Disclaimer: This is not official. Any article (like this) dealing with the Google Content Data Warehouse leak requires a lot of logical inference when putting together the framework for SEOs, as I have done with this article. I urge you to double-check my work and use critical thinking when applying anything for the leaks to your site. My aim with these articles is essentially to confirm that Google does, as it claims, try to identify trusted sites to rank in its index. The aim is to irrefutably confirm white hat SEO has purpose in 2025 – and that purpose is to build high-quality websites. Feedback and corrections welcome.
For over two decades, our work in Search Engine Optimisation (SEO) has been a process of reverse-engineering a black box.
Strategies were built on a foundation of correlation, empirical observation, and the careful interpretation of public guidance.
The March 2024 Core Update, arriving concurrently with an unprecedented leak of Google’s internal Content Warehouse API documentation, represents a fundamental paradigm shift.
This leak, corroborated by sworn testimony from the U.S. Department of Justice (DOJ) v. Google antitrust trial, provides the SEO industry with its first look at the architectural blueprints of Google’s ranking systems, moving the practice from an art of inference to a science of architectural alignment.
My article presents a forensic analysis of one of the most intriguing components revealed in this leak: a protobuf named QualityCopiaFireflySiteSignal.
The central premise is that this specific technical attribute serves as a key enforcement mechanism for Google’s recently evolved “scaled content abuse” policy.
This policy, which rebranded the older “spammy automatically generated content” guidelines, shifted the focus from the method of content creation to the intent and outcome of publishing content at scale.
The leaked documentation primarily consists of property definitions for protocol buffers, or “protobufs”.These are not the scoring functions or algorithms themselves; rather, they are the structured data containers-the very schematics for the information-that Google’s various ranking and demotion systems access and process. Understanding these data structures is akin to an architect studying a building’s foundations; it reveals the principles upon which the entire edifice is constructed.
My investigation builds upon previous analyses published on the Hobo SEO blog in 2025, which have deconstructed other critical components of Google’s quality assessment framework, such as the QualityNsrPQData model and the contentEffort attribute.
This article is the next logical step, connecting the high-level policy against scaled abuse to the specific, site-level data structure seemingly designed to detect it. We will deconstruct the evidence, trace the evolution of Google’s philosophy, and provide strategic imperatives for thriving in this new era of architectural transparency.
A Policy’s Evolution: Why ‘Scale’ Became the Target, Not the Tool
Google’s fight against low-quality, manipulative content is as old as the search engine itself, but its policies have evolved significantly to keep pace with the changing tactics of spammers.
The direct predecessor to the current policy was known as “spammy automatically generated content“. As defined in early 2024, this policy targeted:
“Content that’s been generated programmatically without producing anything original or adding sufficient value; instead it’s been generated for the purpose of manipulating search rankings and not helping users.”
The key term here was “programmatically.”
The focus was on the method of creation. This was effective in an era when automated content was often easily identifiable as machine-generated gibberish or poorly “spun” text. However, the rise of sophisticated generative AI rendered this distinction increasingly obsolete.
Modern AI can produce content that is grammatically correct, coherent, and often indistinguishable from low-effort human writing, creating a grey area that spammers were quick to exploit.
Recognising this, Google updated its spam policies in March 2024, rebranding the section to “scaled content abuse”. The new, method-agnostic definition is far broader:
“When many pages are generated for the primary purpose of manipulating search rankings and not helping users. This abusive practice is typically focused on creating large amounts of unoriginal content that provides little to no value to users, no matter how it’s created.”
This was a strategic and necessary evolution. It future-proofed the policy against any new technology for content generation by shifting the focus to two timeless indicators of spam: the unhelpful outcome (large volumes of unoriginal content) and the manipulative intent (to game search rankings).
Google’s Search Liaison, Danny Sullivan, has been unequivocal about this philosophical shift.
His commentary reveals an awareness that the SEO community was misinterpreting Google’s stance on AI, believing that any content which appeared to be high quality was acceptable. Sullivan clarified the reality:
“…we don’t really care how you’re doing this scaled content, whether it’s AI, automation, or human beings. It’s going to be an issue.”
He further cautioned against the flawed definition of “quality” that some were adopting, noting that AI is proficient at creating “really nice generic things that read very well” but which do not necessarily provide unique value or originality.
This directly addresses the problem of AI being used to flood the web with plausible-sounding but ultimately unhelpful content.
