{"id":57450,"date":"2026-06-20T23:38:01","date_gmt":"2026-06-21T03:38:01","guid":{"rendered":"https:\/\/overcentral.com\/en\/?p=57450"},"modified":"2026-06-20T23:38:01","modified_gmt":"2026-06-21T03:38:01","slug":"google-url-analysis-ranking-secrets","status":"publish","type":"post","link":"https:\/\/overcentral.com\/en\/google-url-analysis-ranking-secrets\/","title":{"rendered":"Analysis of 3.7M Google URLs Reveals Ranking Secrets and Blacklists"},"content":{"rendered":"

An analysis of over 3.7 million internal Google URLs has pulled back the curtain on the search giant\u2019s ranking architecture, revealing everything from the specific systems that score and re-rank results to the manual blacklists used to suppress controversial content. Researchers at Resoneo<\/a> took an unconventional approach: rather than attempting to bypass login screens or access restricted pages, they treated the URLs themselves as the primary data source, applying classic open-source intelligence techniques to extract information from file paths, hostnames, and query parameters.<\/p>\n

The underlying principle is both simple and powerful. Even without opening a single internal page, the mere existence and naming conventions of Google\u2019s internal URLs reveal sensitive information about the systems, teams, and projects that power the world\u2019s most influential search engine. A path name shows which infrastructure components Google relies on, how engineering teams are organized, and what internal initiatives exist \u2014 all before a single byte of page content is loaded.<\/p>\n

What emerged from the data is a detailed map of Google\u2019s internal operations, confirming many suspicions about ranking factors that had previously been inferred from leaks and patents, while also exposing new details about manual editorial controls, experiment rollouts, and even the physical layout of Google\u2019s facilities.<\/p>\n

Twiddlers and Ascorers: The Core Ranking Architecture Confirmed<\/h2>\n

The analysis provided concrete validation for components of Google\u2019s ranking system that had been theorized for years. One of the most significant confirmations involved the primary information retrieval scorer, internally referred to as the Ascorer<\/strong>. Researchers identified a live debug flag embedded in a URL \u2014 g=eng-hip-ascorer<\/strong> \u2014 that directly pointed to this system in active use, moving it from the realm of speculation to documented reality.<\/p>\n

Beyond the initial scoring mechanism, the URLs also shed light on how Google re-ranks results after the first pass. The so-called Twiddler<\/strong> system, a set of re-ranking mechanisms that adjust scores based on specific signals, was visible through internal documentation URLs. Researchers found references to guides for writing Twiddlers stored under a directory called SuperRoot, indicating that this is a structured, documented framework within Google\u2019s engineering environment \u2014 not an ad-hoc collection of patches.<\/p>\n

For SEO professionals and search observers, this is the first direct evidence from internal URLs that confirms the layered nature of Google\u2019s ranking process: an initial retrieval and scoring pass, followed by a re-ranking layer that applies additional signals and adjustments before the final results are served.<\/p>\n

Manual Intervention and Editorial Blacklists<\/h2>\n

One of the most striking findings from the URL analysis is that Google\u2019s ranking is far from purely algorithmic. The data reveals extensive manual curation through explicitly maintained blacklists and demotion files, challenging the narrative that machine learning alone determines what users see.<\/p>\n

Perhaps the most revealing discovery was a file named youtube_controversial_query_blacklist<\/strong>. The URL contained 42 distinct revision tokens (?cl=… parameters), each corresponding to a manual update of the blacklist. By analyzing these tokens, researchers could identify precisely when sensitive search queries were added to the list. A particularly vivid example emerged from the aftermath of the 2017 Las Vegas shooting: parameters such as \/40mandalay<\/strong> and \/40shooter<\/strong> appeared in the URL, indicating that specific query terms related to the attack had been manually blacklisted at that time.<\/p>\n

This demonstrates that Google maintains a live, manually curated list of controversial search terms for YouTube, updated through a formal revision process \u2014 and that the URL structure itself can expose the timing and nature of those updates without ever accessing the file contents.<\/p>\n

Two Tiers of Spam Enforcement<\/h3>\n

The analysis also revealed that Google applies two distinct levels of enforcement against problematic content, with separate infrastructure for each. The researchers identified two parallel Googlebot lists stored in different directory paths: badurls_spamindex<\/strong>, used for complete removal of content from the index, and badurls_demoteindex<\/strong>, used for lowering the ranking of content without removing it entirely.<\/p>\n

