Google Leak 2026 Update: What Google Admitted Under Oath About Rankings and AI Answers

The Google leak broke in 2024, but the rest of the story came later. Google executives described key ranking systems under oath, a federal court put them in writing, and Google's AI answers now lean on the same ranking family. This is what still holds up two years later.

This Google leak 2026 update starts with a surprise: the big news is not a new leak, it is how much of the old one turned out to be true. In May 2024, thousands of pages of internal Google documentation went public and exposed the Google leak ranking signals, the named data fields Google Search keeps about clicks, sites, topics, and links.

Since then, court records from the United States v. Google antitrust case have backed up key parts of that picture with sworn testimony, and the same ranking systems now help feed the AI answers people read inside Google. If you want the first wave of findings, start with our breakdown of the Google 2024 algorithm leak. This guide picks up where that one stopped.

What are the Google leak ranking signals?

The Google leak ranking signals are attributes listed in Google’s internal Content Warehouse API documentation. An automated bot pushed those files to a public GitHub repository on March 13, 2024, and they stayed up until May 7, 2024, which was long enough for people to copy them. According to the 2024 Google Search documentation leak entry on Wikipedia, the files describe 2,596 modules with 14,014 attributes. An attribute is simply a named piece of data, like siteAuthority or hostAge. Think of each one as a label on a box. The leak shows the labels. It does not show how heavy each box is.

That last point is the one most people skip. The documents describe what Google can store and compute, not how much each signal counts in a ranking. They also cover products beyond web search, including YouTube, Google Assistant, and Google Books, so not every field in the Google algorithm leak is a Search ranking factor.

What changed between the 2024 Google leak and 2026?

The biggest change between the 2024 Google leak and 2026 is that a federal court case filled in what the documents left out. The Google algorithm leak landed in the middle of the U.S. government’s landmark search antitrust case against Google, and the two stories keep confirming each other. The United States v. Google trial ran from September to November 2023, months before the files went public. Testimony from that trial had already described click-based systems like NavBoost. The leak then put internal names on those systems, and the court opinions of 2024 and 2025 added details Google had never published. That is the core of the Google leak 2026 story: less mystery, more confirmation.

What did the Google antitrust trial confirm about NavBoost and clicks?

NavBoost is real, it runs on click data, and Google’s own people described it under oath. In the August 2024 opinion, Judge Amit Mehta wrote that NavBoost “pairs queries and documents through memorizing user click data.” In plain words, it helps Google remember which pages people clicked after typing a certain search. The same opinion says that before 2017 Google trained NavBoost on 18 months of user data, and since then on 13 months. It also mentions Glue, which it calls a “relative” of NavBoost.

The leak lines up with that testimony. It lists click fields such as goodClicks, badClicks, and lastLongestClicks. The names suggest what they track: a good click seems to end the search, a bad click sends the person straight back to the results, and the last longest click points to the result that finally solved the problem. The field names come from the leak, while the system itself comes from sworn testimony, which is why NavBoost sits at the top of any list of Google leak ranking signals you can trust.

Scale is what makes this click data so powerful. The court found that Google receives nine times more queries each day than all of its rivals combined, and that 13 months of Google user data equals more than 17 years of data on Bing. Even Google’s newer deep-learning ranking models, like RankBrain and RankEmbed, “were designed with user data and continue to be trained on it,” though with less volume, according to the same opinion.

Which Google leak ranking signals actually hold up in 2026?

The Google leak ranking signals that hold up best in 2026 are the ones confirmed by a second source, either Google’s public documentation or sworn court testimony. The rest are real fields with an unknown weight. Sorting them this way keeps you from building a strategy on a name that might barely matter. Google’s public ranking systems guide is the easiest cross-check, because it lists the systems Google openly admits to using.

The pattern is useful. Signals about links, originality, and exact match domains were never really secret, because Google already documents systems for them. The real news in the Google algorithm leak was the click memory, the site-level scores, and the topical focus measures, which Google had not described in public before.

How do the Google leak ranking signals connect to AEO?

The Google leak ranking signals connect to AEO (Answer Engine Optimization) because Google’s AI answers pull from the same index and the same ranking family. Google’s own guide to AI features and your website says a page must be indexed and eligible to show a snippet to appear as a supporting link in AI Overviews or AI Mode, with no extra technical requirements. In other words, if classic Search cannot find and trust your page, the AI layer will not cite it either.

Google also explains that AI Overviews and AI Mode may use a “query fan-out” technique, which means running several related searches across subtopics at the same time. Each of those mini-searches is a new chance to be retrieved. A page that answers one clear sub-question per section gives the system more places to grab. Google’s public passage ranking system, which scores individual sections of a page, points the same way.

What is FastSearch, and how does it ground Google’s AI answers?

FastSearch is the Google technology that connects ranking signals to AI answers most directly. In the September 2025 remedies opinion, citing testimony from Google Search head Elizabeth Reid, the court wrote that to ground its Gemini models, Google uses a proprietary technology called FastSearch, which “is based on RankEmbed signals” and returns short, ranked lists of web results that a model can use to build a grounded answer.

The opinion adds that FastSearch is faster than regular Search because it retrieves fewer documents, and that its quality is lower than fully ranked results. RankEmbed is one of the models the 2024 opinion says is trained with user data. The court does not list every product that uses FastSearch, but Gemini is the model family behind Google’s AI features, so signals shaped by real searchers help decide which pages an AI model even gets to read.

