New court documents from the DOJ v. Google antitrust case have revealed a rarely discussed truth about Google’s search rankings: clicks matter, but they’re not treated equally.

Google has long collected data on how often a result is shown versus how often it’s clicked—a signal known internally as Impressions-to-Clicks Ratio. But now we know they also created bias correction mechanisms to avoid overvaluing links just because they were higher up on the page.

What the DOJ Docs Show

The redacted exhibit, first shared by Gagan Ghotra, includes this striking passage:

“An early signal measured how many times a link was shown vs how many times it was clicked. It was found that this measure was biased by link position… The use of [REDACTED] would create a system that reinforced this ranking… [REDACTED] modification developed by a Google engineer that calculated [REDACTED] (and avoiding bias created by) link position.”

In plain English: Google acknowledged that just because a result was in position #1 and got more clicks didn’t mean it was actually more relevant. So they built an internal adjustment system to normalize click data by position.

This reveals that Google has been working to account for positional bias—the fact that higher-ranked results get more attention simply because they’re higher up, not necessarily because they’re better.

Why This Is a Big Deal

For years, SEOs have debated whether user signals like CTR (click-through rate) play a role in rankings. Google has remained vague. But this document confirms:

  1. Google does collect impression and click data
  2. Google found it biased and built a solution to counteract that bias
  3. Therefore, click data is used—but only after being corrected for ranking position

This strengthens the case for optimizing titles and descriptions to attract clicks, but also shows Google is carefully modeling user behavior beyond the raw numbers.

Additional Revelations From the DOJ Hearing Notes

The same batch of documents also shed light on internal Google ranking tools and frameworks:

  • Q*: Pronounced “Q-star,” this is Google’s internal metric for document quality
  • Debugger Window: Google engineers can view live weights of ranking signals for any query
  • Okapi BM25: Google’s traditional relevance scoring method
  • RankEmbed: A neural dual encoder model that matches queries and documents using vector embeddings
  • Navboost: A QD (query-document) table tracking user activity frequencies—not just query matches

What’s Really Meant by “User Activity”?

As Richard Hearne noted on X, the most important takeaway is Google’s definition of activity. It’s not limited to clicks—it includes behavior beyond the SERP, such as:

  • Scrolling in Google Discover (stored as topic-based interactions)
  • Return visits
  • Dwell time
  • Session patterns
  • Topic clustering via user journeys

Gagan Ghotra confirmed this, stating that Google tracks Discover behavior not as page scrolling, but as topic scrolling. This suggests that Google’s AI is increasingly clustering content semantically, not just by URL structure or keyword proximity.

How to Apply This as an SEO

Here are some strategic takeaways for SEOs and content strategists:

1. Don’t Obsess Over Position-Based CTR

Google knows top positions get more clicks just because of visibility. Instead, focus on engagement and content clarity that earns corrected clicks—meaning genuine relevance signals.

2. Optimize for Topic Clusters

With Discover and Search integrating more semantic understanding, group your content around themes, not just keywords.

  • Use internal linking to guide topical journeys
  • Include related subtopics in pillar content
  • Monitor which content is surfaced together via tools like GSC and Discover Insights

3. Write for the Click—But Deliver the Value

Google’s models are getting better at judging content not by clicks alone, but by what happens after the click. Think bounce rate, scroll depth, and revisit behavior.

4. Structure Content for RankEmbed & Summarization

As Google leans more on models like RankEmbed and AI Overviews, structured, concise, and well-labeled content becomes essential. Use:

  • Headers that mirror user questions
  • Summary boxes or TL;DR sections
  • Clear definitions and explanations of terms

Final Thought

Google isn’t just ranking documents—it’s modeling attention, interest, and value. The DOJ documents confirm that click data is part of the equation, but it’s filtered through sophisticated systems to eliminate noise and bias.

As Google integrates AI deeper into its stack, SEO will be less about tricking algorithms and more about earning genuine engagement across the full user journey.

The future of search isn’t just about relevance. It’s about resonance.