
By Tom Morgan · Research audit · Updated 14 November 2026 · About 14 minutes to read
Search for AI podcast tools, and you will find hundreds of pages promising to cut editing time by 60, 80, even 90 percent. Try to find the show behind any of those numbers and the trail goes cold.
We tried again on 14 November 2026. The claims we found ranged from roughly 2 to 10 hours saved per episode. Not one came with a named show, a stated baseline, and a method you could check.
So this piece does the reverse. It collects the cases that have a name attached, the platform changes that rewrote the job in the last twelve months, the benchmarks that measure how wrong the machines are, and the public failures. Then it grades every figure by how far you can trust it.
How this was made. This is a reading-and-research piece. We did not run these tools ourselves, and where a number is a vendor’s own claim we say so. Platform status (Spotify, Apple) was re-checked on 7 November 2026. Survey and case figures are shown as published, with their dates.
The answer in 60 seconds
| Job | Verdict | Evidence | What stays human |
|---|---|---|---|
| Transcription | Use it. Expect roughly 10-15% word errors on real podcast audio | A | Names, brands, numbers |
| Filler words, silence, noise | Use it | B | Cuts that touch a claim |
| Chapters | Use it, but write your own titles | A | Specific, searchable titles |
| Show notes | Use it, from your corrected transcript only | B | Every name, number, quote |
| Clips | Use it to find candidates | B | Choosing and approving |
| Handoffs and distribution | Best-measured payoff, though the measured cases are 2023 non-AI automation | B | One personal line per guest |
| Writing what the show says, or voicing it | Don’t | A | Everything |
What the evidence supports, and what it doesn’t
- AI pays where the work is mechanical. Transcripts, filler words, noise, and first-draft chapters are the safe end. The hours saved are plausible, but unaudited. We found no independent time study.
- No case in this article is both AI-era and measured. The named, measured gains (about 5 hours a week, about 1.5 hours per episode, about 70% of guests sharing) come from 2023 automation of handoffs. AI can draft the messages on top of that, but nobody has published what that adds.
- Every public failure has the same shape. The AI was asked to generate claims instead of processing what a host actually said. Fabricated quotes followed.
How we graded the evidence
Most podcast-AI articles stack vendor claims until they look like consensus. We tagged every number instead.
| Grade | Meaning |
|---|---|
| A | Primary data, independent measurement, platform documentation, or reporting based on internal documents. Method visible. |
| B | A named case or survey, usually published by a vendor or partner, self-reported or with an undisclosed sample. |
| C | Anonymous or unsourced. Fine as a hint, not as a figure to repeat. |
Five questions to ask of any “AI saved me X hours” claim:
- Is the show named, so you can listen to it?
- Is the before-state stated (hours, and which tasks were included)?
- Who measured it, and who sells the tool?
- Does the task list match yours (editing only, or editing plus notes, clips, and scheduling)?
- Is there a method, or just a number?
A claim that fails two or more of these is Grade C. Feel free to reuse the grading in your own work. We ask only that you link back.
What changed in the last 12 months
Production advice from 2023 is stale. The platforms moved the goalposts, and two of the moves change what you should automate.
| Date | Change | Why it matters |
|---|---|---|
| 15 Jul 2025 | YouTube’s “inauthentic content” rule took effect. It targets mass-produced, repetitive uploads. Commentary, reaction and clip content stays eligible when it adds significant original value. | AI clip farms are the target. Reviewed clips of your own show are not. |
| Sep 2025 | Spotify began generating transcripts, chapters, and guest and topic tags for published episodes on eligible shows. | The discovery layer now runs on machine-made metadata. Yours should be better. |
| Nov 2025 | Apple announced automatic chapters from iOS 26.2 for English episodes without chapters, skipping trailers and episodes under 10 minutes. Labeled “Automatically created”. | The chapter job moved from creating to correcting. |
| Mar 2026 | Apple Podcasts added HLS video streaming (iOS 26.4) on iPhone, iPad, Vision Pro and the web. Mac and Apple TV support was still pending at launch. | Video is a first-class feed in Apple Podcasts, not a YouTube side project. |
| 14 May 2026 | Spotify’s Distribution API went live for video on Libsyn, Podigee, Audioboom, Audiomeans and Podspace. Spotify for Creators and Megaphone were announced to adopt Apple’s HLS “later this year”. | One upload can reach both big apps, and video goes to Spotify directly, not through RSS. |
| 20 May 2026 | Spotify restated its ban on AI impersonation of other podcasters and extended its verified badge to podcasts. | Voice cloning of anyone but yourself is a policy risk. |
The Distribution API row is the one most automation guides miss. Spotify says video must be uploaded to Spotify directly, while RSS keeps carrying audio. Any Zap or Make scenario that triggers on an RSS item will carry your audio and silently skip your video. Check what your host exposes before you wire anything. And if you host on Spotify for Creators or Megaphone, check whether the Apple HLS integration has shipped for your show, because at the time of the announcement Spotify had not given a date.