This modern policy is not a new invention but the culmination of a long-standing battle
It echoes the work of Matt Cutts, former head of Google’s webspam team, who for years fought against scaled, low-value content in forms like article directories and manipulative guest blogging networks. Cutts consistently warned against any tactic that produced a “ton of useless content” purely for the sake of acquiring links or rankings.
The core principle-penalising low-effort content created for machines rather than people-has remained constant.
The “scaled content abuse” policy is simply the latest and most robust articulation of that principle, supported by John Mueller’s consistent advice that quality is a holistic, site-wide consideration, not just a page-level attribute.
Deconstructing the Name: QualityCopiaFireflySiteSignal
The name of the protobuf itself – QualityCopiaFireflySiteSignal – is not an arbitrary string of code. Within Google’s engineering culture, naming conventions are often highly descriptive.
A forensic, word-by-word analysis of this name provides a powerful indication of its function.
- Quality: This is the overarching context. The signal is part of the vast ecosystem of quality assessment systems within Google. It directly connects to the public-facing goal of surfacing high-quality content and the numerous pageQuality attributes found throughout the leaked documentation.
- Copia: This is arguably the most direct piece of evidence. Copia is Latin for ‘abundance’, ‘plenty’, or ‘profusion’. In the context of a system designed to enforce a policy against “scaled” abuse, this term is a perfect fit. It is the architectural label for the problem of excessive volume that the policy explicitly targets.
- Firefly: This is the most evocative component. While a definitive explanation is unavailable, a plausible hypothesis is that it refers not to Adobe’s AI tool, but to the Firefly Algorithm. This is a nature-inspired, metaheuristic method used to solve complex optimization problems by modeling the flashing behavior of fireflies. Such an algorithm would be well-suited for a system designed to find faint signals of manipulation (the “brightest” fireflies) within the vast and noisy dataset of the web index.
- SiteSignal: This final component is critical. It indicates that the assessment is aggregated and applied at a site-wide or domain level, not just on a per-page basis. This aligns perfectly with the policy’s focus on “many pages” and corroborates the long-standing advice from representatives like John Mueller that Google evaluates the overall quality of a website. It also fits within the broader architecture revealed by the leak, which includes numerous other site-level metrics like siteAuthority, siteFocusScore, and hostNSR. This suggests that Google is looking for systemic, domain-wide patterns of scaled abuse, rather than just penalising individual low-quality pages.
The name itself, therefore, tells a story.
It describes a system that assesses site-wide quality (Quality, SiteSignal) by looking for patterns of excessive volume (Copia) using a sophisticated (potential) heuristic algorithm (Firefly) to identify abuse. And perhaps, to identify quality too.
The purpose of this article is to focus on the parts of it that could be used to identify scaled content abuse.
Technical Analysis: The Attributes of QualityCopiaFireflySiteSignal
The leaked documentation provides a succinct, powerful summary of the module’s purpose: “fireflySiteSignal – Contains Site signal information for Firefly ranking change.“ This single line confirms its role in altering rankings.
The protobuf definition then provides the exact data points that constitute this signal. This is the raw input. By analysing each attribute, we can understand precisely how Google quantifies a site’s behaviour to detect scaled abuse.
| Attribute | Description |
| dailyClicks | A count of the total number of clicks the website receives from search results on an average day. |
| dailyGoodClicks | A subset of dailyClicks, this counts clicks that Google considers “good,” suggesting the user found the page useful. |
| dataTimeSec | A timestamp (in seconds) indicating when this specific set of data was generated. |
| firstBoostedTimeSec | A timestamp marking the first time the site received a ranking boost. |
| impressionsInBoostedPeriod | A count of impressions the site received during a specific period when its ranking might have been temporarily boosted. |
| latestBylineDateSec | The most recent publication date that Google has extracted from an article’s byline on the site. |
| latestFirstseenSec | A timestamp for the last time Google’s crawler first discovered a new page on this site. |
| numOfArticles8 | The number of pages identified as high-quality articles, based on an internal scoring system (a score of 0.8 or higher). |
| numOfArticlesByPeriods | A list tracking the number of new, high-quality articles found in successive 30-day periods. |
| numOfGamblingPages | A specific counter for the number of pages on the site that are identified as being related to (at least) gambling. |
| numOfUrls | The total number of unique URLs (pages) from this site that Google has discovered. |
| numOfUrlsByPeriods | A list showing the number of new URLs discovered in successive 30-day periods, tracking the site’s growth velocity. |
| recentImpForQuotaSystem | A measure of recent impressions, specifically used to manage internal Google system resources (quotas). |
| siteFp | A unique “site fingerprint” (hash value) that serves as a consistent ID for the site for experiments and internal analysis. |
| totalImpressions | The total number of times any page from the site has been shown to a user in the search results. |
User Engagement & Performance Metrics
These attributes measure how users interact with the site in Google’s search results, providing a ground-truth signal of whether the scaled content is actually helpful.
