Melky 
Why Google's 'New' Updates Are Decades Old
Too many SEOs chase shiny new terms like “EEAT” or “Query Fan-Out,” coined by Google’s PR and product teams, to sound innovative.
But if you dig into the work of their research and engineering teams, you’ll find these ideas are rooted in years-old patents.
This is the powerful argument made by Koray Tuğberk Gübür, founder of Holistic SEO. As he puts it:
What Google now calls ‘Query Fan-Out’ is exactly what we’ve called ‘Query Network’ for over 3 years; built on the foundation of Query Augmentation… If you study Google’s patents, every ‘new’ announcement becomes a déjà vu moment. You won’t be surprised; you’ll say, ‘I saw this coming.
During his talk at SEO Rej, Koray presented a masterclass three years ago on the patents that form the bedrock of semantic search.
That presentation, “Semantic Search Engine & Query Parsing,” is a roadmap to understanding Google not through its marketing, but through its engineering.
In this series, I will dissect that presentation, exploring the foundational patents that govern how Google understands, processes, and ranks information.
By the end, you’ll see today’s “innovations” for what they are: the logical evolution of principles established over a decade ago.
I will start where the search engine does: Query Parsing.
This is the introduction to the “AI Search? It’s Just Search, Rebranded” series, exploring Google’s foundational patents that power modern search.
The Foundation: How Google Deconstructs Your Search
Before Google can find an answer, it must first understand the question. This process, known as Query Parsing, is the first and most critical step.
As Koray’s presentation (Slide 6) outlines, it’s the process of “understanding the different sections of a query.”
Tuple Extraction Visualized
This isn’t about just matching keywords.
It’s about identifying components:
-
Entity-seeking Query: The search is looking for a specific “thing” (e.g., “Eiffel Tower”).
-
Substitute or Synonym Term: The search engine understands that “car” and “automobile” can be interchangeable in certain contexts.
-
Canonical Query: A single, standardized query version that represents many close variations (e.g., “restaurants near me,” “food close by,” “places to eat in my area” can all be canonicalized to a single intent).
-
Query Character: This affects the entire design of the search engine results page (SERP). Is the intent informational, navigational, or transactional? Is it dominant (everyone searching this wants the same thing) or minor (there are multiple possible intents)?
This seems advanced, but the principles are laid out in a US Patent from 2004, detailed on Slide 7: “Multi-Stage Query Processing.”
This patent describes a system that:
-
Deletes stop words (“the,” “a,” “in”).
-
Stems concrete words (reduces “running” to “run”).
-
Expands words with synonyms and co-occurrence (understanding that “engine” often appears with “car”).
This is the primitive ancestor of modern query understanding. But the same patent family goes deeper.
You can see how these components work in practice when Google processes a real query like “when was martin luther king jr born.”
Beyond Parsing: Query Breadth and Historical Analysis
The patent for “Query Breadth” (Slide 8, filed 2004) explains how Google handles “unknown entities.” If it doesn’t recognize a term, it analyzes the documents where that term appears alongside known terms.
The jump to the next level of understanding required a shift from simple grouping to mathematical modeling, which led to Context Vectors and RankBrain.
This helps it build a “query breadth” and infer meaning from context, a foundational block of named entity recognition.
Then comes “Query Analysis”, which introduces the critical role of historical data.
The same patent family describes scoring documents based on:
-
Selection Over Time: How frequently a document is chosen for a query.
-
Hot Topics: Rising queries can boost the visibility of documents that contain them.
-
Freshness of Documents: Crucially, this is defined as the date of the information on the page, not just the document’s last modification date. This is a key distinction many SEOs miss today when discussing freshness.
-
Staleness as a Positive Signal: For some queries, a document with a long history of being relevant (historical data) can be a positive signal.
-
Overly Broad Pages: The patent warns against pages that try to rank for too many discordant queries, calling it a signal for spam, an early, algorithmic definition of a lack of topical focus.
It All Starts with the Query
What does this all mean?
The sophisticated way Google interprets your search today, identifying entities, understanding nuanced intent, using historical data to weigh results, and even spotting spammy, unfocused pages, isn’t the result of some magical “new” AI.
It’s the product of decades of engineering, built upon foundational patents filed when the web was still in its adolescence.
The “new” models like BERT and MuM are incredibly powerful, but they are more advanced engines running on these same, time-tested railway tracks.
When you understand these fundamentals, you move from reacting to buzzwords to anticipating the logical next steps in Google’s search evolution.
In my next article in this series, I will explain the “Midpage Query Refinements” patent, the blueprint for how Google moved from simply answering a query to actively suggesting what you should search for next.
Share Article
If this article helped you, please share it with others!
Some content may be outdated