Sources: SEJ on YouTube’s policy; Spotify for Creators, “Expanding our video ecosystem” (creators.spotify.com); MacRumors and 9to5Mac, 14 May 2026; PPC Land on the Distribution API; Podnews on Apple chapters and its FAQ; RightsTech on the impersonation ban. Grade A for what each platform says it did.
Why the workload keeps growing
Edison Research’s Infinite Dial 2026 put monthly podcast use among Americans 12 and older at 58% (about 167 million people), with 45% listening weekly. For 35- to 54-year-olds, monthly use was 68%. And 57% of Americans 12+ have both listened to and watched a podcast.
| Measure (US, 12+) | Result | Grade |
|---|---|---|
| Consumed a podcast in the last month | 58% (~167M) | A |
| Consumed in the last week | 45% (~130M) | A |
| Both listened and watched | 57% | A |
| Users who streamed a Spotify video podcast (Q3 2025, Spotify-reported) | 390M+, up 54% year on year | A |
Platform habits followed. In Edison’s Q3 2026 Podcast Metrics, YouTube led among weekly consumers, with Spotify at 17% and Apple Podcasts at 9% as the app used most. A separate August 2026 Sounds Profitable survey of 5,473 US adults put YouTube at 40%, Spotify at 17%, Apple Podcasts at 8%, and Netflix 5%. The surveys differ, so read them as direction, not one series.
The consequence for production: a video-first show needs more outputs per episode (clips, captions, chapters, a transcript page) than an audio feed did three years ago, in the same hours of the day. That gap is what automation is trying to fill.
Sources: Barrett Media on Infinite Dial 2026; Tubefilter; Podder summary; Barrett Media on Q3 2026 Podcast Metrics; NetInfluencer on the Sounds Profitable survey; Podnews on Spotify’s Q3 2025 remarks.
Four cases, graded
| Case | Era | AI involved? | Measured result | Grade |
|---|---|---|---|---|
| 1. My First Million | Undated customer story | Partly (Studio Sound) | None published | B |
| 2. This American Life | 2011-2014 | No (human transcripts) | About 6.7% more search traffic, attribution only | B |
| 3. Sit Down Startup | 2023 | No (classic automation) | About 70% guest sharing, about 5 h/week | B |
| 4. Zisner’s automation podcast | 2023 | No (classic automation) | About 1.5 h per episode | B |
Read that table first. It is the honest state of the public record: the best evidence predates modern AI, and the AI-era evidence has no numbers.
Case 1: A three-a-week show on a 48-hour clock
My First Million (HubSpot Podcast Network) · Grade B · Descript customer story
The show publishes three episodes a week in audio and video. According to Descript’s published story, recording to launch is a 48-hour window, with feedback rounds inside it. Senior producer Arie Desormeaux and lead editor Ezra Bakker Trupiano describe the old problem as version chaos: notes scattered across files, timestamps drifting between edits.
What changed was less AI cleverness than structure. Editing happens against a transcript, stakeholders comment in the same file, and the episode publishes from the tool. The AI-specific piece is Studio Sound, which the story presents as a fix for noisy guest audio.
What it proves: transcript-based editing and shared review make a tight cadence survivable. What it does not prove: the story publishes no hours-saved figure. Anyone quoting a percentage from it invented one.
Descript’s customer page lists other headline claims, for example, a 70% cut in video production time at Cloudinary (training content, not a podcast) and SaaStr publishing three to four videos a day. Same grade, same caveat.
Sources: Descript’s HubSpot story; Descript customer stories.
Case 2: Do transcripts move search traffic?
This American Life, 2011 to 2014 · Grade B · Study run by 3Play Media, a transcription vendor. Counts and method published.
This is still the best public dataset on the question, and it predates modern AI transcription by a decade. The show transcribed its archive between April and October 2011, then published transcripts a week after each new episode. The analysis window ran January 2012 to March 2014, with the early months excluded to let the effect settle. Google Analytics counted about 15.8 million unique visitors.
Of the search visitors, 6.26% landed on a transcript page. And 405 of the site’s 10,826 external links pointed at transcript pages.
