A podcast editing budget can be wrong even when the plan price is correct. Descript separates the duration of media ingested from the AI actions run against that media, so a workflow estimate should track both.
Two different metering questions
Descript’s usage guide explains that Media minutes relate to uploaded and recorded content, while AI Credits cover actions such as Studio Sound, speech generation and Underlord. They should not be added directly because their units measure different activity. For example, recording a long conversation creates media-duration exposure, whereas applying several AI effects creates action-based usage. Account-level plan prices and changing exchange rates are intentionally excluded; those belong in AI 10.
Seats and pooled capacity
The provider says Editor seats contribute capacity to shared drive pools, while some other user roles cannot invoke the same metered AI tools. Model the production system around actual editors, shared drives and permission boundaries, not just the number of people allowed to view a project. If access must be revoked for a departing contractor, check whether their external exports and linked project assets remain accessible.
A calculator based on units rather than rates
For each project, record imported source duration M minutes, number of AI cleanup actions E, expected generated narration G minutes and agent/edit actions A. Then map each to the provider’s applicable meter or entitlement. If a feature costs a variable amount per use or by output length, retain a separate cost basis and avoid multiplying all minutes by a single universal token price. Track failed retries, revisions and alternate model choices as their own billable events.
Do not confuse transcription with editing quality
An inexpensive single-pass transcript may still require hours of editorial QA if the audio contains multilingual turns, multiple speakers or proper names. Conversely, an effective restoration tool may reduce time without improving factual correctness. Measure correction minutes, abandoned edits and publishing delays in the actual workflow. No savings percentage can be established without a tested baseline and measured results.
Example capacity plan
A weekly podcast imports one ninety-minute meeting and two twenty-minute interviews, then uses noise cleanup on selected segments and generates five minutes of scripted intro narration. List each upload separately, distinguish the recorded sessions from generated content, and itemize every repeat AI action. Compare the totals with the plan’s currently published allotments and top-up policy, using AI 10’s verified snapshot before approval. Monitor both remaining capacity and editorial approval status.
Pricing, permissions and date scope
All changing statements are scoped to documentation consulted by 25 October 2026. Model and account support, rights, language coverage, endpoint configuration and regional access can change. This article does not assert first-hand testing, an independent benchmark, present-day tariff, guaranteed authorship or unconditional commercial rights. Recheck the exact subscription, license and supported region before production.