User Behavior Signals
Google patents confirm that clicks and dwell time are used to test and re-order search results. Learn what the patents actually say — and what they don't.
What you'll learn
- Understand how Google patent US9740750B1 uses a genetic algorithm to test result orderings
- Learn the difference between what a patent proves and what it merely suggests
- Know which behavioral signals are confirmed inputs: clicks and dwell time
- Avoid overclaiming "Google ranking factors" based on patent analysis alone
Google Tests Result Orders Using a Genetic Algorithm
Patent US9740750B1 describes a system Google uses internally to find the best order in which to apply ranking adjustments ("adjusters") to a base set of search results. It does not describe specific ranking factors — it describes a method for testing sequences of adjustments against each other.
The system works like evolution: it creates several "score improvement lists" (different orderings of the same adjusters), tests pairs of them against historical query logs, and lets the winner "breed" with other high-scoring lists through crossover and mutation. Over many iterations, the system converges on an optimal ordering.
How a "Winner" Is Decided
For each test, the system takes a real historical query, applies list A and list B to the same base results, and checks which ordering placed the result users actually clicked (and spent time on) higher. The winning list gets +1 point, the loser -1. Lists with high scores breed together; low-scoring lists mutate or are discarded.
| Component | Role |
|---|---|
| Score improvement lists | Ordered sequences of ranking adjusters being tested |
| Fitness tester | Compares two lists against real click/dwell-time data |
| Breeder | Combines successful lists (crossover) or mutates weak ones |
| Historical query logs | Real user clicks and dwell time — the ground truth for "winning" |
What This Confirms — and What It Doesn't
This patent confirms two things directly: Google uses historical click data and dwell time as validation signals, and Google's ranking algorithm is not static — it is continuously tested and evolved. It does not reveal which specific adjusters exist, which on-page or off-page factors they touch, or how any single page attribute affects ranking.
Be careful with claims like "Google's ranking algorithm literally breeds itself." That's a fair paraphrase of the mechanism, but it is not a factor you can directly optimize for. The practical takeaway is narrower: behavioral signals matter as a validation layer, not a factor list.
Key Takeaways
- Google patent US9740750B1 describes a genetic algorithm for testing ranking-adjustment orderings, not a list of ranking factors
- Clicks and dwell time are the confirmed behavioral inputs used to judge which test ordering "wins"
- The algorithm evolves continuously — Google does not use one static, fixed ranking formula
- The patent gives no detail on what the adjusters themselves are or how they score individual pages
- Use patent analysis to understand Google's mechanics, not as a literal SEO checklist