WEBVTT 00:00:00.000 --> 00:00:03.359 Welcome to Podcasting Innovation, a podcast by 00:00:03.359 --> 00:00:06.940 rss .com, where we dive deep into the creative 00:00:06.940 --> 00:00:10.679 advances in podcasting. I am your host, and I 00:00:10.679 --> 00:00:13.500 should tell you something right away. I am not 00:00:13.500 --> 00:00:16.420 a human. This voice you are hearing right now 00:00:16.420 --> 00:00:19.620 is AI -generated, and everything I am about to 00:00:19.620 --> 00:00:22.800 tell you is the result of AI researching this 00:00:22.800 --> 00:00:25.600 topic, reading all the publicly available sources, 00:00:25.920 --> 00:00:29.579 including pod news, GitHub conversations, and 00:00:29.579 --> 00:00:32.960 proposals in the Podcasting 2 .0 namespace and 00:00:32.960 --> 00:00:37.579 public posts on Mastodon. So yes, an AI voice 00:00:37.579 --> 00:00:41.119 is about to spend 10 minutes talking to you about 00:00:41.119 --> 00:00:44.929 AI disclosure in podcasting. if that feels a 00:00:44.929 --> 00:00:49.229 little ironic good you are paying attention let 00:00:49.229 --> 00:00:52.310 us get into it there is a term gaining traction 00:00:52.310 --> 00:00:56.990 in the podcast industry ai slop it refers to 00:00:56.990 --> 00:01:01.009 mass -produced ai generated podcast content created 00:01:01.009 --> 00:01:04.450 without meaningful human curation editorial oversight 00:01:04.450 --> 00:01:08.349 or quality control Shows where the voice, the 00:01:08.349 --> 00:01:11.569 script, the music, and sometimes even the concept 00:01:11.569 --> 00:01:15.129 are entirely generated by AI, then published 00:01:15.129 --> 00:01:18.150 at industrial scale to capture programmatic advertising 00:01:18.150 --> 00:01:21.349 revenue. Think of it as the podcast equivalent 00:01:21.349 --> 00:01:25.510 of SEO spam. Content that exists not to inform 00:01:25.510 --> 00:01:29.030 or entertain, but to game monetization systems. 00:01:29.569 --> 00:01:33.900 How big is the problem? As reported by PodNews, 00:01:34.200 --> 00:01:37.500 the most widely read daily podcast industry newsletter, 00:01:37.879 --> 00:01:41.000 some operations are now producing thousands of 00:01:41.000 --> 00:01:44.340 episodes per week with minimal staff, at production 00:01:44.340 --> 00:01:48.599 costs as low as $1 per episode. These shows become 00:01:48.599 --> 00:01:51.219 profitable after reaching just 20 listeners, 00:01:51.459 --> 00:01:55.120 thanks to programmatic ads. To put that in perspective, 00:01:55.500 --> 00:01:58.439 an independent podcaster might spend 10 hours 00:01:58.439 --> 00:02:02.319 producing a single episode, researching, scripting, 00:02:02.340 --> 00:02:06.260 recording, editing, mixing. These operations 00:02:06.260 --> 00:02:09.740 produce thousands in the same time frame. Is 00:02:09.740 --> 00:02:13.139 all of that inherently wrong? Not necessarily. 00:02:13.680 --> 00:02:16.659 If real listeners find value in the content, 00:02:16.900 --> 00:02:20.360 that is a legitimate question. But here is the 00:02:20.360 --> 00:02:24.639 concern. There is often no disclosure. Listeners 00:02:24.639 --> 00:02:26.919 have no way of knowing the voice they are hearing 00:02:26.919 --> 00:02:30.280 is synthetic. There is no editorial oversight 00:02:30.280 --> 00:02:34.099 catching factual errors, and podcast directories 00:02:34.099 --> 00:02:36.840 are being flooded, pushing independent creators 00:02:36.840 --> 00:02:40.639 down in search results. Meanwhile, a separate 00:02:40.639 --> 00:02:43.000 but related conversation is happening around 00:02:43.000 --> 00:02:46.599 companies like Inception Point AI, headed by 00:02:46.599 --> 00:02:50.740 Janine Wright, formerly COO of Wondery, whose 00:02:50.740 --> 00:02:53.340 