Read it carefully. It is an attribution analysis, not a controlled experiment. The site was already strong (the study notes a PageRank of 8), and the vendor sells transcription. The effect, about 6.7% more search traffic, is real but modest. The cheap lesson for a small show: publish the full transcript as plain HTML text on your own episode page.
Source: 3Play Media, “This American Life: Boosting Podcast SEO with Transcription”.
Case 3: Automate the timing, write the sentence yourself
Sit Down Startup (Zendesk), run by Adam O’Donnell · Grade B · Zapier customer story, 19 Jan 2023
O’Donnell runs the podcast on top of a full-time role. Three automations carry it:
- When an interview time passes in Airtable, a Zap drafts a referral request in Gmail within minutes of the recording ending. He found guests respond far less if asked even an hour later.
- When he ticks a box for an episode, a calendar hold goes to the guest and their team for launch day.
- Two days before publishing, an email gives the guest sample wording for a social post.
Reported results: around 70% of guests now share the episode once live, and about five hours a week saved. The five hours cover all the automated work, not editing. He still adds a personal sentence to every referral draft before sending. That is the pattern to copy.
Source: Zapier customer story.
Case 4: One tracked link per platform
David Zisner’s automation podcast (LeAndolini Solutions) · Grade B · Make community post, 13 Mar 2023
Zisner keeps episodes in Airtable and lets Make do distribution. For each episode, the scenario builds a separate tracking URL per streaming platform, shortens it, pulls a post from a library of pre-written copy, and publishes to WhatsApp, Facebook, and LinkedIn. It also drafts the newsletter, which he still sends by hand.
Reported results: about 1.5 hours saved per episode, and a place among the top 20% of most-shared podcasts on Spotify. The ranking claim has no method behind it, so we do not lean on it. The tracking link per platform is the useful part, because it shows where listeners actually come from.
Source: Make Community.
The honest gap
Cases 3 and 4 use classic automation, not AI. We searched for named 2025 to 2026 shows publishing a measured, AI-era time saving and found none we could verify. A fresh search on 7 November 2026 turned up first-person blog posts and vendor pages, with claims from about 2 to 10 hours saved per episode, and one vendor page citing 70-80 hours a week saved across an unnamed network of twelve podcasters. All Grade C.
The nearest thing to a verifiable AI-era signal is scale. Podsqueeze, a Portuguese tool that turns uploads into transcripts, show notes, chapters, and clips, told Radio World in June 2025 that it had 100,000 registered users, with ECO and Rádio Comercial among named users. That is adoption, not hours saved, and it comes from the vendor (Grade B).
If you can supply the first verified before-and-after for an AI-era show, you own a gap nobody has filled. The three-episode test later in this article is built for exactly that.
Sources: Radio World on Podsqueeze; Podhome automation docs; About Podcasting episode on Claude, API and Zapier.
What the machines get wrong
Transcripts: plan for one wrong word in ten
Everything downstream (show notes, chapters, clips, search text) inherits the transcript’s mistakes. Two academic benchmarks give a sober picture.
A podcast fact-checking dataset (arXiv 2502.01402) measured Whisper’s word error rate on podcast audio:
A broader benchmark of commercial and open systems (arXiv 2504.05702) found overall word error rates between 9.0% (Rev) and 14.4% (IBM), with Whisper at 9.5%. The same paper flags that newer is not always safer:
The “invented words” figure is the insertion error rate, meaning text the model added that nobody said. The authors note OpenAI later added settings to reduce this.
Practical reading: vendor pages promising 98% accuracy describe read-aloud speech. Your two-guest episode with crosstalk will land nearer 10 to 15% errors. The words that break are proper nouns, brand spellings, and numbers, exactly the words your show notes and search text depend on. Keep a running glossary of guest and product names and fix those first.
Sources: arXiv 2502.01402, Table 4; arXiv 2504.05702, Tables 1 and 3. Grade A.