- dailyClicks and dailyGoodClicks: These are perhaps the most crucial engagement signals. dailyClicks is a raw count of clicks from search, while dailyGoodClicks is a subset that Google deems successful-meaning the user didn’t immediately return to the search results. This is a direct input from the NavBoost system. For a site publishing at scale, the ratio between these two numbers is paramount. A site might generate thousands of pages and get a high volume of dailyClicks through keyword targeting, but if the content is unhelpful, the dailyGoodClicks count will be disproportionately low. A poor ratio is a powerful mathematical signal of user dissatisfaction at scale.
- totalImpressions: This tracks how often the site’s pages are shown in search results. A massive totalImpressions number combined with a low click-through rate and a poor dailyGoodClicks ratio would indicate that while the site is targeting many queries, it is failing to satisfy user intent.
- impressionsInBoostedPeriod and firstBoostedTimeSec: These attributes track when a site has received a temporary ranking boost (e.g., for a news event). A site that repeatedly tries to exploit temporary boosts by publishing large volumes of low-effort content around trending topics could be flagged by these metrics.
- recentImpForQuotaSystem: This is a measure of recent impressions used to manage Google’s internal resources for crawling and processing. A sudden, massive spike in impressions from a site publishing thousands of new pages could trigger resource quotas, flagging the site for review as a potential spam source.
Content & Indexing Metrics
These attributes provide a quantitative measure of the scale and quality of a site’s content production, directly addressing the “Copia” (abundance) aspect of the signal.
- numOfUrls and numOfUrlsByPeriods: This is the most direct measure of scale. numOfUrls is the total number of pages Google has discovered. More importantly, numOfUrlsByPeriods tracks the velocity of new page creation in successive 30-day periods. A site that suddenly goes from creating 10 new pages a month to 10,000 would exhibit a dramatic spike in this metric, a classic footprint of scaled content abuse.
- numOfArticles8 and numOfArticlesByPeriods: These are the critical counter-metrics to the raw URL count. numOfArticles8 counts pages identified as high-quality articles (based on an internal score of 0.8 or higher). This score is likely derived from other quality systems, such as the contentEffort attribute in the QualityNsrPQData model, which uses an LLM to estimate the effort put into a page. A site can publish a huge number of URLs, but if the numOfArticles8 count remains low, it’s a clear signal that the scaled content is of poor quality. The numOfArticlesByPeriods metric tracks the velocity of high-quality article creation, allowing the system to distinguish between a site undergoing a genuine, high-effort content expansion and one engaged in scaled abuse.
- numOfGamblingPages: This is a specific risk-factor attribute. The presence of a high number of gambling-related pages can be a signal for review, especially if it’s on a site whose primary topic is unrelated, which would also be a flag for site reputation abuse. It might be that this is for gambling sites and other high-risk sites, I’m not sure, but it’s noticeable it is only gambling sites… they must be identifying other types of sites too.
Timestamps & Identification
These attributes provide temporal context and a unique identifier, allowing the system to track a site’s behaviour over time.
- dataTimeSec: A timestamp indicating when the data set was generated, allowing for historical analysis of a site’s behaviour.
- latestFirstseenSec and latestBylineDateSec: These are freshness signals. latestFirstseenSec tracks when Google’s crawler last discovered a new page, while latestBylineDateSec is the most recent publication date extracted from an article. A large discrepancy between these could indicate a site is trying to appear fresh by manipulating byline dates without adding genuinely new content.
- siteFp: A unique “site fingerprint” or hash value. This is a crucial identifier that allows Google to track a site consistently across different systems and experiments, ensuring that a site cannot easily escape a negative reputation by simply changing its domain name.
The Public Stance vs. The Leaked Reality: Google’s Statements on Clicks
The confirmation that Google extensively uses click data, as evidenced by attributes like dailyGoodClicks and the underlying NavBoost system, stands in stark contrast to years of public statements from its representatives who have consistently downplayed or denied the use of user engagement signals as a direct ranking factor.
John Mueller, a prominent Search Advocate at Google, has repeatedly dismissed the idea. In one statement, he argued against the viability of using click-through rates (CTR) for ranking:
“If CTR were what drove search rankings, the results would be all click-bait. I don’t see that happening.”