large -scale AI production model has drawn both 00:02:53.340 --> 00:02:57.099 attention and debate. As reported by PodNews, 00:02:57.360 --> 00:03:00.180 Wright has characterized critics of AI -generated 00:03:00.180 --> 00:03:03.340 content by saying, I think that people who are 00:03:03.340 --> 00:03:05.900 still referring to all AI -generated content 00:03:05.900 --> 00:03:10.979 as AI slop are probably lazy Luddites. Which 00:03:10.979 --> 00:03:14.620 raises a fair question. Does asking for transparency 00:03:14.620 --> 00:03:19.360 make you anti -technology? The answer is no. 00:03:19.719 --> 00:03:22.460 And the Luddite reference actually helps explain 00:03:22.460 --> 00:03:25.930 why. The original Luddites were English textile 00:03:25.930 --> 00:03:29.349 workers in the early 19th century. Their concern 00:03:29.349 --> 00:03:32.430 was not technology itself. It was about worker 00:03:32.430 --> 00:03:35.830 pay and output quality. They saw automation being 00:03:35.830 --> 00:03:38.990 used to produce lower -quality goods while displacing 00:03:38.990 --> 00:03:42.229 skilled workers. Caring about quality and fairness 00:03:42.229 --> 00:03:45.189 in the face of rapid change is not the same as 00:03:45.189 --> 00:03:47.930 opposing progress. Because this conversation 00:03:47.930 --> 00:03:51.370 often gets polarized, it is worth being explicit. 00:03:52.199 --> 00:03:55.979 AI in podcasting is not just legitimate. Some 00:03:55.979 --> 00:03:59.479 of it is genuinely exciting. The best example 00:03:59.479 --> 00:04:03.419 is voice cloning for multilingual content. A 00:04:03.419 --> 00:04:06.280 podcaster creates brilliant episodes in Italian, 00:04:06.500 --> 00:04:10.300 but only speaks Italian. With AI voice cloning, 00:04:10.479 --> 00:04:13.539 that same host's voice, with its character and 00:04:13.539 --> 00:04:16.439 personality, can deliver the same episode in 00:04:16.439 --> 00:04:20.579 English, Spanish, French, Mandarin. The journalism 00:04:20.579 --> 00:04:24.930 is human. The creativity is human. AI breaks 00:04:24.930 --> 00:04:28.750 the language barrier. That is not slop. That 00:04:28.750 --> 00:04:31.730 is cultural diversity made possible by technology. 00:04:32.250 --> 00:04:35.250 Or think about a podcaster who writes their own 00:04:35.250 --> 00:04:38.329 script, does all the research, structures the 00:04:38.329 --> 00:04:42.189 episode, but uses AI to generate the audio instead 00:04:42.189 --> 00:04:45.589 of spending hours in a recording studio. The 00:04:45.589 --> 00:04:48.750 intellectual property is entirely theirs. AI 00:04:48.750 --> 00:04:52.170 is just the production tool. And then there is 00:04:52.170 --> 00:04:55.769 post -production, noise removal, filler word 00:04:55.769 --> 00:04:59.550 elimination, audio leveling, transcript generation. 00:04:59.970 --> 00:05:03.990 These are standard techniques. Nobody would call 00:05:03.990 --> 00:05:08.850 using a noise gate AI content. The line is about 00:05:08.850 --> 00:05:12.029 the soul of the content. If a human conceived 00:05:12.029 --> 00:05:15.709 it, created it, and curated it, but AI helped 00:05:15.709 --> 00:05:18.829 produce or distribute it, that is a tool being 00:05:18.829 --> 00:05:22.720 used well. If AI conceived it, generated it, 00:05:22.800 --> 00:05:25.259 and published it with minimal human involvement, 00:05:25.540 --> 00:05:28.480 and there is no disclosure, that is where the 00:05:28.480 --> 00:05:32.100 concern lies. And again, in the interest of full 00:05:32.100 --> 00:05:35.160 transparency, you are listening to a case that 00:05:35.160 --> 00:05:38.420 sits right in the middle. The research and editorial 00:05:38.420 --> 00:05:41.519 direction behind this episode are human. The 00:05:41.519 --> 00:05:44.480 voice delivering it is not. And we are telling 00:05:44.480 --> 00:05:47.959 you that up front. So if the need for transparency 00:05:47.959 --> 00:05:52.100 is clear, Why is there no standard? Actually, 00:05:52.240 --> 00:05:55.040 a lot has been discussed, just not resolved. 