Personalization signals
A signal is a rule that turns something you already know about a profile — an attribute, a segment — into a typed value the recommender can use. Signals are how you tell the model what matters, without writing any code.
Deriving a value
You pick a source — a profile attribute type or a segment — and write a short rule that reads the value along with how much to trust it. Gate on confidence, gate on whether it's verified, or reshape it entirely. A signal that comes back empty simply sits out; it never blocks a save.
Typed values
A signal has a type, and the type decides how the model reads it — and whether it can also group people.
A single label — a favorite topic, a tier. Can also key a cohort.
A set of labels — the topics someone follows. A model feature.
A measured value, bucketed and normalized before the model sees it.
A binary flag — subscriber or not. Can also key a cohort.
Two jobs
Every signal can do two things, and you choose which. As a feature, it feeds the model's picture of a person, sharpening their ranking. As a cohort dimension, it joins a compact key that powers "people like you" — so viewers who share signals share behavioral recommendations.
Feeds the model — sharpens this person's ranking.
Groups similar people for "people like you".
Learned interest
Signals aren't only about what you already store. When someone rates content highly, that interest is written back to their profile as a learned attribute — with a confidence that grows the more they repeat it, but never reaches certainty. Over time, a person's own behavior personalizes their feed.
Keep exploring
How candidates are generated, ranked live, and tuned by feedback.
The candidate generators — trending, co-engagement, and the live model.
Named surfaces that assemble a ranked set for any spot in your app.
The trained recommender, its quality gates, and built-in A/B testing.