MOST OF WHAT YOU KNOW ABOUT GOOGLE RANKING IS OUTDATED

In October 2023, the US Department of Justice put Google on trial. And for the first time ever, Google's VP of Search Quality, Pandu Nayak, testified under oath about how Google's ranking systems actually work.

Not blog speculation. Not influencer theories. Sworn testimony.

Here is what every business owner needs to know.

GOOGLE'S SEARCH STACK: THE 4-STAGE FUNNEL

When you type a query into Google, here is what actually happens, confirmed under oath during the United States v. Google LLC antitrust trial:

STAGE 1: INDEX RETRIEVAL

Google's index stores hundreds of billions of documents. When you search, the system walks through what is called an "inverted index," matching words from your query against a massive postings list. The system retrieves tens of thousands of potentially matching pages.

As Nayak testified: the index is sorted so that likely high-quality pages appear earlier, sometimes sorted by PageRank. The system cannot walk the entire list because it would take too long.

STAGE 2: FIRST-PASS RANKING

Core ranking signals kick in here: topicality signals, PageRank signals, localization signals. Together, they score tens of thousands of documents and extract the top few hundred.

NavBoost, Google's click memorization system, plays a major role at this stage. It memorizes past clicks for all queries received in the prior 13 months (it was 18 months before 2017).

STAGE 3: DEEP LEARNING RE-RANKING

This is where Google's three primary deep learning models take over:

  • RankBrain: Looks at the top 20-30 documents and adjusts their scores. It is too computationally expensive to run on thousands of results, so it only runs on the final shortlist. Trained on click and query data across all languages, then fine-tuned on human evaluator data.
  • DeepRank (BERT-based): Google's application of BERT for ranking. It handles language understanding and common-sense reasoning. Also trained on user data.
  • RankEmbed BERT: Identifies additional documents that traditional retrieval might have missed. Trained on click and query data, fine-tuned on human rater data.

STAGE 4: SERP ASSEMBLY (TANGRAM)

The final stage assembles the search engine results page. A system internally called "Tangram" (previously called "Tetris") uses multiple signals, including a system called "Glue," to determine which search features (knowledge panels, featured snippets, video carousels, etc.) appear alongside the organic results.

THE THREE SIGNALS THAT CONTROL EVERYTHING

According to DOJ Trial Document 1436, Google's ranking revolves around three top-level signal categories:

  • Popularity (P*): Measures how popular a page is based on Chrome browsing data and the number of authoritative links pointing to it. This is not just backlinks. It is real user behavior data from the Chrome browser.
  • Topicality (T*): Measures how relevant a document or web page is to a set of queries. This is where keyword relevance, semantic matching, and content alignment come in.
  • Quality (Q*): Measures a page's quality by evaluating authority and trustworthiness through multiple signals. This is the signal that determines whether your page gets indexed, stays indexed, or gets quietly removed.

GOOGLE DOES NOT UNDERSTAND YOUR CONTENT. IT FAKES IT.

One of the most striking revelations from the trial came from an internal Google exhibit (Trial Exhibit UPX0203). It contained a blunt admission:

We do not understand documents. We fake it. Today, our ability to understand documents directly is minimal. So we watch how people react to documents and memorize their responses.

This changes the game for content creators. Google does not read your blog and think "this is great content." It watches what users do after they click, and it memorizes that behavior.

This is why click-through rate matters more than you think. Why dwell time and pogo-sticking are real signals. Why user satisfaction over time is how Google determines quality.

As Nayak confirmed under cross-examination: "User feedback" means click data. And clicks give a measure of actual user performance that human quality raters cannot replicate.

WHAT THIS MEANS FOR YOUR BUSINESS

Stop optimizing for robots. Start optimizing for clicks and satisfaction. Google's entire deep learning ranking stack is trained on click and query data. If your pages do not earn clicks and keep users engaged, no amount of keyword stuffing will save you.

Links still matter, but not how you think. PageRank is still used to sort the index during retrieval. But Google now measures "Nearest Seed PageRank," the distance from trusted seed pages. Low-quality link building does not move the needle. Earning links from genuinely authoritative sources does.

Quality is the gatekeeper to indexing itself. Your content does not just need to rank. It needs to pass Google's quality threshold to even remain in the index.