Proposal: <podcast:ai> (AI Generation and Assistance Disclosure)

[benjaminbellamy]

AI Generation and Assistance Disclosure

<podcast:ai>

This tag is used to indicate whether and how artificial intelligence was used in the creation of the podcast or episode.

Parent

<channel> or <item>

Count

Multiple

Attributes

• aspect (required): The production aspect being disclosed. See list of standard aspects below.

Values

• yes: AI was used for this aspect
• no: AI was explicitly not used for this aspect
• undefined: The podcaster explicitly indicates that AI usage for this aspect is unknown or undetermined (for example, when using third-party services where AI usage is unclear)
• omitted: When a tag for a specific aspect is not present at all, it means the podcaster has chosen not to disclose information about that aspect

The distinction between undefined and omitting a tag is significant: undefined is an active declaration that the podcaster doesn't know or cannot determine if AI was used (which itself is valuable transparency), while omitting a tag for a specific aspect means the podcaster has chosen not to address that aspect at all.

Standard Aspects

Content-related:

• writing: AI assistance in writing scripts, show notes, or other text content
• title: AI generation or assistance in creating titles
• description: AI generation or assistance in creating descriptions or show notes
• category: AI assistance in categorizing the podcast
• chapters: AI generation or assistance in creating chapter markers
• transcript: AI generation or assistance in creating transcripts
• translation: AI translation of content to other languages

Audio-related:

• voice: AI-generated or AI-modified voices (text-to-speech, voice cloning, voice transformation)
• music: AI-generated music or musical elements
• soundEffects: AI-generated sound effects
• audioEditing: AI assistance in editing audio (removing silences, filler words, etc.)
• audioEnhancement: AI enhancement of audio quality (noise reduction, equalization, mastering)
• audioMixing: AI assistance in mixing multiple audio tracks

Visual-related:

• coverArt: AI generation or assistance in creating cover art or episode artwork
• videoEditing: AI assistance in video editing for video podcasts
• videoEffects: AI-generated visual effects for video podcasts

Production-related:

• research: AI assistance in research and fact-checking
• …

Examples

<podcast:ai aspect="transcript">yes</podcast:ai>
<podcast:ai aspect="audioEnhancement">yes</podcast:ai>

<podcast:ai aspect="voice">no</podcast:ai>
<podcast:ai aspect="music">no</podcast:ai>
<podcast:ai aspect="audioEditing">yes</podcast:ai>

<podcast:ai aspect="writing">yes</podcast:ai>
<podcast:ai aspect="transcript">yes</podcast:ai>
<podcast:ai aspect="audioEditing">yes</podcast:ai>
<podcast:ai aspect="audioEnhancement">yes</podcast:ai>
<podcast:ai aspect="voice">no</podcast:ai>
<podcast:ai aspect="music">no</podcast:ai>

<podcast:ai aspect="coverArt">yes</podcast:ai>

<podcast:ai aspect="voice">yes</podcast:ai>
<podcast:ai aspect="writing">yes</podcast:ai>
<podcast:ai aspect="music">yes</podcast:ai>

<podcast:ai aspect="transcript">undefined</podcast:ai>
<podcast:ai aspect="audioEnhancement">undefined</podcast:ai>

Notes

• When applied at the <channel> level, it indicates the general AI usage policy or common practices for the entire podcast
• When applied at the <item> level, it specifies AI usage for that particular episode, which may override or supplement channel-level declarations
• Podcasters are encouraged to be transparent about AI usage to build trust with their audience
• The absence of a <podcast:ai> tag does not imply any particular stance on AI usage
• Custom aspect values may be used for specialized production workflows not covered by the standard aspects

[picard102]

#731
#669
#663
#531

Some sort of disclosure would be good, but I don't think a tag just for AI is warranted.

[benjaminbellamy]

Thank you for feeding the debate!

I understand your concern, but I believe an AI disclosure tag is essential for the health of the podcast ecosystem.