00:05:55.439 --> 00:05:59.740 The Podcasting 2 .0 namespace is an open -source 00:05:59.740 --> 00:06:03.779 initiative on GitHub that defines new RSS tags 00:06:03.779 --> 00:06:06.959 beyond what Apple's original iTunes namespace 00:06:06.959 --> 00:06:11.100 provides. Tags like podcast colon transcript, 00:06:11.439 --> 00:06:15.480 podcast colon chapters, podcast colon person, 00:06:15.660 --> 00:06:19.149 and many others were born there. Since September 00:06:19.149 --> 00:06:22.949 2024, the community has had four major discussions 00:06:22.949 --> 00:06:26.790 about AI disclosure. None have reached consensus. 00:06:27.550 --> 00:06:32.430 Discussion 663, opened by developer Nathan Gathright, 00:06:32.610 --> 00:06:36.829 proposed a granular tag specifying exactly which 00:06:36.829 --> 00:06:40.959 AI tools were used and for what purpose. James 00:06:40.959 --> 00:06:44.680 Cridland, editor of PodNews, argued that AI use 00:06:44.680 --> 00:06:47.779 is not binary and suggested a percentage -based 00:06:47.779 --> 00:06:51.600 approach. Sam Sethi, founder of the True Fans 00:06:51.600 --> 00:06:55.379 podcast app, pushed for simplicity. A yes or 00:06:55.379 --> 00:06:59.300 no flag, like the explicit content tag. Discussion 00:06:59.300 --> 00:07:04.290 669 opened by Daniel J. Lewis, proposed a broader 00:07:04.290 --> 00:07:07.769 disclosure tag covering not just AI, but also 00:07:07.769 --> 00:07:11.389 sponsorships, compensation, and legal declarations. 00:07:11.990 --> 00:07:15.050 Elegant in theory, but it faces the adoption 00:07:15.050 --> 00:07:18.430 problem. Apps will not implement it until hosts 00:07:18.430 --> 00:07:21.870 use it, and hosts will not use it until apps 00:07:21.870 --> 00:07:27.350 surface it. Discussion 731 Opened again by Nathan 00:07:27.350 --> 00:07:32.129 Gathright in February 2026, proposed adding a 00:07:32.129 --> 00:07:35.470 synthetic attribute to the existing podcast colon 00:07:35.470 --> 00:07:40.170 person tag. Values like cloned for AI cloned 00:07:40.170 --> 00:07:43.649 voices of real people and generated for fully 00:07:43.649 --> 00:07:48.629 synthetic personas. And Discussion 735, opened 00:07:48.629 --> 00:07:52.550 by Benjamin Bellamy from Castapod, proposed a 00:07:52.550 --> 00:07:57.459 dedicated podcast colon AI. tag with per -aspect 00:07:57.459 --> 00:08:01.819 disclosure, colon, voice, music, script, editing, 00:08:02.019 --> 00:08:06.720 cover art, and more. Four proposals, four different 00:08:06.720 --> 00:08:09.899 philosophies, granularity versus simplicity, 00:08:10.420 --> 00:08:14.439 new tags versus existing tags, and one persistent 00:08:14.439 --> 00:08:17.399 chicken and egg problem. Someone has to ship 00:08:17.399 --> 00:08:21.500 first. Some companies have started moving. Spreaker 00:08:22.029 --> 00:08:25.329 owned by iHeartMedia, added an AI disclosure 00:08:25.329 --> 00:08:28.550 checkbox to their dashboard. In the absence of 00:08:28.550 --> 00:08:31.430 formal international guidelines, a simple, generic 00:08:31.430 --> 00:08:35.870 checkbox was the safest approach. RSS .com took 00:08:35.870 --> 00:08:40.409 that a step further. In March 2026, they shipped 00:08:40.409 --> 00:08:44.169 AI Disclosure to all their podcasters and wrote 00:08:44.169 --> 00:08:48.070 the signal directly into the open RSS feed using 00:08:48.070 --> 00:08:53.669 the podcast colon TXT tag. That distinction matters. 