This distinction is significant because it confirms that Google operates a graduated enforcement system. Not all low-quality or policy-violating content receives the same treatment. Some is eliminated entirely, while other content is simply pushed down in the rankings \u2014 a nuance that has important implications for understanding how Google handles spam, misinformation, and borderline content at scale.<\/p>\n

Mendel and Finch: The Experimentation Infrastructure Behind Every Change<\/h2>\n

The URL data also provided an unusually clear view of how Google manages changes to its search engine. Nearly every modification, whether a minor tweak to ranking signals or a major feature launch, passes through a formalized experimentation and rollout pipeline.<\/p>\n

Two platform names appeared repeatedly in the URL paths: Mendel<\/strong> and Finch<\/strong>. These systems handle A\/B testing and staged rollouts, with every experiment logged and documented. Researchers found a URL referencing a KillSwitchExample.gcl<\/strong> file, suggesting that even the ability to rapidly disable a feature in production is managed through the same infrastructure \u2014 a critical safety mechanism for a system handling billions of queries per day.<\/p>\n

The staging and deployment pipeline itself was visible through the URL structure, which consistently followed a progression path: Dev \u2192 Autopush \u2192 Staging \u2192 Preprod \u2192 Prod<\/strong>. This five-stage chain shows that Google maintains multiple isolated environments between development and live deployment, each serving a specific validation purpose.<\/p>\n

AI Mode and Gemini: Revealed Through Staging Hosts<\/h3>\n

Perhaps the most forward-looking finding in the URL analysis concerns Google\u2019s AI initiatives. The existence of AI Mode<\/a><\/strong> and the Gemini<\/strong> development project were visible through staging hostnames long before their official announcements. URLs such as hc-ai-mode<\/a>a-staging.corp.google.com<\/strong> and associated deployment paths made it possible to track the progress of these projects through their infrastructure, years before they became public products.<\/p>\n

This has practical implications for competitive intelligence: the same patterns of staging hostnames and deployment directories that revealed AI Mode and Gemini could, in principle, be used to identify future Google initiatives before they are officially announced, simply by monitoring changes in Google\u2019s internal URL landscape.<\/p>\n

The Human Element: 16,000 Quality Raters and Their Infrastructure<\/h2>\n

Despite massive investments in artificial intelligence and machine learning, Google continues to rely heavily on human judgment for quality control. The URL analysis confirmed the infrastructure supporting approximately 16,000 quality raters \u2014 human evaluators who assess search result quality according to detailed guidelines.<\/p>\n

Raters operate through a platform called raterhub.corp.google.com<\/strong>, a system that the URLs trace back to an internal project codenamed EWOK<\/strong>. More revealing were the complex evaluation URLs such as eval-analytics.corp.google.com\/querygroup?experimentId=…<\/strong>, which show<\/a> that raters are presented with specific groups of search queries \u2014 called query groups<\/strong> \u2014 to evaluate the outcomes of algorithm experiments.<\/p>\n

What is the purpose of query groups in Google’s quality evaluation process? Query groups are curated sets of search queries presented to human raters to assess how well algorithm changes perform across diverse search scenarios. Each experiment ID corresponds to a specific algorithmic change or ranking model being tested, and the rater’s feedback on the query group helps determine whether that change moves to production.<\/p>\n

This reveals a hybrid system where algorithmic experiments are validated through structured human feedback loops, bridging the gap between automated ranking changes and real-world search quality.<\/p>\n

URLs That Map the Physical World<\/h2>\n

One of the more unexpected dimensions of the analysis is how Google\u2019s internal URLs map onto physical infrastructure and organizational structure. The researchers discovered 2,377 internal printers registered under the domain *.printer.in.goog<\/strong>, with names that revealed precise geographic and logistical details.<\/p>\n

Printer names such as 24th-floor-printer<\/strong> or au-syd-erk1a-1-security-truck-entry<\/strong> provide enough information to reconstruct building floor plans, security topologies, and even the physical layout of data centers. The naming conventions are not randomized \u2014 they follow systematic patterns that encode location, function, and access level.<\/p>\n

