Podcast production with generative tools

This skill covers producing a finished, publishable podcast episode when some or all of the audio is machine-generated. It is provider-neutral: it names TTS engines, music libraries, and hosts only as illustrative options, never as the method. The craft — format design, writing for the ear, voice direction, structure, rights, loudness, delivery, disclosure, and QA — is the same regardless of which tool renders the audio.

An episode is done when it (1) serves a defined show and audience, (2) sounds intentional and consistent, (3) uses only audio you have the right to distribute, (4) meets the loudness and file specs of its target platforms, (5) ships correct RSS metadata plus a transcript, (6) carries any legally or platform-required AI disclosure, and (7) has passed a listen-through QA. Skipping any of these is a defect, not a shortcut.

---

1. Scope and activation

Use this skill when the task is a spoken-word audio program: a topic explainer show, an interview series, a daily/weekly news brief, a narrative/documentary episode, a two-host conversational show, an internal briefing feed, or a show bible/format design for any of those.

Do not use it for:

• Music production — songs, beds, or stings as the deliverable (that is a music/audio-generation concern). This skill consumes music and treats it as a rights and loudness problem.
• A single TTS line — an IVR prompt, a notification, one voiceover clip with no show context.
• Video where the picture leads — if the deliverable is a video and audio is a track under it, the video workflow owns pacing and delivery. This skill applies when audio is the product.

If a request is ambiguous ("make an audio version of this article"), confirm whether the output is a one-off narration or an episode of a show. The two need different structure and metadata.

---

2. Format design (do this before writing a word)

The single most consequential early decision is format, because format dictates script style, voice count, structure, and length. Common podcast formats and what each demands (production heuristic, drawn from standard podcast-production practice — see Sources):

Format	Typical length	Voices	Script style	Best for
Two-host conversational	20–60 min	2 (chemistry critical)	Outlined, not fully written	Ongoing relationship with audience; opinion/commentary
Interview	25–75 min	Host + guest(s)	Host questions scripted; answers live	Expertise, guests, evergreen back-catalog
Solo / monologue	5–30 min	1	Fully or mostly scripted	Teaching, essays, focused explainers
Narrative / documentary	20–45 min	Narrator + tape/characters	Fully scripted, structured in acts	Storytelling, high production value
News brief / daily	3–10 min	1–2	Tightly scripted, dense	Recurring, time-sensitive, habit-forming

For a fully-synthetic show (every voice is TTS), the interview and two-host formats are the hardest to make convincing because they depend on turn-taking realism and chemistry; solo, narrative-narrator, and news-brief formats are the most forgiving. Weigh this when advising a format for a synthetic show.

Design the show once, in a short show bible, before designing episodes:

• Premise and audience — one sentence: who it is for and the promise each episode keeps.
• Cadence and length — a length band the format supports and a realistic release schedule.
• Voice identity — which voice(s), their names/personas, and a locked voice configuration so episode N sounds like episode 1 (see §4).
• Signature elements — the intro line, the outro/call-to-action, the music theme, and any recurring segments. These are the show's "sonic branding" and must be reused verbatim.
• Disclosure stance — decided up front (see §6), because it affects the script and metadata.

---

3. Writing for the ear

Scripts for audio are not text documents read aloud. The ear has no scrollbar, cannot re-read, and loses complex clauses. Rewrite prose into speech (production heuristics, widely taught in podcast scriptwriting — see Sources):

• Short sentences, one idea each. Break compound sentences. The listener holds only the last clause in memory.
• Contractions and spoken register. "It's / you'll / here's," not "it is / you will / here is." Write how a competent host actually talks.
• Front-load the point. Say the conclusion, then support it. Never bury the payoff behind a long subordinate clause.
• Signpost transitions aloud. "Three things. First… / But here's the catch… / So what does that mean?" These are the audio equivalent of headings.
• Kill the unpronounceable. Spell out or rephrase acronyms, symbols, URLs, and numbers. "twenty-sixteen L-U-F-S," not "-16 LUFS." "dot com slash join," not "/join."
• Read it out loud. Anything that trips a human reader will trip a synthetic voice too and will bore a listener. For a synthetic show, read-aloud testing is doubly important because you cannot ad-lib a save in post.

