United StatesMedia and entertainment2026 – ongoingScribd, Inc.
Gemini batch classification of uploaded documents for trust and safety
Scribd classified 400M+ documents with Gemini and continued the pipeline for new uploads
Problem
Scribd needed to review a large document collection against content policies. Specialized detection models required substantial engineering, and text-only tools missed visual signals.
Approach
Scribd selected Gemini 2.5 Flash Lite for classification and Gemini 2.5 Pro for a second consistency pass. It submitted PDFs through Gemini Enterprise batch prediction. After completing the historical collection, it continued processing new uploads.
Results
Takeaway
Native PDF input removed preprocessing for most documents. The reported processing volume did not establish classification accuracy.
Categories
- Industry
- Media and entertainment
- Business function
- Production and quality
- Innovation type
- Process innovation
- AI technology
- Generative AI
- Adoption
- Off-the-shelf tool or platform
- Evidence
- Vendor case study
Sources
Source summaries
Google Cloud described Scribd's completed classification backfill and continuing upload pipeline. It reported 400M+ documents classified and more than 99% processed without preprocessing.
Full details are in the original sources.