If someone tells you the Yandex algorithm leak exposed Google's ranking factors, that's not quite what happened. The leak was real, the code was real, and it told us a lot about how Yandex ranks pages. But Yandex and Google are different companies running different search engines, and a leaked Yandex repository can't hand you Google's source code. The honest answer to does Yandex leak apply to Google is: only in broad strokes, and only where Google's own documentation happens to confirm something similar. The evidence here is mixed. Some categories of signals clearly matter to both engines. The specific features, weights, and thresholds in the leaked code are a different story, and they mostly stay Yandex's.
This matters because a lot of Yandex algorithm leak SEO advice has quietly started treating Yandex leak ranking factors as if they were a cheat sheet for Google. If you're trying to figure out which claims to trust, you need to know where that comparison holds up and where it falls apart.
What actually leaked, and when
In January 2023, a torrent of Yandex source-code repositories showed up online. Ars Technica covered it at the time and reported the dump was close to 45 GB. The files inside were dated to February 2022 and appeared to have been taken that July, so this wasn't a live, current snapshot of Yandex's ranking system. It was a chunk of code as it existed roughly six months before the leak, not a controlled release of a complete and current ranking manual.
Search Engine Land's initial report focused on one file containing 1,922 ranking-factor entries. That number got repeated everywhere, but Search Engine Land's own follow-up reporting clarified that the 1,922 figure was just one factor set inside a much bigger codebase. Across multiple files covering different parts of Yandex's query processing and ranking, the broader count came to 17,854 factors, plus roughly 2,000 more sitting in Jupyter notebooks outside the core code. Those are different scopes measuring different things, not two competing counts of the same list. Treat the 17,854 figure as a reported codebase-wide total, not as proof that 17,854 unique, independent, active rules are running in production today.
A separate factor-list site took the leaked repository, converted the entries into JSON, and translated the labels into English. That's a useful index into the code, but a translated variable name is not a full explanation of how a production ranking pipeline actually behaves.
What the leaked factors appear to cover
Once researchers dug into the code, a pattern emerged. The factors clustered into a handful of recognizable families: PageRank and other link-related signals, text and query relevance, content freshness, URL and host characteristics, and user-interaction data pulled from Yandex's own analytics product.
Some of the specifics reported at the time:
- Links and authority. PageRank and link-age signals showed up, along with features tied to where a link sits on a page. Search Engine Journal's February 2023 analysis found a factor suggesting links from a homepage carried more weight than links buried on an internal page.
- Text and query match. The translated list includes weights for how often query words appear in a document's text, in its links, and in different word forms. That's evidence of a feature family existing, not a formula you can copy for keyword placement.
- Freshness. Document-age and update-time features showed up, with extra weight apparently given to freshness on news-style queries.
- URLs and hosts. Reporting pointed to factors tied to numbers in URLs, trailing slashes, capital letters, and host-level reliability. Ars Technica's coverage also mentioned signals around keywords in URLs and overall site structure.
- User behavior. Clicks, dwell time, and data from Yandex Metrika (Yandex's own analytics tool) all appear as inputs. These are specific to how Yandex processes its own click data, not a blanket instruction to chase a higher click-through rate everywhere.
- Other factors. Reports mentioned a factor tied to the presence of advertising on a page, something related to domain-level clickability, and a signal that seemed to favor Wikipedia specifically.
None of this is a complete inventory of what Yandex actually runs today. The 1,922-entry file itself had 988 entries tagged deprecated and 244 tagged unused, which is about 64% of that one list marked as not currently in play. The remaining roughly 690 entries shouldn't automatically be assumed active either, since a factor's real status depends on the code context around it, something a translated label can't tell you on its own.
The counts and coefficients, read carefully
Search Engine Journal also found specific numeric coefficients attached to some factors: FI_MAX_WORD_HOST_CLICKS at roughly +0.345, FI_PAGE_RANK at about +0.183, FI_ADV around -0.251, and FI_DATER_AGE near -0.207. These read like hard numbers, which makes them tempting to quote as if they settle something. They don't. The article's own characterization was that these were initial scoring weights, likely from an early relevance-scoring stage, with the downstream processing not fully understood from the code alone. That means you can't treat them as the final weights in Yandex's complete ranking system, and you definitely can't use them to infer a comparable number for Google.
The same caution applies to every big number this leak produced. A factor name can sound more conclusive than it is. An entry might be deprecated, unused, scoped to one processing stage, dependent on other signals you haven't seen, or only meaningful in combination with a model that wasn't included in the leak at all. Ars Technica noted that Yandex's anti-spam code wasn't part of what leaked, which is one more reason to be cautious about treating this as the whole picture.
Where Google's own evidence shows overlap
Here's where it gets more useful. Google publishes a fair amount about how its own ranking systems work, and some of that documentation covers the same broad territory as the Yandex factors. That overlap is real, but it's a comparison of categories, not of implementations.
