Userplay is launching soonGet Started
Userplay
Discord0.8kGet Started
Browse security docs Toggle navigation

// Security · Architecture

AI Processing

Last updated May 31, 2026

Userplay optionally produces two AI-generated outputs from a session recording:

  1. Audio transcript — a searchable, time-aligned text version of what the tester said, produced by OpenAI’s transcription API.
  2. Video analysis — structured insights (themes, friction points, sentiment, notable moments, session summary) produced by Google Gemini.

Both paths are disabled if the studio has opted out of AI processing for the workspace or the specific playtest. When opted out, recordings are still captured and stored — only the AI processing steps are skipped.

When AI Processing Runs#

The transcription path runs only if:

  • The playtest has AI processing enabled (the default), and
  • The recording contains an audio track (the tester consented to microphone capture).

The video analysis path runs only if:

  • The playtest has AI processing enabled, and
  • The recording exists in Mux (and typically a transcript has been produced).

What the Subprocessors Receive#

OpenAI (transcription):

  • Audio segments extracted from the playtest recording.
  • A Userplay-owned API key.
  • Metadata required for the API call (model name, language hint if any).

OpenAI does not receive tester PII beyond what may be present in the audio, nor any video frames or identifiers linking the audio to a specific Userplay workspace.

Google Gemini (video analysis):

  • Relevant frames or short clips from the recording.
  • The associated transcript text (if available).
  • A structured prompt and output schema.
  • A Userplay-owned API key.

Gemini does not receive tester credentials or any identifiers linking the frames to a specific Userplay workspace.

No Model Training#

Userplay’s agreements with OpenAI and Google exclude API submissions from model training. Userplay does not train its own machine learning models on customer data.

Failure Handling#

If transcription or video analysis fails (network error, rate limit, model error):

  • Failures are retried with exponential backoff.
  • Persistent failures surface as a non-blocking warning on the recording in the UI.
  • The underlying recording remains fully viewable; only the transcript or insights panel is absent.

Opting Out#

AI processing can be disabled per workspace or per playtest. Contact support@userplay.io to configure a workspace-level opt-out, or use the per-playtest toggle in the playtest settings. When opted out, no audio or video is sent to OpenAI or Gemini.

Optional Chrome Extension#

For web-based and AI-driven games, studios can enable an optional Chrome extension that captures post-gameplay in-game telemetry (conversations, model calls, structured events the game emits). The extension is double opt-in — both the studio and the tester must explicitly enable it. The extension operates only after the gameplay session ends and never captures credentials, cookies, or browsing history outside the playtest context. Captured telemetry is stored alongside the recording and follows the same retention policy.