A recent U.S. Justice Department deposition featuring a Google engineer has shed light on some of the inner workings of Google’s search ranking systems. Although much of the information is high-level and generalized, it offers valuable insights into how Google evaluates and ranks web pages.
Hand-Crafted Signals in Google’s Algorithm
One key takeaway is the emphasis on “hand-crafted” signals — ranking factors that are manually designed and tuned by Google’s engineering teams rather than being fully automated. This approach allows Google to maintain transparency and control over its algorithm, making it easier to troubleshoot and improve when issues arise. Unlike Bing, which reportedly uses more automated machine learning methods, Google prefers this structured, human-driven process for most of its ranking signals.
The ABC Signals: Anchors, Body, and Clicks
The deposition introduced what’s referred to as the ABC signals , which form the foundation of topical relevance:
- A – Anchors : Refers to links pointing to a page (i.e., backlinks or anchor text).
- B – Body : Involves how well the content on a page matches the search query.
- C – Clicks : Measures user engagement metrics such as dwell time and whether users return quickly to the search results page (SERP).
These three components help determine a document’s relevance to a given query, forming what Google calls a “base score” for topicality.
Page Quality: A Static Yet Crucial Factor
Another notable point is that page quality is largely considered a static factor across queries. Once Google determines a site is trustworthy and high-quality, it retains that status across various searches. However, there are exceptions where query-specific relevance can override or refine this general quality signal — for example, directing users to a more technical source when the query demands specialized knowledge.
Interestingly, the engineer acknowledged that people still complain about the quality of search results, and the rise of AI-generated content has only made things worse. Because the quality score is mostly static and tied to the site rather than the query, it can be reverse-engineered with enough data.
eDeepRank: Making LLM-Based Rankings Transparent
Google also uses an advanced system called eDeepRank , which leverages large language models (LLMs) like BERT to better understand content and relevance. However, because LLMs can act like a “black box,” Google engineers have developed ways to break down their outputs into interpretable components — essentially making AI-based rankings more transparent and actionable.
PageRank Still Plays a Role
Although updated over the years, PageRank remains part of Google’s ranking system. It functions as a signal measuring how close a page is to known authoritative sources (or “seed sites”) within a topic. Pages closer to these trusted sources are generally seen as more credible.
A Mysterious Popularity Signal Based on Chrome Data
Perhaps one of the most intriguing revelations is the mention of a redacted popularity signal derived from Chrome data . While the exact nature of this signal isn’t specified, it hints at the possibility that Google may use real-world browsing behavior to assess a website’s popularity — a factor that could influence rankings indirectly.
Some speculate this might relate to leaked Chrome APIs, though the engineer clarified that while certain internal documents may name components of the ranking system, they don’t provide enough detail to reverse-engineer the full algorithm.
Final Takeaway
This deposition provides a rare, behind-the-scenes look at how Google approaches search ranking. It highlights the balance between hand-crafted engineering and evolving technologies like AI, while reaffirming the importance of relevance, quality, and user behavior in determining search results.
From ABC signals to LLMs and Chrome data, Google’s system remains complex — but intentionally designed to be understandable and maintainable by its engineering teams.