Facebook Introduces Graph Search

Mark Zuckerberg announced Facebook Graph Search on January 15, 2013 at an event at Facebook’s Menlo Park headquarters, releasing it in a limited US English beta. Graph Search was led by Tom Stocky and Lars Rasmussen — Rasmussen was the co-founder and original engineer of Google Maps (sold to Google 2004) and creator of Google Wave (launched 2009, discontinued 2010), who joined Facebook in 2010. The system accepted natural-language queries structured around Facebook’s Open Graph data model: “Photos of my friends taken in Paris,” “Friends of friends who work at Google who went to Harvard,” “Restaurants in New York my friends have visited,” and “People who like The Dark Knight and Star Wars.” The queries were parsed into structured graph traversals over Facebook’s social graph — nodes (people, pages, photos, places, events, groups) connected by typed edges (likes, tagged-in, lives-in, works-at, attended, friend-of) — rather than keyword-based text retrieval. Bing web search was integrated as a fallback when queries lacked sufficient social data to produce useful results.

The engineering challenge was enforcing Facebook’s permission model across every possible query result. Each object in Facebook’s graph had an access control list determining who could see it: a photo tagged “friends only” was invisible to anyone outside that friend circle, even if their friend appeared in it. Graph Search results had to be filtered through these ACLs before being returned — meaning the same query could return entirely different results for different users based on their individual permission scopes, and the filtering had to operate in real time against potentially millions of candidate results. Facebook’s infrastructure at the time processed approximately 1 billion+ queries per day on its primary graph; Graph Search added a structured query layer on top requiring separate index structures and computation pipelines distinct from the News Feed ranking and notification systems. Privacy audits of Graph Search revealed that the feature made previously “findable in theory” information practically discoverable: users who had liked specific pages years earlier (embarrassing brands, controversial content, personal interests shared in a different social context) found those likes exposed through structured queries like “People who like [sensitive topic] who live in [city]” — including queries that could be used to identify LGBT individuals in regions where such identity was sensitive or illegal.

Facebook began winding down Graph Search in June 2019, removing People, Photos, Posts, and Videos search tabs in the US, citing low engagement relative to the engineering investment required to maintain the system. The decision reflected broader challenges: the Open Graph data model had never been adopted widely enough by third-party websites to fulfill its full vision (Mark Zuckerberg had announced Open Graph at f8 in April 2010 as a way to extend Facebook’s structured data across the entire web via meta tags), and Facebook’s own user behavior had shifted toward mobile (where Graph Search’s complex query interface was difficult to use) and toward visual/video content (Stories, Reels) rather than text-and-links content types that graph traversal could search effectively. Graph Search’s ambition — treating a social network as a structured database queryable with natural language — prefigured similar structured data projects at other companies, though the privacy implications it surfaced influenced subsequent social network transparency and data minimization policies.