Here's why:

AI is already being used extensively across podcast production—transcription being the most obvious example, but also in audio enhancement, chapter generation, show notes writing, and increasingly in content creation itself.
This isn't a future concern; it's happening right now at an industrial scale.

We're seeing operations churning out 3,000 AI-generated podcast episodes per week.
This isn't experimentation—it's a content factory model that threatens to flood the ecosystem.

Some listeners won't care about AI usage. But many will.
More importantly, podcasters who craft their content with care and intention deserve a way to distinguish their work from mass-produced AI content.
Without a disclosure mechanism, there's no way for listeners to make informed choices, and no way for quality creators to differentiate themselves.

The Podcast 2.0 namespace has always been about giving creators and listeners more control and transparency. AI disclosure fits squarely within that mission.
We created tags for transcripts, chapters, and funding—all optional, all valuable to those who want them.
This is no different.

This tag is optional. Podcasters who don't want to use it don't have to.
But those who want to be transparent about their production methods—whether they use AI or explicitly don't—should have that option.

The time to address this is now, before the ecosystem is overwhelmed.
The music industry is already struggling with AI-generated content flooding platforms like Spotify and major streaming services, with debates over authenticity, royalties, and artist protection.

With Podcasting 2.0, we have an opportunity to get ahead of this challenge and maintain control over our ecosystem before we face the same struggles.

Additional context:

• Matt Deegan: 3,000 podcast episodes a week of AI
• The Podcast Factory Making 3,000 Episodes a Week
• What Creatives Don't Understand About AI
• ABC Media Watch on AI Podcasts
• Inside the AI Slopocalypse
• Podnews Weekly: What's the Deal with Inception Point AI?

[picard102]

I would still think a general disclosure tag, and amending the person tag to have an AI role would be the best path forward.

[benjaminbellamy]

I see your point about using a general disclosure tag and the person tag with an AI role.

That said, I still believe a dedicated AI tag is necessary. The <podcast:person> tag serves a different purpose—it's about crediting contributors. Having both tags could absolutely make sense and serve complementary roles.

But here's the thing: disclosing that AI voices were used shouldn't force podcasters to fill in <podcast:person> tags. Those are two separate concerns—one is about production methods, the other is about attribution.

Does that make sense?

[nathangathright]

disclosing that AI voices were used shouldn't force podcasters to fill in <podcast:person> tags. Those are two separate concerns—one is about production methods, the other is about attribution.

I agree. A standalone tag would be cleaner, but after seeing no traction on my first disclosure proposal, which was more production-method centric, the feedback I received pushed me to focus solely on disclosing synthetic voices. It was intentionally a drastic narrowing of scope to see if that had a better chance of gaining traction.

A crucial point to emphasize. I think every proposal needs to present why apps/listeners/hosts would find it valuable even if there’s very low adoption rates. I think the success of every tag is dependent on how they address that scenario.

[theDanielJLewis]

There is currently no legal requirement to disclose AI use. But there is legal requirement to disclose financial compensation and affiliations. These required and AI disclosures can be handled with the <podcast:disclosure> tag (#669).

There's no need to create a separate tag only for AI disclosure. It would be better to have a single disclosure tag that can handle all disclosures instead of separate single-purpose tags that do similar things.

So I'd love to see how you'd envision merging these ideas with <podcast:disclosure>.

[Marzal]

I agree that it would be better to add all the cases from #735 (comment) to the <podcast:disclosure> tag (#669).

Tag adoption is complicated so I would prefer more capable tags if the corpus is on the same topic. (fewer tags that can do more). So they are easy to extend one there is some adoption.

And I also thing that the follow tag pointing to a JSON could be used for a more than just follow links. There is probably a good line to draw between when all can be put into a more general metadata.json and when is better to increase the feed size.

transcripts -> single purpose file
chapters -> probably could have a bit more info there.
metadata (instead of follow) -> a lot could be put here. IMHO things that are more than X lines for the feed. So we don't have the feed pointing to 7 external files.