For more than a decade, SEO professionals have built their strategies around a simple premise: Google evaluates only a narrow slice of the web—roughly 20 to 30 candidate pages—before deciding which result earns the coveted spot on the first page. That limitation isn’t a matter of policy; it’s a technical constraint. Recent disclosures from Google’s own research labs suggest the company may soon have a way to cut the computational cost of evaluating far more pages, a development that could reshape the entire SEO landscape.
Why Google Currently Limits the Candidate Set to 20‑30 Pages
The restriction stems from the sheer amount of processing power required to run Google’s most advanced ranking components, especially RankBrain, the deep‑learning system introduced in 2015. During the United States v. Google trial in October 2023, DOJ counsel Kenneth Dintzer cross‑examined Pandu Nayak, Google’s Vice President of Search, and received four unequivocal confirmations:
Q: RankBrain looks at the top 20 or 30 documents and may adjust their initial score. Is that right?
A: That is correct.Q: And RankBrain is an expensive process to run?
A: It’s certainly more expensive than some of our other ranking components.Q: So that’s, in part, one of the reasons why you just wait until you’re down to the final 20 or 30 before you run RankBrain?
A: That is correct.Q: RankBrain is too expensive to run on hundreds or thousands of results?
A: That is correct.
These statements make it clear that the bottleneck is not a strategic choice but a hardware limitation. Running a neural network on every possible match for a query would require orders of magnitude more CPU cycles and energy, something that would be unsustainable at Google’s scale.
The New Research Technique That May Reduce Evaluation Costs
In a paper released by Google’s research division earlier this year, the company outlined a novel approach designed to shrink the computational footprint of large‑scale ranking. While the full technical details are still under peer review, the gist is that Google is experimenting with a two‑stage filtering system that can prune irrelevant results much earlier in the pipeline.
The proposed workflow looks roughly like this:
- Stage 1 – Coarse Scoring: A lightweight model scans the entire index and assigns a rough relevance score to every document. Because the model is intentionally simple, it can process millions of pages in seconds.
- Stage 2 – Fine‑Grained Ranking: Only the top‑N candidates from Stage 1 (where N could be several hundred) are passed to the full RankBrain architecture for deep evaluation.
- Dynamic N Adjustment: The system learns, over time, how many candidates are needed for a given query type, allowing it to expand or contract the candidate pool on the fly.
By front‑loading the filtering with a cheap
Leave a Comment