The Problem: A Stateless SERP#

As Koray highlights, ambiguous queries create a “stateless SERP instance”. The term is an unhelpful engineering term for a results page because it doesn’t understand the user’s context.

Queries involving homonyms (“jaguar”), general terms (“repair”), or improper context lead users to a dead end, forcing them to go back and try again.

The “Midpage Query Refinements” patent was Google’s solution. It outlined a system to dynamically provide suggestions on the results page, helping users refine their search without starting over.

Google built the Midpage Query Refinements patent on four key functional blocks:

• a Matcher

• a Clusterer

• a Scorer, and

• a Presenter

Think of it as a knowledgeable librarian:

1. The Matcher: Gathers all the different ways people ask for something (e.g., “jaguar,” “jaguar info,” “what is a jaguar”).

2. The Clusterer: Groups these varied queries together. But here’s the magic: it also groups the documents that successfully answer these queries. It’s creating topical buckets.

3. The Scorer: Analyzes each bucket’s “center of gravity” to understand its core meaning. Is this bucket about cars or animals?

4. The Presenter: Shows the user the different buckets as suggestion links, allowing them to choose their path.

The Secret Sauce: Centroids and Semantic Clusters#

How does the “Scorer” know what a cluster is about?

The concepts of Semantic Grouping and Centroids come into play to understand what the scorer knows about a cluster.

A centroid is a mathematical concept representing a cluster’s “center” or average meaning. By analyzing the words, phrases, and ideas in all the queries and documents within a cluster, Google can calculate a vector representing its core topic.

For the query “jaguar,” the system would identify at least two major clusters:

• Cluster A (Automotive): Contains documents and queries with terms like “car,” “engine,” “dealership,” “F-TYPE,” and “luxury.” Its centroid would be distinctly automotive.

• Cluster B (Zoological): Contains documents and queries with terms like “animal,” “big cat,” “jungle,” “prey,” and “panthera onca.” Its centroid would be distinctly zoological.

When you search for “jaguar,” Google presents refinements derived from the centroids of these pre-computed clusters.

It’s using the collective knowledge embedded in documents across the web to help you clarify your intent.

A Living Legacy#

Midpage Query Refinements is not some forgotten, dusty patent.

As Koray’s presentation shows, the patent was refreshed in 2017, with updates to the “Scorer” method, making it even better at choosing representative queries for each cluster.

The inventors, Paul Haahr and Steven D. Baker, are legends in search.

Years before the famous “RankBrain” announcement, Baker wrote a pivotal blog post for Google titled “Helping computers understand language,” which explained the concept of using synonyms and query context.

Paul Haahr is renowned for his “How Google Works” presentations, which have given SEOs rare glimpses under the hood.

Their work on this patent is a foundational piece of Google’s semantic journey.

What does this mean for you?#

Understanding the “Midpage Query Refinements” patent changes how you approach content strategy.

Stop thinking in keywords, start thinking in clusters. If you want to rank for “jaguar” (the animal), your page cannot just repeat the keyword. It must be a comprehensive resource that helps Google place it firmly within the zoological cluster. It must contain the co-occurring terms (big cat, jungle, prey) that define that contextual domain.

Topical authority is algorithmic. The strength of your “document cluster” determines whether Google will use your content to answer a query. The more comprehensively you cover a topic, the closer your page is to that topic’s centroid, and the more likely a search engine will see you as an authority.

“People Also Ask” is not random. The questions in PAA boxes are often refinement queries generated from these semantic clusters. By analyzing your main topic’s related searches and PAA questions, you are essentially reverse-engineering the query clusters Google has already identified. Covering these in your content is a direct way to align with the search engine’s understanding.

The search suggestions we see every day are not a new trick. They fulfilled a vision in 2003: to make the search engine a collaborative partner in the user’s journey for information.

In the following article, I will cover an interesting topic: Context Vectors.

And explain how Google moved beyond clustering to understanding the mathematical relationships between words, paving the way for the groundbreaking technology of RankBrain.