Similarly, personal employee pages hosted at www.corp.google.com\/~login<\/strong> reveal, without ever being accessed, which engineers worked on which projects. A path like ~daepark\/public\/mustang-suggest\/<\/strong> is sufficient to link a specific engineer to the Mustang<\/strong> ranking system \u2014 one of Google\u2019s core ranking components. Internal bookmark systems, using the go\/<\/strong> URL shortener, further expose the existence of specialized classification algorithms, such as go\/ymyl-classifier-dd<\/strong>, a reference to the “Your Money or Your Life” (YMYL) content classification system that Google uses to apply higher quality standards to pages that could impact a user’s health, finances, or well-being.<\/p>\n

What the URL Landscape Reveals About Google\u2019s Priorities<\/h2>\n

Stepping back from the individual findings, the aggregate picture from the 3.7 million URLs tells a broader story about Google\u2019s operational priorities. The company operates a search ecosystem that is simultaneously deeply automated and heavily manually curated. Algorithmic scoring through Ascorer and Twiddler systems handles the bulk of ranking decisions, but those decisions are constantly being overridden, refined, and audited through manual blacklists, human rater feedback, and staged experimentation pipelines.<\/p>\n

The infrastructure revealed by the URLs also highlights the immense operational complexity behind what users experience as a simple search box. The five-stage deployment pipeline, the dedicated experimentation platforms, the rater infrastructure, and the physical asset naming conventions all point to an organization that has systematized nearly every aspect of its operations \u2014 including how it manages the systems that manage the systems.<\/p>\n

For anyone tracking the evolution of search technology, the significance of this analysis is clear. Google\u2019s internal URL patterns are not just administrative artifacts; they are a real-time signal of organizational priorities, project status, and infrastructure philosophy. The fact that AI Mode and Gemini were visible through staging hosts before their public launches suggests that similar patterns could reveal future initiatives \u2014 if one knows where to look.<\/p>\n

The broader lesson from the Resoneo analysis is that in an era of increasingly locked-down systems and login-gated interfaces, the metadata itself \u2014 the names, paths, and parameters that companies leave visible \u2014 can be just as revealing as the content behind the walls. For Google, a company built on the principle of organizing the world’s information, the irony is that its own internal organization has proven just as susceptible to discovery.<\/p>\n","protected":false},"excerpt":{"rendered":"

An analysis of over 3.7 million internal Google URLs has pulled back the curtain on the search giant\u2019s ranking architecture, revealing everything from the specific systems that score and re-rank results to the manual blacklists used to suppress controversial content. Researchers at Resoneo took an unconventional approach: rather than attempting to bypass login screens or […]<\/p>\n","protected":false},"author":7,"featured_media":84562,"comment_status":"closed","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"fifu_image_url":"https:\/\/cards.overcentral.com\/cards\/en\/57450.png","fifu_image_alt":"Analysis of 3.7M Google URLs Reveals Ranking Secrets and Blacklists","footnotes":""},"categories":[31],"tags":[],"class_list":["post-57450","post","type-post","status-publish","format-standard","has-post-thumbnail","category-technology"],"fifu_image_url":"https:\/\/cards.overcentral.com\/cards\/en\/57450.png","fifu_image_alt":"Analysis of 3.7M Google URLs Reveals Ranking Secrets and Blacklists","_links":{"self":[{"href":"https:\/\/overcentral.com\/en\/wp-json\/wp\/v2\/posts\/57450","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/overcentral.com\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/overcentral.com\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/overcentral.com\/en\/wp-json\/wp\/v2\/users\/7"}],"replies":[{"embeddable":true,"href":"https:\/\/overcentral.com\/en\/wp-json\/wp\/v2\/comments?post=57450"}],"version-history":[{"count":0,"href":"https:\/\/overcentral.com\/en\/wp-json\/wp\/v2\/posts\/57450\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/overcentral.com\/en\/wp-json\/wp\/v2\/media\/84562"}],"wp:attachment":[{"href":"https:\/\/overcentral.com\/en\/wp-json\/wp\/v2\/media?parent=57450"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/overcentral.com\/en\/wp-json\/wp\/v2\/categories?post=57450"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/overcentral.com\/en\/wp-json\/wp\/v2\/tags?post=57450"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}