Chapters and show notes: from creating to correcting
Apple’s automatic chapters and Spotify’s generated ones mean you no longer start from a blank page. The catch is quality. Generated titles can run generic, so read them before listeners do. Chapters you provide override the automatic ones on both platforms, and you can opt out.
| Apple Podcasts | Spotify | |
|---|---|---|
| Generates | Chapters from the transcript; timed links to mentioned shows | Transcripts, chapters, guest and topic tags |
| Limits | English; skips trailers and episodes under 10 minutes | Eligible shows; chapters currently English-focused |
| Your chapters | Always shown instead | Shown instead |
| Opt out | Apple Podcasts Connect, or supply your own | Per episode or whole catalog |
Hosting platforms sell the same convenience. Buzzsprout’s Cohost AI produces transcripts, titles, descriptions, chapter markers, and social posts, and Buzzsprout advertises it from $10 a month (check current pricing). Reports on quality conflict: one reviewer on Riverside’s blog (a competitor) found accuracy inconsistent, while Podfeet’s host found the transcript clean and the chapter markers slightly dull. Those are two single reports (Grade C). The finding is that results depend on your audio.
Sources: Podnews Apple chapters FAQ; RSS.com help on chapters; Spotify for Creators; Buzzsprout; Riverside on Buzzsprout; Podfeet.
Two public failures and a practitioner’s warning
The Washington Post, December 2025. The Post launched AI-generated personalized podcasts on 10 December. According to an internal readout Semafor reviewed, between 68% and 84% of generated scripts failed the Post’s quality metric across three rounds of testing. Four staff described errors from mispronunciations to misattributed or invented quotes, and the tool sometimes turned a source’s quote into the paper’s own position. The product team recommended launching anyway, labeled beta. Grade A.
Inception Point AI and the flood. The company told Podnews it releases about 3,000 episodes a week with eight staff. Semafor reported a cost near $1 an episode. A Podcast Index snapshot reported by Bloomberg and Gizmodo found 39% of new podcasts created in one day were AI-generated, with Inception Point behind at 23.6% of new output. People do listen: TheWrap reported 12 million lifetime downloads and 400,000 subscribers. For a human show, the lesson is that you now compete with volume for search slots, so specific, corrected metadata and a real host are the things a factory cannot copy. Grade A.
A practitioner’s list. Pacific Content, a podcast agency, documented its own AI stumbles in August 2025. A chatbot invented names, companies, and email addresses once it passed about 40 entries in a list of 400. AI-written scripts read like generic LinkedIn posts. And post-production tools could not rescue bad source audio. Its advice: fact-check every AI claim against the source, treat output like an intern’s work, and fix the recording at the source. Grade B.
The pattern across all three: AI as a processor of what a human said is holding up. AI as a source of what a show says is not.
Sources: Semafor, Media Copilot and NPR on The Washington Post; Semafor on Inception Point; Podnews interview; Gizmodo; TheWrap; Pacific Content.
What listeners will accept
Edison Research’s Evolving Ear study (reported by Inside Radio, 28 Jan 2026) found listeners accept AI behind the scenes, for brainstorming, research, editing and admin, and largely resist it replacing the human on air. Roughly a quarter of newer listeners and one in five long-time listeners have tried an AI-narrated show. No sample size was published (Grade B).
Sounds Profitable’s Podcast Landscape 2025 fielding asked US adults who had ever consumed a podcast how they would react if a favorite show used AI-generated voices:
Listeners are also the reason clips deserve a place in the workflow. A Sounds Profitable study published 20 July 2026 (sample size not stated, Grade B) found about 89% of podcast listeners watch clips on social at least sometimes, and 71% would follow a host they like into short-form. Its authors warn that these are not conversion rates: about 30% say clips often or always lead them to an episode.
Put the two together and a rule emerges: use AI to find and format the moments, and keep the human voice on every one.
Sources: Inside Radio on Edison; Sounds Profitable on AI voices; Podcast News Daily on the clips study.
Who actually uses AI? Depends on who you ask
Surveys put podcaster AI use anywhere from a quarter to three quarters. They are not measuring the same thing: the samples differ (a vendor’s users, a host’s customers, a blog’s readers), and so do the years.
Two details are worth keeping. In RSS.com’s survey, the most common AI uses were transcripts (35%) and show notes or descriptions (33%), the low-risk end. And Spotify found larger-audience creators nearly twice as likely to use AI tools as beginners. Use cases cluster where mistakes are cheap to catch.
Sources: Spotify AI creator insights; The Podcast Host; RSS.com survey; Rain News on Descript’s study.