How much to script depends on format. Fully script the high-risk moments regardless of format: the hook/cold open, segment transitions, sponsor reads, and the close. For conversational and interview shows, outline the middle so the talk stays natural. For solo, narrative, and synthetic shows, script fully — a synthetic voice reads exactly what you give it, so the script is the performance.

When the script is the performance (fully-synthetic), also write performance direction into the script: mark intended pauses, emphasis, and emotional tone per paragraph, because those become your TTS instructions in §4.

---

4. Multi-voice casting and TTS direction

This is the craft that separates a convincing synthetic episode from an obviously robotic one. The goal is voices that are distinct from each other, consistent across episodes, correctly pronounced, and — for dialogue — plausibly reactive to each other.

4.1 Casting and consistency

• Lock voice configuration per persona. Save the exact voice identifier plus every generation parameter (stability/consistency, similarity, style/expressiveness, speed) as the persona's "voice lock." Reuse it for every episode. Voice drift — the same nominal voice sounding different across episodes or even across a long file — is the top continuity failure in synthetic shows (practitioner observation, corroborated across TTS vendor guidance — see Sources).
• Cast for contrast. In a two-voice show, pick voices that differ in pitch, pace, and timbre so listeners can tell who is speaking without name tags. Two similar voices are worse than one.
• Segment long scripts by speaker/role and generate each speaker's lines in that speaker's locked config, then assemble. This preserves consistency better than switching voices mid-request on engines that support only one primary voice per call.
• Higher stability/consistency settings reduce drift and expressiveness together. For a narrator that must sound identical for 30 minutes, bias toward stability. For a character that needs range, accept more variation and regenerate takes until consistent (heuristic).

4.2 Pronunciation control

• Fix names, jargon, and foreign words explicitly. Do not hope the model guesses. Use the engine's pronunciation mechanism: phoneme markup (IPA or CMU Arpabet via SSML <phoneme>), a custom pronunciation/lexicon dictionary, or, as a last resort, phonetic respelling in the text ("KAI-roh" for "Cairo"). Documented fact: SSML <phoneme> supports IPA and CMU Arpabet on major engines; some newer expressive models honor a lexicon or respelling but not full SSML — verify per engine (verified 2026-07-10 against ElevenLabs, Google Cloud TTS, and Azure Speech docs).
• Build a per-show pronunciation glossary in the show bible (recurring names, the show title, sponsor names) and apply it every episode so "the host's name" never changes pronunciation.
• Numbers, dates, units, and symbols are read inconsistently. Normalize them in the script ("July tenth, twenty-twenty-six"), which also helps human clarity.

4.3 Emotional pacing and performance

• Punctuation is your primary pacing tool. Periods = full stop; commas = short breath; ellipses = a hanging pause; question marks lift the final pitch. Sentence length sets rhythm — short sentences accelerate, long ones slow down. Around 140–160 words per minute reads as natural, engaged speech (production heuristic — see Sources).
• Apply emotion at the paragraph/section level, not word-by-word. Set a tone for a passage; use per-word emphasis sparingly. Stacking many emotion directions or switching every sentence produces unnatural tonal lurches (practitioner heuristic corroborated across TTS best-practice docs — see Sources).
• Use the engine's expressive controls deliberately: SSML <prosody> (rate/pitch/volume), <break time="…"> for engineered pauses, <emphasis>, or an engine's audio/emotion tags. Where the engine is prompt-driven rather than SSML-driven, put the direction in a delivery note the engine reads. Verify which markup your chosen engine actually honors before authoring — support is uneven (verified 2026-07-10).
• Generate multiple takes and choose. Synthetic delivery is stochastic; the first take is rarely the best. Budget for regeneration.

