AI4E
IntermediateReviewed AI-assisted guide

The 15-Minute Podcast Repurposer

Turn a reviewed podcast transcript into accurate show notes, a platform-neutral social thread and a carousel draft without losing timestamps or speaker meaning.

One-time setup

35–50 minutes once

Each run

15–30 minutes per episode after transcription

AI-assisted and reviewed by AI4E · last reviewed September 20, 2026. Reviewed AI-assisted guide — not represented as manually tested.

Intended outcome

A human-approved content pack containing timestamped show notes, a seven-post social thread and an eight-slide carousel outline, all traceable to the recording.

Best for

  • Interview and educational podcasts with a final audio master
  • Teams repurposing one approved episode across several channels
  • Episodes where quotes and timestamps must remain verifiable

Prerequisites and accounts

  • Descript account and permission to process the recording
  • Claude account or another synthesis tool with a suitable data policy
  • Canva account and an existing brand template
  • Final audio file, speaker names, episode title and approved call to action

Setup and handoffs

Complete these in order. Availability reflects the reviewed setup; provider plans can change.

  1. 1DescriptFree tier

    Create and verify the timestamped transcript

    Import the final audio, assign speakers, correct names and specialist terms, then check quoted passages against the waveform before export.

    Open Descript

    Handoff in

    Final MP3/WAV plus speaker list and spelling notes.

    Exact handoff out

    UTF-8 plain text in `[HH:MM:SS] Speaker: exact words` format. Keep one timestamp at every speaker change and at least every 30–60 seconds; never strip timecodes in later steps.

    Numbered setup

    1. 1.1Create a project and import the final audio master.
    2. 1.2Set the correct language and assign every speaker.
    3. 1.3Correct names, brands, numbers and specialist terms.
    4. 1.4Listen to every passage likely to be quoted and correct it against the recording.
    5. 1.5Export plain text with speaker labels and timestamps enabled.

    Human approval point

    A person confirms speaker identity, names and all candidate quotations against the recording before the transcript leaves Descript.

    If this connection fails

    • Wrong speakers: relabel one known segment for each speaker, then review every speaker change.
    • Drifting timestamps: retranscribe from the final audio master rather than an edited transcript.
    • Missing timecodes: enable timestamps in export; do not reconstruct them with AI.
  2. 2ClaudeFree tier

    Draft the content pack without changing meaning

    Use the verified transcript and the complete prompt below. The model must retain timestamps, label paraphrases and refuse to invent quotations.

    Open Claude

    Handoff in

    The verified timestamped transcript, episode title, audience, tone and call to action pasted below the prompt.

    Exact handoff out

    Markdown with the exact headings `SHOW NOTES`, `SOCIAL THREAD`, `CAROUSEL`, and `VERIFICATION TABLE`. Every factual claim or direct quote includes its source timestamp.

    Numbered setup

    1. 2.1Start a new private chat for the episode.
    2. 2.2Paste the prompt, then replace every bracketed field.
    3. 2.3Paste the complete verified transcript without removing timestamps.
    4. 2.4Generate the draft and retain the verification table.
    5. 2.5Do not pass any output to design until a person completes quotation review.

    Complete AI prompt

    You are repurposing a podcast from a verified timestamped transcript. Preserve the speakers' meaning and voice.

RULES
1. Use only information present in the transcript. Do not invent facts, quotations, speakers or timestamps.
2. A direct quotation must reproduce the transcript wording exactly and include [HH:MM:SS] and the speaker name.
3. Put paraphrases in your own words without quotation marks and cite the supporting timestamp.
4. If a speaker, word or timestamp is uncertain, write [VERIFY: reason] instead of guessing.
5. Keep all timestamps unchanged through the output.
6. Do not imply endorsement, certainty or causation beyond what the speaker said.