00:08:53.929 --> 00:08:56.870 When Disclosure lives only in a hosting dashboard, 00:08:57.330 --> 00:08:59.990 it helps the hosting company moderate internally. 00:09:00.429 --> 00:09:04.250 When it lives in the RSS feed, it travels with 00:09:04.250 --> 00:09:07.740 the podcast. Every app that reads the feed can 00:09:07.740 --> 00:09:11.899 see it. It is portable, verifiable, and platform 00:09:11.899 --> 00:09:15.480 agnostic. The implementation is deliberately 00:09:15.480 --> 00:09:19.720 simple. A checkbox per episode. If checked, the 00:09:19.720 --> 00:09:24.139 episode signals true. If not, false. Identical 00:09:24.139 --> 00:09:27.059 logic to the explicit content tag, which has 00:09:27.059 --> 00:09:30.700 worked well for nearly two decades. Apple Podcasts 00:09:30.700 --> 00:09:33.620 has also taken a clear position. Their content 00:09:33.620 --> 00:09:36.940 guidelines state that creators using AI to generate 00:09:36.940 --> 00:09:40.179 a material portion of a podcast's audio must 00:09:40.179 --> 00:09:43.379 prominently disclose this in the audio and metadata. 00:09:43.799 --> 00:09:47.840 They add that creators must not use AI to mislead 00:09:47.840 --> 00:09:50.740 or fabricate narratives. This is significant. 00:09:51.019 --> 00:09:54.240 The largest podcast platform in the world has 00:09:54.240 --> 00:09:57.399 declared that AI disclosure is required, not 00:09:57.399 --> 00:10:00.610 optional. And that validates the urgency of having 00:10:00.610 --> 00:10:03.289 a machine -readable way to carry this signal 00:10:03.289 --> 00:10:06.509 in the feed. So where does the industry go from 00:10:06.509 --> 00:10:10.950 here? The honest answer is forward, imperfectly. 00:10:10.970 --> 00:10:15.570 And that is fine. The explicit content tag started 00:10:15.570 --> 00:10:18.870 as a simple Boolean too. It has served the industry 00:10:18.870 --> 00:10:22.389 for almost 20 years. No one argues it is perfect. 00:10:22.889 --> 00:10:25.889 but it works it gave apps a signal to filter 00:10:25.889 --> 00:10:28.629 on and it gave creators a simple way to self 00:10:28.629 --> 00:10:32.190 -declare ai disclosure can follow the same path 00:10:32.190 --> 00:10:36.269 start simple iterate with the industry formalize 00:10:36.269 --> 00:10:39.730 when consensus emerges what matters most right 00:10:39.730 --> 00:10:43.070 now is that the signal starts appearing in open 00:10:43.070 --> 00:10:47.870 rss feeds not locked inside dashboards not hidden 00:10:47.870 --> 00:10:51.250 in proprietary systems in the open decentralized 00:10:55.169 --> 00:10:58.929 The more hosting companies that adopt some form 00:10:58.929 --> 00:11:01.809 of disclosure, even using different technical 00:11:01.809 --> 00:11:04.970 approaches, the faster this becomes a real standard. 00:11:05.250 --> 00:11:08.470 And the sooner podcast apps have reason to build 00:11:08.470 --> 00:11:10.870 filtering and display features for listeners. 00:11:11.370 --> 00:11:15.710 Nobody is asking for AI bans. Nobody is asking 00:11:15.710 --> 00:11:19.230 for penalties. The ask is simple. a voluntary, 00:11:19.509 --> 00:11:22.309 good -faith declaration that costs nothing to 00:11:22.309 --> 00:11:26.250 make. Check the box or do not. The signal exists. 00:11:26.690 --> 00:11:29.970 It is in the open feed. And that alone changes 00:11:29.970 --> 00:11:33.350 the conversation. The sooner this lives in open 00:11:33.350 --> 00:11:37.769 RSS feeds, the better. This has been Podcasting 00:11:37.769 --> 00:11:41.120 Innovation. Thanks for listening. And yes, you 00:11:41.120 --> 00:11:44.299 just spent 10 minutes listening to an AI voice 00:11:44.299 --> 00:11:47.779 talk about AI transparency. If that does not 00:11:47.779 --> 00:11:50.100 prove the point, I am not sure what does.