4.4 Host chemistry in generated dialogue

Chemistry between synthetic hosts must be engineered, because the voices are generated independently and do not actually react. Techniques (heuristics):

• Write the reactions in. Interruptions ("—wait, say that again"), back-channels ("mm-hm," "right,"), and callbacks to earlier lines. If it is not in the script, it is not in the audio.
• Vary turn length. Real conversation is uneven — a long point, a two-word reply. Scripting even, equal turns is a tell.
• Control the seams in assembly. Turn-taking realism lives in the gaps. Tighten or overlap the joins between speaker clips so the reply does not sound like it was recorded in a different room a week later. Consistent room tone under both voices sells co-presence.
• Do not fake a live remote. If you add fake "over the phone" filtering to sell realness, that edges toward deception; keep it clearly a produced show.

---

5. Episode structure and assembly

5.1 Structure

A conventional episode spine (production heuristic; adapt to format):

1. Cold open (optional, 10–30 s) — the single strongest moment or hook, before any branding. Earns the listen. Strong for narrative and social-clip-driven shows.
2. Intro / theme (5–20 s) — show name, host, one-line promise, over the theme music. Keep it short and identical every episode (sonic branding).
3. Episode tease — what this episode delivers, in one or two sentences.
4. Body / segments — the content, broken into clearly transitioned segments. Signpost each segment change with a spoken transition and, optionally, a short music sting.
5. Ad slots — see §5.3. Mark them structurally so they can be inserted/removed or dynamically served.
6. Outro (15–30 s) — recap, call to action (subscribe/share/link), sign-off, theme out.

Narrative shows use an act structure inside the body (setup → complication/turn → resolution or open question) rather than flat segments.

5.2 Assembly and editing

• Level the dialogue first, then place music and SFX under it. Speech intelligibility wins over music every time.
• Duck music under speech (sidechain or manual automation) so beds sit roughly 12–18 dB below the voice during talk and come up in the gaps (heuristic; tune by ear).
• Trim synthetic artifacts — clipped word-onsets, unnatural breaths, and swallowed syllables are common in TTS. Cut or regenerate the offending line.
• Consistent room tone / silence between clips. Dead-digital silence between assembled TTS clips sounds unnatural; a low consistent floor reads as one continuous recording.
• Match levels across segments so the intro, body, and ads are not wildly different volumes before you do the final loudness pass (§7).

5.3 Ad slots

• Baked-in vs. dynamic. Baked-in ads are part of the file forever; dynamically inserted ads are stitched at request time by the host and can be updated or removed. If the show will run ads long term, structure the timeline with clean, silent insertion points so ads can be dynamic.
• Disclosure carries into ads. A synthetic-voice ad read, or an AI-generated endorsement, may trigger the same disclosure and consent duties as the show (see §6), plus advertising-law rules about endorsements. Never synthesize a real person appearing to endorse something without rights.

---

6. Fully-synthetic vs. hybrid, and the disclosure duty

6.1 The production decision

• Fully-synthetic — every voice is TTS. Cheapest and fastest to iterate; best for solo, narrative-narrator, and news-brief formats; weakest for spontaneity and true interview dynamics.
• Hybrid — recorded human voice(s) plus synthetic segments (e.g., a real host with a synthetic co-host, synthetic narration around recorded interview tape, or synthetic reconstruction of unavailable audio). Best of both, but multiplies the rights and disclosure surface: every recorded human needs a release, and every synthetic voice needs a licensed/consented source.

Choose hybrid when the show needs a real human's authority, spontaneity, or an actual guest, and fully-synthetic when scale, consistency, or cost dominate and the format tolerates it.

6.2 Disclosure and consent obligations

Treat these as requirements, not style choices. They are the highest-risk part of a synthetic podcast.