Google's Search Central ranking-systems guide describes its ranking as happening primarily at the page level using varied signals, with site-wide signals and classifiers also playing a role. The guide names specific systems, including link analysis and PageRank, freshness systems, and neural matching for query understanding. Separately, a Department of Justice trial exhibit recorded a conversation with a Google engineer that described Navboost, a signal based on how often users click a result for a given query, segmented by location and device, using a rolling window of the most recent 13 months of data. That same exhibit described PageRank as still contributing to page-quality signals and mentioned Twiddlers, a system that re-ranks results after an initial set has already been selected.
Put side by side, a few categories line up:
| Yandex leak category | Independent Google evidence | Fair conclusion |
|---|---|---|
| PageRank and link signals | Google names link analysis and PageRank in its own guide, and the trial exhibit says PageRank still feeds quality signals | Both engines document a link/PageRank-related concept, not identical implementations |
| Query and text relevance | Google's documentation describes query meaning and relevance, with named systems for text understanding | Both need to connect queries to relevant content, using their own separate methods |
| Freshness | Google documents freshness-related ranking systems; Yandex's code shows age and update-time features | Freshness is a shared problem area for both engines |
| Clicks and user interaction | Navboost uses click activity segmented by location and device over a defined time window | There's a documented click-derived signal family on both sides, not the same data or system |
| URL and host signals | Google's guide names domain-name wording as one of many relevance factors | A narrow, specific Google signal is confirmed; Yandex's broader list of URL quirks is not |
That's the limit of what you can honestly claim. Both engines appear to care about links, freshness, relevance, and clicks as broad categories. Neither the leak nor Google's documentation tells you that the two engines treat any of those categories the same way underneath.
Why the details don't transfer
The core issue is architecture. Yandex and Google are two independently built search systems, not two copies of one shared algorithm. The leak exposed fragments of a particular Yandex codebase, including multiple factor files for different parts of its query-processing pipeline. Google's own documentation describes a different set of systems, classifiers, and query-sensitive weighting, built by a different company with a different history and a different scale of data. The Yandex leak never touched Google's source code, because it wasn't Google's code to leak.
There's also a model-history wrinkle worth knowing. Yandex's own documentation describes MatrixNet, a machine-learning method it says was used for ad-click prediction starting in 2011, built on more than 120 factors. That page doesn't establish that the same model or factor set was the production ranker for organic web search represented across every leaked file, and Search Engine Journal's analysis noted that CatBoost had reportedly superseded MatrixNet by 2017 anyway. So even within Yandex's own history, the leaked code captures one point on a system that had already moved on by the time anyone saw it.
This is also a good moment to separate correlation from confirmed mechanics. A frequently cited 2005 study on clickthrough data found that when researchers deliberately reversed a result ranking, users clicked on less relevant results than they did under a normal or swapped ordering. That's evidence that result quality and position shape what people click, which is a useful thing to know about user behavior in general. It's not proof of how either Yandex or Google uses click logs in 2026. Keep those two kinds of evidence, behavioral correlation and confirmed mechanism, in separate buckets whenever someone cites this study to make a bigger claim than it supports.
The verdict
Pulling the Yandex algorithm leak SEO conversation together: the most defensible way to put it is that the Yandex algorithm leak is a useful case study in the kinds of signals one real search engine combines, and it works as a comparison checklist against what Google independently confirms about itself. It is not evidence that Google uses the same factors, the same weights, or the same thresholds. Broad categories transfer only where Google's own sources independently back them up. The specific implementation details, numeric coefficients, and claims about which factors are currently active do not transfer from Yandex to Google on the evidence available.
What this doesn't support: that the leak revealed Google's algorithm, that it exposed actual Google ranking weights, that it proved a specific tactic works on Google, or that every factor named in the leak is still live anywhere. The separate 2024 Google leak is its own story and a different source entirely, worth reading about on its own terms rather than folding into this one.
If you're deciding what to actually do with any of this, the practical move is to treat the Yandex leak as background context on how complex and probabilistic modern ranking algorithms tend to be, not as a literal playbook. When you want to know whether something affects your Google rankings specifically, check Google's own documentation or direct evidence about Google, the kind you'd use to settle whether backlinks still matter or whether click-through rate functions as a ranking factor. That's a slower habit than quoting a leaked coefficient, but it's the one that holds up when someone asks you to defend the claim later.
A lot of the AI search coverage sitting around these older leaks would benefit from the same discipline on the AI-citation side too: understanding what engines like ChatGPT or Perplexity are actually weighing when they decide what to cite, rather than guessing from an unrelated leak six degrees removed. DeepSmith tracks exactly that, how your brand shows up across AI engines and which pages they're citing when it happens, so you're working from your own data instead of a decade-old repository from a different company's search engine.
