Bits in Extensions – policy clarification for AI-generated output referencing other opted-in users

Hi all,

Before investing in development, I’d like to confirm a few policy points for a Bits-enabled video overlay Extension. I’m describing the mechanic generically:

  • Viewers explicitly opt in (identity shared via Request Identity Link) to be “eligible participants”. They can opt out at any time.
  • The Extension shows a pre-populated list of recently active, opted-in participants (no free-text input, no custom usernames).
  • A viewer uses a fixed Bits product (single SKU) to activate an effect on a selected participant. - The EBS takes that participant’s most recent chat message (read via EventSub channel.chat.message with broadcaster authorization) and generates a short text output with an AI model, inside strict, Extension-defined content rules plus automated moderation.
  • The output is shown briefly on stream with both usernames (activating viewer + participant) and read via TTS. Audio plays through a broadcaster-added browser source, not inside the Extension iframe.
  • The broadcaster can require manual approval before display, skip/remove items, set cooldowns, block participants, or disable the feature.

Questions:

  1. Does using another user’s existing chat message as AI input fall under 6.2.8 (“experiences specified by broadcasters or end users”), given the viewer only selects from a pre-populated list?
  2. Is that chat message considered “user content” under Section 7? If so, is an optional broadcaster pre-approval mode sufficient for 7.2?
  3. Is AI-generated output (unique per transaction) acceptable as a Bits product, as long as the product itself is fixed and Extension-defined?
  4. Is it acceptable for a Bits-activated experience to be delivered visually/audibly via a broadcaster’s browser source rather than inside the Extension, given 2.4?
  5. Would a variant where the participant is selected randomly by the Extension (viewer has no choice) be viewed more favorably?
  6. Are there restrictions on sending chat message text from opted-in users to third-party AI/TTS providers, beyond disclosing it in the privacy policy?

Any guidance from staff or devs who’ve passed review with similar mechanics would be much appreciated. Thanks!