• Voice cloning requires documented consent from the voice owner. Cloning or replicating a real person's voice without permission exposes you to right-of-publicity and voice-rights liability. As of 2026 multiple U.S. states protect voice as a distinct likeness right — Tennessee's ELVIS Act (effective 2024) is the first to name AI voice replicas explicitly, and California, Illinois, New York and others have related statutes; the federal TAKE IT DOWN Act (signed May 2025) targets nonconsensual synthetic depictions (documented fact, verified 2026-07-10 — see Sources; not legal advice — confirm current law for your jurisdiction and use). Consent must be specific, written, and documented; a verbal "yes" does not meet the bar. Never clone a public figure, a guest, or a co-host's voice without a signed release scoped to the use.
• Platform AI-disclosure policies apply to synthetic voices (documented facts, verified 2026-07-10):
  • YouTube requires creators to disclose realistic altered or synthetic content at upload, explicitly including synthetic/cloned voices and AI voiceovers that could mislead a viewer into thinking a real person spoke. A "How this content was made" label may be shown. Mass-produced, low-effort AI content risks demonetization under 2025 monetization updates.
  • Spotify does not down-rank AI-assisted content per se but bans unauthorized voice clones, aggressively removes spam/low-quality mass-produced audio, and is adopting a DDEX-based AI disclosure standard surfaced in-app. Its "Verified" program excludes profiles that primarily represent AI personas.
  • Apple Podcasts does not currently mandate a generic "this is AI" label for synthetic voices in the base RSS spec, but its content policies still prohibit impersonation and require rights to all audio. (Verify current policy at publish time — platform policy is volatile.)
• When in doubt, disclose. A brief, honest note ("voices in this episode are AI-generated" in the show notes and/or a spoken line) is cheap insurance and increasingly expected. Regulatory momentum (e.g., transparency laws emerging in 2025–2026) is toward more mandatory disclosure, not less.
• Hybrid shows: label which parts are synthetic if a listener could otherwise be deceived about a real person's words — for example, a synthetically reconstructed quote must be marked as a recreation, never presented as authentic tape.

---

7. Loudness and delivery standards

This is a hard-numbers area. Get it wrong and platforms turn your show up or down, or reject it.

7.1 The target

Documented facts (verified 2026-07-10):

• Apple Podcasts recommends preconditioning so overall loudness is around -16 dB LKFS (= LUFS) with ±1 dB tolerance, and true peak ≤ -1 dB FS, measured per ITU-R BS.1770. (LKFS and LUFS are the same unit.)
• The commonly cited creator targets are -16 LUFS for a stereo file and -19 LUFS for a mono file. The mono figure is lower on purpose: a mono file played through a stereo player is perceived roughly 3 LU quieter, so mastering a mono file to ~-19 LUFS makes it sound like a -16 LUFS stereo file. (Production heuristic reconciling the two numbers; the ~2–3 LU perceptual offset is documented in AES TD1008 — see Sources.)
• Spotify normalizes podcasts to -14 LUFS at playback with a -1 dBTP ceiling; YouTube normalizes toward roughly -14 LUFS as well. These platforms apply gain at playback rather than re-encoding your file.
• AES TD1008 (2021, supersedes TD1004) — the streaming-loudness recommendation — sets a distribution target of -18 LUFS for speech/"assorted" content and advises keeping integrated loudness above -20 LUFS; it notes speech is perceived ~2–3 LU louder than music at the same measured loudness. Note this is guidance for the distributor's normalization, not the creator's master file.

Practical resolution (heuristic): Master one stereo file to -16 LUFS integrated, true peak ≤ -1.0 dBTP as the portable default that sits inside Apple's window and is only gently adjusted by Spotify/YouTube. If delivering mono, target -19 LUFS for equivalent perceived loudness. Never chase Spotify's -14 by squashing dynamics — platforms turn quiet content up; loud, over- limited content cannot be turned back down cleanly and sounds fatiguing. Preserve dialogue dynamic range; a spoken-word show does not need to be as loud as a mastered song.

7.2 File and encoding specs

Documented facts — Apple Podcasts (verified 2026-07-10):

• RSS-delivered episode audio: MP3 or AAC. Mono 64–128 kbps, stereo 128–256 kbps, at 44.1/48 kHz.
• Spoken-word content is often fine in mono — smaller files, and most listening is on a single speaker or earbud. Reserve stereo for shows with meaningful music/SFX staging.
• Loudness is set before encoding: lossy compression does not change measured loudness, so precondition first, then encode.
• (Apple's high-resolution WAV/FLAC specs apply to subscriber audio uploaded to Podcasts Connect, not to the public RSS enclosure.)

Deliver a single file per episode via the <enclosure> in the RSS item (URL + byte length + MIME type, all three required).

---

8. Metadata, chapters, transcripts, and RSS

A podcast is an RSS feed; the audio is just the enclosure. Getting the feed right is as much of the deliverable as the audio (documented facts about the spec, verified 2026-07-10 — see Sources).

