Open any app on your phone. Scroll for thirty seconds. Everything you see, every recommendation, every suggested connection, every search result, was shaped by a structure you've never seen and probably never thought about.
That structure is a graph.
Two graphs that run the internet
There are two types of graphs quietly powering most of the digital products you use every day, and they do very different things.
A social graph maps relationships between people. When Facebook suggests someone you might know, or when Tinder surfaces a potential match, or when LinkedIn recommends a connection, they're traversing a social graph: a web of nodes (people) and edges (relationships, interactions, mutual friends). Facebook's social graph infrastructure, called TAO, handles over 10 billion queries per second. Every "people you may know," every news feed ranking, every dating suggestion passes through it.
A knowledge graph maps relationships between facts. When Google shows you a card about a person, a place, or a concept alongside your search results, that's a knowledge graph at work. It connects entities (people, places, events, ideas) and the relationships between them (who founded what, what's located where, what's related to what). Amazon uses knowledge graphs to power product recommendations. Spotify uses them to connect artists, genres, moods, and listening patterns.
The social graph knows who you are. The knowledge graph knows what things mean. Together, they shape nearly every personalized experience on the internet.
Why this matters for product people
If you build products, understanding graphs isn't just a technical curiosity. It's a lens for thinking about personalization, discovery, and recommendation at a structural level.
At Tinder, the social graph isn't just "who's connected to whom." It encodes proximity, activity patterns, mutual interests, and implicit signals from swipe behavior. The richer the graph, the better the recommendations. But the graph is only as good as the data that feeds it, which is why onboarding (getting new users to express preferences and behave naturally) matters so much for match quality.
Knowledge graphs are becoming especially important as AI gets smarter. Large language models are powerful but they hallucinate. They generate plausible-sounding answers that are sometimes wrong. Knowledge graphs provide a structured, verifiable layer of facts that can ground AI outputs in reality. As generative AI adoption accelerates, the demand for knowledge graph infrastructure is growing fast, driven by the need to make AI systems more reliable and trustworthy.
The invisible shapes everything
The reason I find graphs fascinating isn't the technology itself. It's that they're invisible by design. No user ever thinks, "I'm querying a social graph right now." They just see a suggestion that feels relevant, a match that feels right, a search result that answers their question.
The best product experiences feel like the product understands you. Graphs are how that understanding is built, one relationship, one connection, one edge at a time. When someone asks me what makes a great recommendation engine, my answer is always the same: it starts with the graph.