In another hangout, he went further, suggesting Google doesn’t even have visibility into on-site user actions, which would preclude their use as a ranking signal:
“So in general, I don’t think we even see what people are doing on your web site. If they are filling out forms or not, if they are converting and actually buying something… So if we can’t see that, then that is something we cannot take into account. So from my point of view, that is not something I’d really treat as a ranking factor.”
Gary Illyes, another analyst on the Google Search team, has echoed this sentiment, often describing click data as unreliable for direct ranking purposes. He has referred to clicks as a “very noisy signal” and stated that using them directly would be problematic due to manipulation and scraping activities. In a particularly blunt dismissal, Illyes was quoted as saying:
“Dwell time, CTR, whatever Fishkin’s new theory is, those are generally made up crap. Search is much more simple than people think.”
These public denials created a long-standing debate within the SEO community. The leaked documentation and DOJ trial testimony have now provided concrete evidence that resolves this debate, confirming that while Google may not use raw CTR as a simplistic, direct input, it absolutely uses sophisticated, aggregated, and normalised click data via systems like NavBoost to evaluate and re-rank search results.
A fascinating footnote to this history of denial lies in Illyes’ choice of words. His dismissal of click-based theories as “made up crap” takes on a layer of profound irony when viewed through the lens of the leak. The documentation reveals a ranking system module explicitly named “Craps,” which is defined as the system that processes “click and impression signals.”
It is, in essence, the very system that handles the data Illyes was publicly dismissing. The metrics it processes-goodClicks, badClicks, and lastLongestClicks-are direct, quantifiable measures of user satisfaction that serve as sophisticated proxies for the very concepts of CTR and dwell time that were being derided.
Whether this was a deliberate, meta-textual joke on Illyes’ part-a hidden admission veiled in dismissive language-is impossible to know. Look to the sentence structure itself (“…made up crap. Search is much more simple…”) as a potential, albeit highly speculative, nod to the truth.
Regardless of intent, the coincidence is striking.
It serves as a perfect encapsulation of the dynamic between Google’s public relations and its internal engineering reality: the very “crap” being publicly derided was, in fact, a named component of the ranking architecture.
A System of Systems: How Firefly Connects to the Quality Ecosystem
The QualityCopiaFireflySiteSignal should not be viewed as a standalone, monolithic algorithm. The architecture revealed by the leak makes it clear that Google’s quality assessment is a sophisticated, multi-stage pipeline composed of many interconnected systems.
Firefly’s role is likely that of a high-level aggregator or a decision-making system that acts upon a confluence of signals fed to it from other, more specialised modules.
A spam action is a significant event, and Google’s engineering relies on cross-verification. The Firefly system likely synthesises inputs to make a final determination. For example:
- Initial Flag: The numOfUrlsByPeriods attribute shows a massive spike in new pages. This is the “Copia” signal.
- Quality Check: The system checks numOfArticlesByPeriods. It sees that despite the huge number of new URLs, the number of high-quality articles is flat. This suggests the new content is low-effort. This is corroborated by a low average contentEffort score from the QualityNsrPQData system.
- User Validation: The system then looks at the engagement metrics. It sees a high dailyClicks count but a very low dailyGoodClicks count. This is the user-behavioural confirmation from NavBoost that the content, despite attracting clicks, is not satisfying users.
- Verdict: With corroborating signals from content velocity (Copia), content quality (QualityNsrPQData), and user dissatisfaction (NavBoost), the Firefly system can conclude with high confidence that the site is engaged in scaled content abuse and apply a site-wide demotion.
Mapping Policies to Signals: The Law and The Evidence
The module and the policy document are not just related; they are two sides of the same coin. They represent the “what” and the “how” of Google’s search quality enforcement. Think of it like this:
- Spam Policies Document (The Law): This document is the rulebook. It’s a public declaration of “what” Google considers to be manipulative or low-quality behaviour. It defines the violations (e.g., Scaled Content Abuse, Thin Affiliation).
- QualityCopiaFireflySiteSignal (The Evidence): This module is part of the data-gathering and enforcement mechanism. It’s a collection of quantitative signals that act as “how” Google’s automated systems can detect the violations described in the rulebook. It’s the evidence used to determine if a site is breaking the rules.