• Feed-level required elements: title, description, language, at least one <itunes:category>, <itunes:explicit>, artwork (Apple requires cover art typically 1400×1400 to 3000×3000 px, RGB JPEG/PNG), and an owner/contact. Missing <itunes:explicit> or <itunes:category> gets a feed rejected by Apple.
• Episode-level required elements: title, a <enclosure> (URL, length in bytes, MIME type), a globally unique <guid> that never changes, and a publish date. Changing a GUID duplicates the episode for subscribers.
• Chapters — two mechanisms: (a) embedded in the file (ID3v2 CHAP frames in MP3, MP4 chpl atoms in M4A), or (b) linked from the feed via the Podcasting 2.0 <podcast:chapters> tag pointing to an external JSON chapters file. Chapters improve navigation and are surfaced by several clients.
• Transcripts — the Podcasting 2.0 <podcast:transcript> element links a transcript file from the RSS item. WebVTT (.vtt) is the preferred format because it supports speaker labels (via cue identifiers), timing, and light styling, and has the widest client support; SRT is accepted but simpler. Major clients (Apple, and Spotify as of late 2025) render feed-linked transcripts. For a synthetic show you effectively have the transcript already — it is your script — so shipping one is nearly free and removes any excuse not to.
• Namespaces: declare the itunes and podcast (Podcasting 2.0) namespaces in the feed to use those tags; validate the feed before publishing.

---

9. Accessibility

• Ship a transcript for every episode. It is the core accessibility feature of podcasting — it serves Deaf and hard-of-hearing listeners, improves discoverability/SEO, and is trivially cheap for a synthetic show (your script). Link it via <podcast:transcript> (§8).
• Speaker-labeled transcripts (WebVTT cue identifiers) matter for multi-voice shows so a reader can follow who said what.
• Clear show notes with a summary, key timestamps/chapters, and link text that makes sense out of context.
• Avoid audio-only information that a transcript cannot carry — if a laugh, tone, or sound is load-bearing for meaning, note it in the transcript ("[laughs]", "[phone rings]").

---

10. Rights and safety checklist

Every audio element must be one of: original, licensed for this use, or public-domain/appropriately Creative-Commons-licensed with attribution honored.

• Music. You need rights to both the composition and the specific recording (master). Buying a song, or a personal streaming subscription, does not grant podcast/sync rights. Use a royalty-free / production-music library license that explicitly covers podcasts, a direct sync license from the rights holders, or genuinely license-clear music. "Royalty-free" means one fee, not free-of-charge, and not free-of-license-terms — read the license scope (episodes covered, ad-supported use, term) (documented fact — see Sources).
• Sound effects — same principle; use SFX libraries whose license covers commercial podcast distribution.
• Voices — every recorded human needs a release; every cloned/synthetic voice needs a licensed or consented source and must respect the TTS provider's commercial-use terms (some voices/tiers are non-commercial). Re-confirm the provider grants you rights to distribute generated audio commercially.
• Third-party audio / clips — quoting tape, songs, or another show requires permission or a valid fair-use/fair-dealing basis; do not assume short clips are automatically fine.
• Disclosure/consent for synthetic and cloned voices — see §6. This is the one that creates legal exposure, not just a takedown.

---

11. Pre-publish QA (do not skip)

Run this before every release. A synthetic show especially needs a human (or careful agent) listen-through, because generation errors are silent until you hear them.

1. Full listen-through end to end, at listening volume, on earbuds. Catch mispronunciations, voice drift, clipped words, robotic seams, awkward pacing, and level jumps.
2. Pronunciation check against the show glossary — every name, the show title, the sponsor.
3. Structure check — intro/outro present and identical to prior episodes; segments transition; ad slots correct; cold open lands.
4. Loudness/peak verification with a meter — integrated LUFS at target (~-16 LUFS stereo / ~-19 LUFS mono), true peak ≤ -1 dBTP, no clipping (§7).
5. Encoding/format — correct codec, bitrate, channels, sample rate; file plays on a phone.
6. Metadata — title, unique unchanged GUID, description, chapters, correct enclosure length in bytes, artwork, categories, explicit flag (§8).
7. Transcript attached, accurate, speaker-labeled (§9).
8. Rights — every music/SFX/voice element cleared (§10).
9. Disclosure — AI/synthetic disclosure present where required by law or platform, and consent on file for any cloned voice (§6).
10. Feed validation — run the feed through a validator; confirm it parses before it goes live.