The audience for that AI layer is huge. Google reported more than 1.5 billion monthly AI Overviews users in April 2025, more than 2 billion in July 2025, and more than 2.5 billion by May 2026, while AI Mode passed 1 billion monthly users in its first year. The systems exposed by a 2024 leak now sit upstream of answers that billions of people read every month.

What happens to click signals when AI answers show up?

Clicks on regular results drop when an AI summary appears, and that matters because NavBoost learns from clicks. A Pew Research Center study of 900 U.S. adults, based on their real browsing in March 2025, found that people clicked a regular search result in 8% of visits when the page showed an AI summary, compared with 15% when it did not. They clicked a link inside the AI summary itself in just 1% of visits, and they ended their browsing session more often: 26% of the time, versus 16% without a summary.

Google tells a different side of the story. Its AI features guide says clicks from results pages with AI Overviews are higher quality, meaning people tend to spend more time on the site they visit. In July 2026, Google also said it sends billions of clicks to websites every week through AI features in Search. Both things can be true at once: fewer clicks overall, but clicks with more intent behind them.

What nobody outside Google knows is how click-based systems adjust when fewer people click. Google has not said how NavBoost or Glue treat a search where the AI answer solved the problem and no one clicked at all. That is an open question, not a reason to panic, and anyone who claims to know the answer is guessing.

Do the Google leak ranking signals and GEO research point the same way?

Yes. On content quality, the Google leak ranking signals and GEO (Generative Engine Optimization) research point in the same direction, even though they come from completely different places. The leak lists fields such as contentEffort, a model-based estimate of how much work went into a page, and OriginalContentScore. On the research side, the paper “GEO: Generative Engine Optimization” by Aggarwal and colleagues, presented at the KDD 2024 conference, tested nine ways of rewriting web content across 10,000 queries to see what made a generative engine cite a source more.

Adding quotations raised visibility by about 41% over unoptimized text. Adding statistics raised it by about 31%, and citing sources by about 27%. Keyword stuffing, one of the oldest tricks in SEO, made things worse and cut visibility by about 8%. Google’s own spam policies have treated keyword stuffing as spam for years. That is the overlap: effort, evidence, and originality help in both worlds, and padding hurts in both.

One caveat matters here. The GEO study used a research setup built on GPT-3.5, not Google’s live systems, so it shows a direction, not a Google rule. We apply those findings with care in our Generative Engine Optimization work, always paired with real measurement.

Why do question-style pages match how AI answers get triggered?

AI answers show up most often for questions, which is why question-and-answer formatting matters so much for AEO. In the same Pew study, 18% of all Google searches produced an AI summary. That share jumped to 53% for searches with 10 or more words, and to 60% for searches that started with a question word like who, what, why, or how.

This is where QFO (Question-Format Optimization) earns its place. A heading written as a real question, followed by a first sentence that answers it directly, matches the kind of search most likely to trigger an AI summary. It also matches how query fan-out breaks a big question into smaller ones. We explain the full method in our guide on how to write content that AI summaries pull from.

What should you do with the Google leak ranking signals in 2026?

The practical side of the Google leak 2026 story comes down to six moves, each tied to a signal that has at least some outside support.

Fix pages that already rank before you publish new ones. NavBoost remembers how people react to your results over a rolling 13-month window. A page that ranks but sends visitors back to Google keeps teaching the system the wrong lesson, and that memory takes months to change.

Keep your site focused on a clear topic. Fields like siteFocusScore and siteRadius suggest Google measures how tightly a site sticks to its core subject and how far each page drifts from it. Our topical authority work starts by mapping that territory before adding a single page.

Add proof, not padding. Put original numbers, real quotes, and named sources on your key pages. That lines up with contentEffort and OriginalContentScore in the leak, and with the biggest wins in the GEO study.

Write one answer per section. Query fan-out and passage ranking both reward sections that answer one question well on their own. You can see how that works in practice on our Answer Engine Optimization page.

Protect indexing and snippet eligibility. AI features only cite pages that are indexed and snippet-eligible, so a stray noindex tag, a page blocked in robots.txt, or an overly strict nosnippet setting can remove you from classic results and AI answers at the same time. A technical SEO audit catches these problems before they cost you traffic.

Treat leak fields as hypotheses. Test your changes and measure them in Google Search Console instead of chasing an attribute name. The leak tells you where to look, not what will happen.

What does the Google leak still not tell us in 2026?

Even in 2026, the Google leak does not reveal weights, formulas, or how signals combine, and that limit shapes everything else. When Google confirmed the documents were authentic on May 29, 2024, it also warned against assumptions based on out-of-context, outdated, or incomplete information. That warning protected Google, but it also had a point.

Three more limits are worth keeping in mind. First, the files are a snapshot from early 2024, before AI Mode launched, so some systems have likely changed since then. Second, even basic names are debated: NSR, a site-level quality signal, has been expanded in more than one way by different analysts, and Google has never confirmed which reading is right. Third, the court case is not over. Both Google and the Department of Justice have appealed parts of the ruling, so the public record about Google leak ranking signals may still grow.

The leak tells you what Google measures. The court record tells you what Google relies on. Neither one tells you the weights.