Trace Foundation is a student-founded nonprofit based in Massachusetts that builds free, publicly accessible information tools. Their first project, AI Trace, is a community-driven database that helps people research how companies use technology in their products and services. The organization was started by students at UMass Amherst and Colorado State University after finding that most people had no reliable, centralized place to look up this kind of information. Everything they build is free, has no ads, and is maintained by community contributions.
AI Usage at a Glance
Mar 24, 2026
Creative GenPractice documented: Trace Foundation used an AI image generation tool to create a stylized illustration from a personal photograph for a LinkedIn post announcing the AI Trace platform launch in March 2026.
Practice DocumentedView practice →Apr 8, 2026
Data AnalysisPractice documented: AI Trace uses AI to scan news sources daily across 10 industries, reading articles to identify which company is mentioned and what AI practice is described, then flagging findings for a human moderator to review.
Practice DocumentedView practice →Apr 8, 2026
Data AnalysisPractice documented: AI Trace uses AI to power its search function so that results are matched by meaning rather than exact word overlap — for example, a search for 'AI voice acting' can surface results described as 'synthetic speech in games.'
Practice DocumentedView practice →Apr 8, 2026
ProductivityPractice documented: AI Trace uses AI to help moderators review public submissions about company AI practices, automatically suggesting which company a report refers to, whether the practice is already in the database, and what category it belongs to.
Practice DocumentedView practice →AI Trace uses AI to scan news sources daily across 10 industries, reading articles to identify which company is mentioned and what AI practice is described, then flagging findings for a human moderator to review.
The system fetches articles via RSS feeds from industry-specific publications. Each article is processed by Anthropic's Claude Haiku model using industry-aware prompts that recognize domain-specific AI terminology. Extracted information is surfaced to human moderators as candidate entries rather than being published automatically.
AI Trace uses AI to scan news sources daily across 10 industries, reading articles to identify which company is mentioned and what AI practice is described, then flagging findings for a human moderator to review.
AI Trace uses AI to power its search function so that results are matched by meaning rather than exact word overlap — for example, a search for 'AI voice acting' can surface results described as 'synthetic speech in games.'
Have evidence about Trace Foundation's AI practices? Submit a report.
Submit a report →AI Trace is free and nonprofit. Support our work
AI Trace uses AI to power its search function so that results are matched by meaning rather than exact word overlap — for example, a search for 'AI voice acting' can surface results described as 'synthetic speech in games.'
The platform converts every company profile and practice entry into a vector embedding using OpenAI's text-embedding-3-small model. When a user searches, the query is converted into the same format and compared against stored embeddings using cosine similarity. This runs alongside traditional keyword matching and fuzzy text matching. Embeddings are generated once per content update and stored in the database. The embedding model does not retain or train on user queries.
AI Trace uses AI to help moderators review public submissions about company AI practices, automatically suggesting which company a report refers to, whether the practice is already in the database, and what category it belongs to.
The system makes three sequential API calls to Anthropic's Claude Haiku model: one for entity matching (identifying the company), one for duplicate detection (checking existing entries), and one for category classification. User-submitted text is passed separately from system instructions to reduce prompt injection risk. All AI suggestions are presented to human moderators as editable fields; no suggestion is accepted without human approval.