The time-saved claims, audited
Here is every editing-time number we found, and what stands behind it.
| Claim | Where | What backs it | Grade |
|---|---|---|---|
| Editing cut from 12 hours to 2 per episode | Tool-review blog | Unnamed podcaster, no method | C |
| Editing cut 60-80% | Podcast-AI blog | “Industry estimates”, no source | C |
| 3 hours to 45 minutes per episode | Review site | One unnamed weekly show | C |
| 2.5 hours vs 6.5 hours, same interview | Comparison blog | Described as a test, no named testers or data | C |
| Post-production averages 3-4 hours per episode | CodeWords blog | No source; vendor sells automation | C |
| 70-80 hours a week saved across 12 podcasters | Vendor landing page (found 7 Nov 2026) | Unnamed network, no method | C |
| 2 to 10 hours saved per episode | First-person blogs (found 7 Nov 2026) | Self-reported, no named shows | C |
| About 5 hours a week saved (all automation) | Zapier story | Named host, self-reported, 2023 | B |
| About 1.5 hours per episode (distribution) | Make community | Named host, self-reported, 2023 | B |
The claims all point the same way, which makes them plausible. None is audited. Until an independent study exists, measure your own.
The workflow
This is our synthesis of the cases above, not a tested recipe. Each step says what a machine should do and what you keep.
- Record clean, locally if you can. Every downstream error starts here, and Pacific Content found no tool could rescue weak source audio. Machine: noise cleanup, levelling. You: decide whether to re-record.
- Transcribe first and edit against the transcript. Load a glossary of names and terms if your tool allows. Machine: transcript, filler words, silences. You: proper nouns, numbers, anything that changes meaning.
- Listen to every cut that touches a claim. Automatic edits are safe on “um” and unsafe near a sentence a guest might regret. Spot-check the first 60 seconds and anywhere more than a few words vanished.
- Draft notes from your corrected transcript only. Ask the model to summarize, structure, and pull timestamps. Never ask it to recall the episode or add context. This is the line The Washington Post crossed. You: check every name, number, and quote against the text.
- Publish your own chapters and transcript. Supply corrected chapters and a VTT or SRT file, and put the full transcript as HTML text on your own episode page.
- Clip with a human gate. Clip tools score moments for likely performance. That score is the tool’s prediction, not a measurement. Review each clip for context before it goes out under your name, and add your own commentary or framing so it clears YouTube’s originality bar.
- Automate the handoffs, and check your host. Trigger from “episode published” or “interview finished”. Send guests a calendar hold and share-ready copy, give each platform its own tracking link, and draft the newsletter for your review. Confirm whether your trigger carries video before you rely on it.
- Measure before you believe. The next two sections show how.
What the math looks like
Take a show that publishes 4 episodes a month with 4 hours of post-production each, a $25 value per hour, and a $40 monthly tool bill. These are assumptions, not findings. Swap in your own.
| Scenario | Time cut | Hours saved per year | Value at $25/hr | Tool cost per year | Net per year |
|---|---|---|---|---|---|
| Cautious | 20% | 38.4 | $960 | $480 | $480 |
| Middle | 40% | 76.8 | $1,920 | $480 | $1,440 |
| Optimistic | 60% | 115.2 | $2,880 | $480 | $2,400 |
Formula: episodes per month × hours per episode × time cut × 12.
Break-even time cut = (monthly tool cost × 12) ÷ (episodes per month × 12 × hours per episode × value of an hour).
At these inputs, the tool bill ($480) equals 19.2 hours of your time, and your baseline is 192 hours a year, so a 10% time cut pays for the tool. Everything above that is yours. If your honest measurement comes in under your break-even figure, cancel the subscription.
The three-episode test
Nobody has published a clean AI-era time study for podcasts, so make your own. It takes three episodes before and three after.
| Step | What to log |
|---|---|
| Baseline, episodes 1-3 | Minutes per task: edit, transcript fix, show notes, chapters, clips, scheduling |
| Change one thing | For example, transcript-based editing, or AI show notes. Not five at once |
| Test, episodes 4-6 | The same task list, same timer |
| Quality check | Count errors you caught in AI output before publishing |
| Decide | Keep the change if net time saved beats your tool cost and the error count stays low |
If you publish the result, include your show name, your audio setup, the tools, and the raw minutes. That makes it a Grade B case, which is better than anything currently ranking. Link back to this article and use our contact page, and we will add verified, credited results in the next update.
Platform checklist for this week
- Open Apple Podcasts Connect and Spotify for Creators, read the automatic chapters on your last three episodes, and rewrite any generic title.
- Upload corrected transcripts (SRT or VTT), so both apps show your version.
- Put each full transcript on your own episode page as plain text.
- Ask your host whether its automation exposes video, and whether it is on Spotify’s Distribution API list.
- If you host on Spotify for Creators or Megaphone, check whether Apple HLS distribution has reached your show yet.
- Give every platform its own tracking link.
What we would skip
- Fully AI-written show notes with no transcript check.
