Classify
our mission
Classify exists to give the open web a layer of understanding it has never had. We read the web at the page level, keep that reading current as the web changes, and make it available to the systems that run on it, whether they buy media or run agents
the web still gets sorted from the outside in
Search, recommendation, advertising, and AI training all run on the open web, and almost everything that sorts it starts at the domain and works down. A site gets one label and every page on it inherits that label, so a gear guide on a news site gets bought and sold as news.
Advertisers end up buying sites when the brief asked for pages, and an agent that lands on the same page has no idea what it's holding. Every product downstream inherits the same gap.
Language models changed what a system can know about a single page, and that reading can now run across the whole web and keep running as pages change.
Nobody had built the infrastructure to do that at the scale of the web and serve the result as a call, so we did, and it runs today as ContentGraph.
one page, one record
Every page in ContentGraph carries its own classification with scores, and the record updates as the page does. Atlas reads that record and turns a brief into a segment, and Phoenix hands the same record to a developer or an agent
A label for the whole site says nothing about the page under it, so a gear guide on a news site gets sold as news and a recipe on a retailer gets sold as retail, and the precision you paid for never gets past the front door
Words drift and pages change, so a list written for one campaign is wrong by the next one. The taxonomy has to come out of the page itself, and it has to get rebuilt every time the page moves
research and transparency
We apply the same rigor to reading the web that frontier labs apply to training on it. Accuracy, latency, and coverage get benchmarked before a number reaches a customer, and benchmarks publish with their methodology and their dates
the pledge
The index holds pages and never people. There are no cookies, no identifiers, and no PII anywhere in it, because the signal lives in the content. And where sponsored results exist they are labeled and kept apart, so nothing an advertiser pays for changes what the index returns
based in the bay area
We build in the place where the infrastructure of the web was built the first time, alongside the people building the models that read it now. The problem is large and the neighborhood is good at large problems
join the project
We're early, and we want to hear from people who care about how information is structured at scale. Builders of AI systems that need richer input, publishers who want sharper classification, advertisers tired of blunt targeting, and anyone obsessed with the shape of the web
Talk to Classify