Skip to content

Tell admins and editors that audio and video transcription uses OpenAI #206

Open

Description

@smithjw1

In Ideation, an uploaded audio or video source is transcribed by sending the raw file (up to 25 MB) directly to OpenAI's Whisper endpoint, api.openai.com/v1/audio/transcriptions, using the OpenAI key from Connectors (MediaProcessor::call_whisper_api() in includes/integrations/class-media-processor.php). This happens whichever AI provider the admin selected. The summary of the transcript then goes through the selected provider.

Nothing in the UI says so, which causes two problems:

  • A silent send to a provider the admin didn't choose. A site that selected Anthropic or Google but also has an OpenAI key in Connectors sends its media to OpenAI without anyone being told. That breaks Workflows' promise of configurable, transparent AI.
  • A late, confusing failure. A site with no OpenAI key accepts the upload, then the background job fails the source with "OpenAI API key not configured," although the site's selected provider is configured and working.

The WordPress AI client has no speech-to-text capability today, so transcription can't simply follow the selected provider. This issue makes the current behavior visible. It doesn't change which service transcribes.

What to do

  • Where audio and video sources can be added in Ideation, say plainly that transcription uses OpenAI.
  • Say the same in the AI settings where the provider is chosen, next to the provider selection.
  • With no OpenAI key in Connectors, still accept audio and video uploads, but warn before upload that they won't be transcribed, and tell administrators where to add the key. The source then shows a clear error, and Retry transcribes it once a key is added. Image and PDF sources are unaffected. (Changed from the original plan, which refused these uploads: after testing, a missing key shouldn't block an upload.)
  • Keep the server-side check in call_whisper_api(), so a file that reaches the job without a key still fails with a clear message.
  • Update docs/SPEC.md (Outbound calls).

Done when

  • An admin who has selected a provider other than OpenAI can see, before any upload, that audio and video go to OpenAI.
  • On a site with no OpenAI key, an editor is warned before uploading audio or video that it won't be transcribed, and an administrator is also told where to add the key. The upload is accepted, the source shows why it wasn't transcribed, and Retry transcribes it after a key is added.
  • Image and PDF source processing behave as before.

Activity

  1. added this to the Beta 3 milestone on Nov 17, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions