United StatesMedia and entertainment2025 – ongoingScribd, Inc.
Claude-generated descriptions for user-uploaded documents on Scribd and Slideshare
Scribd used Claude to write descriptions for millions of uploaded documents and reported 7% more sign-ups and subscriptions
Problem
Scribd and Slideshare host large amounts of user-uploaded content. About 70% of it had low-quality or missing metadata, mainly descriptions, which made relevant material harder to find.
Approach
Scribd tested models, used content operations staff for human quality reviews, and weighed throughput and cost. It chose Claude for its balance of quality, cost and high-volume inference. The generated descriptions were added to document pages and later used in homepage recommendations. Search integration was planned.
Results
Takeaway
Human review of sample output let Scribd compare models on quality before it weighed cost. The company did not give a baseline, a time period or a control group for the 7% figure.
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
Anthropic's customer story describes how Scribd, Inc. used Claude to generate metadata for user-uploaded documents on Scribd and Slideshare. It reports that the work addressed the 70% of content lacking quality metadata, scaled to over 100 million documents, and came with a 7% increase in sign-ups and subscriptions. The page carries no date; the Internet Archive first saved it on December 9, 2025.
Full details are in the original sources.