a 40-minute nuanced discussion becomes a 30-second clip with the nuance amputated. Anything in a clip that could read as clinical or financial advice out of context goes past you before it goes anywhere. Your VA runs the workflow; what your audience is told remains a practitioner decision. Guest release calls and permissions sit with you too.

What it costs and where to start

Descript-native work, episode edits by transcript, filler and Studio Sound passes, clips from your templates, show notes and hand-off to hosting, sits on our admin tier at $12-17 AUD an hour excl GST. Where you want genuine video production, motion graphics, grading, the things Descript itself does not do, that is specialist territory at $18-25. Most placements run 10-15 hours a week, roughly $500-1,100 a month, and because a weekly episode only fills part of that, Descript editing usually shares the role with inbox, admin or social scheduling. That is not theory: a July enquiry from a solo health practitioner listed Descript podcast editing as one task inside a 15 hour a week role, which is the shape that works.

Placement takes 7-10 business days, with 5-7 days supervised inside your Descript workspace, including at least one full episode edited under review, before anything ships solo. A $500 refundable deposit credits against your first month, there is a 30-day recalibrate-or-replace guarantee, and no lock-in beyond 14 days notice. We have made 87+ Australian placements since 2024, and Jenn takes every discovery call herself.

If you sell creative work, the creative industry page covers the wider role a Descript VA usually sits inside. The podcast editing task page goes deeper on the editing workflow itself, and the VA cost guide has the full pricing picture. Otherwise, book a discovery call and bring your messiest unedited episode.