OUTPUT EXACTLY
# SHOW NOTES
- 80-word episode summary
- 5 timestamped chapter titles in [HH:MM:SS] Title format
- 3 key takeaways, each with a supporting timestamp

# SOCIAL THREAD
- 7 platform-neutral posts, each under 300 characters
- Post 1 states the episode topic; post 7 uses the supplied call to action
- Add a source timestamp in parentheses to each factual post

# CAROUSEL
- 8 slides: hook, six evidence-based points, call to action
- For each slide provide: headline | body (max 35 words) | source timestamp

# VERIFICATION TABLE
Columns: asset | claim or quote | speaker | timestamp | direct quote or paraphrase | verification needed

EPISODE TITLE: [EPISODE TITLE]
AUDIENCE: [AUDIENCE]
TONE: [TONE]
CALL TO ACTION: [CALL TO ACTION]
VERIFIED TRANSCRIPT:
[PASTE TIMESTAMPED TRANSCRIPT]

    Human approval point

    A person opens the recording at every cited timestamp, verifies each direct quote word for word and checks every paraphrase for preserved meaning.

    If this connection fails

    • Invented quote or timestamp: reject the output and rerun with only the relevant transcript excerpt.
    • Timestamps disappear: confirm the pasted transcript contains timecodes and repeat rule 5 at the end of the prompt.
    • Voice sounds generic: provide two approved examples of previous show notes, without personal listener data.
  3. 3CanvaFree tier

    Lay out the approved carousel

    Paste only approved carousel copy into a duplicate of your brand template; retain source timestamps in presenter notes or the review document.

    Open Canva

    Handoff in

    Approved `CAROUSEL` rows as `slide number | headline | body | source timestamp`, plus approved episode artwork.

    Exact handoff out

    Eight-slide PDF or platform image set, plus an internal review copy retaining each slide’s source timestamp.

    Numbered setup

    1. 3.1Duplicate the approved brand template.
    2. 3.2Paste one approved row per slide without rewriting quotations.
    3. 3.3Add episode art only if you hold usage rights.
    4. 3.4Store source timestamps in notes or the review filename.
    5. 3.5Export a review PDF before platform-specific images.

    Human approval point

    A person checks clipping, contrast, branding, rights, quotations and the final call to action before publishing.

    If this connection fails

    • Copy overflows: shorten a paraphrase; never shorten a direct quote without rechecking the audio.
    • Fonts or assets are locked: use the approved free brand template or obtain the required Canva plan.
    • Timestamp trail is lost: keep a separate review table keyed by slide number.

Example input

A 28-minute interview with Maya Chen about reducing meeting load, exported with verified speaker labels and timestamps, aimed at operations leaders.

Expected final output

Timestamped show notes, a seven-post social thread and an eight-slide carousel outline; each substantive point maps to the transcript, and direct quotes appear in a verification table for human sign-off.

Human approval points

  • Approve the corrected transcript before synthesis.
  • Verify every direct quote and timestamp against the audio.
  • Approve meaning, tone, claims and rights before design.
  • Approve the final rendered assets before publishing.

Privacy and security

  • Do not upload confidential, embargoed or unconsented recordings to third-party tools.
  • Voice transcription can misidentify speakers, names and numbers; the recording remains the source of truth.
  • Never publish an AI-produced quotation without listening to that passage.
  • Confirm music, artwork, guest and likeness rights before reuse.

What can be automated

  • Transcription draft, filler-word suggestions and content-format drafting
  • Moving approved structured copy into a prepared production queue

What cannot be automated

  • Consent, rights clearance or deciding whether private remarks should be published
  • Quotation verification, meaning preservation, reputational judgement and final publishing approval

Test safely before real data

  1. 1Use a 2–3 minute internal recording containing invented names and no confidential information.
  2. 2Plant one difficult name, one number and one deliberate correction to expose transcription errors.
  3. 3Verify every generated quote and timestamp against the audio.
  4. 4Export drafts to a private folder; do not connect automatic publishing during the test.
  5. 5Only use a real episode after the complete content pack passes human review.

Other workflow stacks

Inbound Lead Auto-ResponderDeep Market Research Brief