---

12. Worked examples

The following are illustrative examples, not mandatory templates. They show how the decisions above compose on a real brief. Adapt the specifics.

Example A — Fully-synthetic daily 5-minute news brief

Intent: A recurring weekday "3-minute-ish" brief summarizing one industry's news, single synthetic host, published to Apple/Spotify/YouTube.

Format & bible: News-brief format (forgiving for synthetic). One host persona "Ava," a mid-pitch, steady voice locked at high stability (news must sound identical daily). Glossary of recurring company and person names with phoneme entries. Fixed 8-second music intro, 6-second outro with a "subscribe" CTA. Disclosure stance: state "This is an AI-generated briefing" in the show description and once in the outro.

Script (writing for the ear), abridged:

[COLD OPEN — no music]
Three things moved the market today, and the third one nobody saw coming.

[THEME 8s, duck under]
This is The Daily Brief for Tuesday, July eighth. I'm Ava. Here's what matters.

[BODY]
First. [company], read "ACK-mee", reported earnings after the bell...
(short sentences. One idea each. A pause... before the turn.)

[OUTRO, theme up]
That's your brief. Follow the show so tomorrow's finds you automatically.
This is an AI-generated briefing from [publisher]. See you tomorrow.

TTS direction: Generate the whole body in Ava's lock. Emotion at paragraph level: neutral- authoritative for the body, a lift on the outro CTA. Numbers/dates spelled out in the script. Two takes per day; pick the cleaner. Loudness: master mono to -19 LUFS, TP ≤ -1 dBTP; encode MP3 mono 96 kbps 44.1 kHz. Delivery: unique GUID per day; <podcast:transcript> VTT built from the script; chapters unnecessary at this length. Disclosure: in description + spoken outro (covers YouTube's synthetic-voice disclosure and general transparency).

Likely failure modes: drift if the voice lock is not reused; "ACK-mee" reverting to a spelling pronunciation without the phoneme entry; over-limiting to chase Spotify's -14 and sounding harsh.

Example B — Hybrid two-host conversational tech show

Intent: Weekly 35-minute show. One real human host (recorded) plus one synthetic co-host persona. This is the hard case: chemistry and disclosure both matter.

Decision: Hybrid, because the human brings spontaneity/authority and the synthetic co-host adds a consistent "explainer" foil cheaply. Consent/rights: the synthetic co-host uses a voice the producers licensed for commercial use (documented, provider terms confirmed); if it were modeled on any real person, a signed release would be mandatory — it is not, it is a wholly synthetic persona. Disclosure: the synthetic co-host is introduced honestly as an AI co-host in episode 1 and in every show description; a "How this content was made" disclosure is toggled on at YouTube upload because a synthetic voice is present.

Chemistry engineering: The human's side is recorded live around a scripted skeleton. The synthetic co-host's lines are written to react — interruptions, back-channels ("right, right"), callbacks to what the human just said — then generated in the co-host's voice lock and edited into the gaps with matched room tone so the exchange sounds co-present. Turn lengths are deliberately uneven.

Assembly & loudness: Level both voices to match, duck the bed 12–15 dB under talk, master stereo to -16 LUFS, TP ≤ -1 dBTP; encode AAC stereo 128 kbps. Delivery: chapters (JSON via <podcast:chapters>) for the segment structure; speaker-labeled WebVTT transcript distinguishing the human host and the AI co-host; clean silent ad-insertion points for dynamic ads.

Likely failure modes: the synthetic replies sounding recorded "elsewhere" (fix room tone and tighten seams); even, equal turn lengths reading as fake; forgetting the platform AI-disclosure toggle; the synthetic co-host reading a sponsor endorsement without the extra care ad reads require.