- Voice cloning of anyone but yourself without disclosure. Spotify has restated its impersonation ban, and 47% of surveyed adults say AI voices would make them less likely to stay with a favorite show.
- Auto-posting clips nobody reviewed.
- Any tool whose headline is a percentage you cannot trace to a named show.
Frequently asked questions
Can AI fully edit a podcast?
Not unattended. It handles transcripts, filler words, silences, and noise well. Cuts that change meaning, clip selection, and fact checks still need a person, and word error rates of 10 to 15% on real podcast audio mean names and numbers need a pass.
Do I still need chapters if Apple and Spotify generate them?
Yes, if you want useful titles. Chapters you provide override the automatic ones on both platforms, and you can opt out of the automatic version.
Do transcripts help podcast SEO?
The best public evidence shows a modest gain, about 6.7% more search traffic for This American Life from 2011 to 2014. It is attribution, not a controlled test. The cost of trying is low: publish the transcript as HTML text.
What is the safest way to use AI for show notes?
Feed it your corrected transcript, ask it to summarize and structure only what was said, and check every name, number, and quote against the text before publishing.
Is it safe to use AI clips on YouTube?
YouTube’s inauthentic content policy targets mass-produced, repetitive uploads. Commentary and clip content remain eligible when they carry significant original value, so review each clip and add your own framing.
Can I send my video podcast to Spotify through RSS?
No. Spotify says video must be uploaded to it directly, through Spotify for Creators or a hosting provider on its Distribution API. RSS keeps carrying audio, so automations that trigger on RSS will skip your video.
How much time will AI actually save me?
Nobody has published an independent study. Named shows report about 1.5 hours per episode on distribution and about 5 hours a week across all automation, both from 2023 and both without AI. Use the three-episode test and the break-even math above.
Update log
- 14 Nov 2026: Re-checked Spotify and Apple video status; added Apple HLS (March 2026); moved the evidence gap to the top; added the case-era table, the five-question claim check, the break-even formula and the fresh audit rows.
- Next update: when Spotify for Creators and Megaphone ship Apple HLS distribution, and when a verified AI-era before-and-after case is submitted.
Cite this article
Morgan, Tom. “AI for Podcasters in 2026: What Editing, Show Notes and Distribution Automation Actually Deliver.” Updated 14 November 2026. Includes a three-grade evidence system (A, B, C) for podcast-AI claims.
Sources
Sources were reviewed in November 2026. Where a figure arrived through secondary coverage, the coverage is named. Prices and features change, so check vendor pages.
Audience and platforms
- Edison Research, Infinite Dial 2026 (via Barrett Media, Tubefilter and Podder)
- Barrett Media on Edison Q3 2026 Podcast Metrics; NetInfluencer on the Sounds Profitable survey
- Podnews: Apple Podcasts AI-generated chapters; Podnews Apple chapters FAQ; RSS.com help on chapters
- Spotify for Creators: automatic transcripts and chapters; “Expanding our video ecosystem with new partners and supported platforms” (creators.spotify.com)
- MacRumors, “Spotify to Adopt Apple’s Technology for Video Podcasts” (14 May 2026); 9to5Mac, same date; PPC Land on the Video Distribution API; RightsTech on the impersonation ban
- Search Engine Journal / related coverage on YouTube’s inauthentic content policy (July 2025 rename from “repetitious content”)
Cases
- Descript: HubSpot / My First Million customer story; Descript customer stories
- 3Play Media: This American Life study
- Zapier: Zendesk executive podcast / Sit Down Startup (Adam O’Donnell) story; Make Community: David Zisner
- Radio World / related coverage on Podsqueeze; Podhome automation docs; Buzzsprout on Cohost AI
Accuracy and failures
- arXiv papers: 2502.01402 and 2504.05702 (Whisper/podcast ASR benchmarks)
- Semafor on The Washington Post AI podcasts; Media Copilot; NPR
- Semafor on Inception Point; Podnews interview; Gizmodo; TheWrap
- Pacific Content: AI failures in podcasting (practitioner notes)
Listeners and adoption
- Inside Radio on Edison’s Evolving Ear; Sounds Profitable on AI voices; Podcast News Daily on the clips study
- RSS.com survey; Spotify AI creator insights; The Podcast Host; Rain News on Descript’s survey
A few secondary or older references (e.g., exact Zapier/Make customer posts, some vendor landing pages, certain 2026 survey write-ups) are harder to pin to a single permanent URL and may have moved; the primary platform announcements and major studies above are the most stable links.