Example C — Narrative documentary episode with reconstructed audio

Intent: A 30-minute story where an unavailable historical quote is voiced synthetically.

The critical rule: A synthetically reconstructed voice or quote must be disclosed as a recreation — never presented as authentic archival tape. Mark it in the narration ("what follows is an AI re-creation of the words in the letter, not a recording") and in the transcript. Cloning a specific real person's voice would additionally require rights/consent from their estate; a neutral "reader" voice avoids that and is the safer choice unless rights are secured. Everything else follows the narrative act structure (§5.1), full scripting (§3), and the standard rights, loudness (-16 LUFS stereo), transcript, and QA passes.

---

Sources

Volatile facts (platform specs, policies, laws) verified 2026-07-10. Standards and craft heuristics are labeled inline. Do not treat legal notes as legal advice; confirm current law and policy at publish time.

Standards & loudness

• Apple Podcasts — Audio requirements (formats, bitrates, -16 dB LKFS ±1, -1 dB FS true peak): https://podcasters.apple.com/support/893-audio-requirements
• AES TD1008 (2021), Recommendations for Loudness of Internet Audio Streaming and On-Demand Distribution (supersedes TD1004; -18 LUFS speech, keep above -20 LUFS, speech ~2–3 LU louder than music): https://aes2.org/wp-content/uploads/2024/01/20210924_TD1008_v3.13.pdf and https://aes.org/technical-council/technical-document-aestd1008/
• Production Advice, summary of AES TD1008: https://productionadvice.co.uk/td1008/
• Spotify — Loudness normalization (-14 LUFS, -1 dBTP): https://support.spotify.com/us/artists/article/loudness-normalization/

RSS / metadata / transcripts / chapters

• Apple Podcasts — Podcast RSS feed requirements: https://podcasters.apple.com/support/823-podcast-requirements
• Podcasting 2.0 / Podcast Namespace spec (transcript, chapters, etc.): https://podcasting2.org/docs/podcast-namespace/1.0 and https://github.com/Podcast-Standards-Project/PSP-1-Podcast-RSS-Specification

Platform AI-content policies

• YouTube — Disclosing altered or synthetic content (blog): https://blog.youtube/news-and-events/disclosing-ai-generated-content/ ; "How this content was made" help: https://support.google.com/youtube/answer/15447836
• Spotify — verification/AI-persona and spam enforcement (secondary, practitioner reporting): https://www.musicbusinessworldwide.com/spotify-extends-verified-by-spotify-badges-to-podcasts-further-cracking-down-on-ai-impersonators/

Voice rights & disclosure law (not legal advice)

• Tennessee ELVIS Act overview (Art and Media Law): https://artandmedialaw.com/elvis-act/
• Right-of-publicity / voice-cloning risk map (Holon Law): https://holonlaw.com/entertainment-law/synthetic-media-voice-cloning-and-the-new-right-of-publicity-risk-map-for-2026/
• Deepfake & AI voice-cloning laws by state (Recording Law): https://www.recordinglaw.com/us-laws/deepfake-laws/

TTS direction (best-practice docs, provider-neutral synthesis)

• ElevenLabs — Text-to-speech best practices: https://elevenlabs.io/docs/overview/capabilities/text-to-speech/best-practices
• Microsoft Azure — SSML voice/prosody/phoneme reference: https://learn.microsoft.com/en-us/azure/ai-services/speech-service/speech-synthesis-markup-voice
• Google Cloud Text-to-Speech — SSML reference: https://docs.cloud.google.com/text-to-speech/docs/ssml

Music/SFX rights (practitioner references)

• The Podcast Haven — How to license music for a podcast: https://thepodcasthaven.com/how-to-license-music-for-a-podcast/
• Art and Media Law — Podcast music licensing: https://artandmedialaw.com/podcast-music-licensing/

Format, structure & scriptwriting (practitioner references)

• Buzzsprout — How to write a podcast script: https://www.buzzsprout.com/blog/write-podcast-script-examples
• Riverside — Podcast structure: https://riverside.com/blog/podcast-structure
• Podnews — LUFS/LKFS for podcasters: https://podnews.net/article/lufs-lkfs-for-podcasters