The signals within the QualityCopiaFireflySiteSignal module can be used directly to detect or flag potential violations of the spam policies. Here’s a breakdown of how specific policies could be monitored using these signals:
| Spam Policy Violation | How the QualityCopiaFireflySiteSignal Module Could Detect It |
| Scaled Content Abuse | This policy is about generating many low-value pages. The module is perfectly designed to spot this. Signal: A massive increase in numOfUrlsByPeriods without a corresponding increase in numOfArticlesByPeriods. Indication: This creates a poor ratio of quality content to total content, a strong sign of automated, low-value page generation. |
| Thin Affiliation | This content provides little original value, leading to a poor user experience. User engagement signals would reveal this. Signal: A large gap between dailyClicks and dailyGoodClicks. The site might get clicks, but users immediately bounce back to Google because the content is unhelpful. Indication: High click count but low “good click” count is a classic sign of user dissatisfaction. |
| Site Reputation Abuse | A good site starts hosting low-quality, third-party content to abuse its reputation. Signal: A previously stable site shows a sudden spike in numOfUrlsByPeriods and a gradual decline in its dailyGoodClicks ratio as user trust erodes. Indication: The module’s time-series data can detect this negative change in a site’s quality profile over time. |
| Hacked Content / User-Generated Spam | Often, a hack or spam attack involves injecting thousands of new, spammy pages onto a legitimate site. Signal: A sudden, anomalous explosion in numOfUrls that is completely out of character for the site’s history (tracked in numOfUrlsByPeriods). Indication: This is a huge red flag that something is wrong, prompting further automated analysis or human review. |
| Doorway Abuse | These pages are frustrating intermediaries that don’t satisfy the user’s need. Signal: Just like Thin Affiliation, this would result in a poor dailyGoodClicks to dailyClicks ratio. Users click, realize it’s a useless doorway page, and leave. Indication: The site is getting traffic but failing to satisfy users. |
Strategic Imperatives for the Post-Leak Era
The revelation of the specific attributes within QualityCopiaFireflySiteSignal moves SEO strategy beyond generic platitudes. “Create great content” is no longer sufficient advice. The goal now is to build websites whose fundamental architecture of value aligns with Google’s own now-visible blueprint. This new paradigm can be described as “Architectural SEO”-structuring a site’s content, authority, and user experience in a way that is legible and favourable to Google’s core data-gathering architecture.
- Focus on the GoodClicks Ratio, Not Just Clicks: The strategy must extend beyond securing the initial click. The entire user journey must be optimised to generate positive user behaviour signals. This involves creating a clear information architecture, ensuring fast page load times, and, most importantly, crafting content that comprehensively and efficiently resolves the user’s query. The goal is to prevent the “pogo-sticking” that generates badClicks and serpDemotion signals, instead cultivating the long clicks and successful outcomes that validate your content’s quality.
- Prioritise Increasing Your numOfArticles8 Count: The primary strategic goal must be to increase the proportion of your site’s content that Google would classify as high-effort. This means shifting focus from content volume to content irreplaceability. Invest in content that contains original research, unique data, expert insights, and a perspective that cannot be easily replicated by an AI tool scraping the top ten search results. Every piece of content should be an asset that is difficult and expensive for a competitor to reproduce, thereby increasing its likely contentEffort score.
- Manage Your Publication Velocity: The numOfUrlsByPeriods attribute makes it clear that sudden, unnatural spikes in content production are a red flag. Content strategy should be sustainable and consistent. If you plan a major content expansion, ensure it is matched by a corresponding increase in high-quality (numOfArticles8) pages to avoid triggering abuse signals.
- Audit and Prune Low-Quality Content: The existence of these site-level signals confirms what John Mueller has advised for years: the overall quality of a website matters. A large volume of low-quality, unhelpful pages can drag down the entire domain. Conduct rigorous audits to identify pages that likely have low contentEffort scores and poor engagement signals. Improve them, consolidate them into more comprehensive resources, or remove them entirely.
Conclusion
Ultimately, the QualityCopiaFireflySiteSignal is the technical manifestation of a philosophy Google has held for over a decade.
The leak did not change the rules; it simply revealed the scorecard.
For SEO professionals, the path to durable success is not about finding loopholes or chasing algorithmic fads
It is about building websites whose fundamental value proposition is so clear and robust that it aligns perfectly with the architectural principles of a search engine that is, and always has been, trying to identify and reward true quality.
The era of the black box is over; the era of architectural alignment has begun.
Disclosure: I use generative AI when specifically writing about my own experiences, ideas, stories, concepts, tools, tool documentation or research. My tool of choice for this process is Google Gemini Pro 2.5 Deep Research. I have over 20 years writing about accessible website development and SEO (search engine optimisation). This assistance helps ensure our customers have clarity on everything we are involved with and what we stand for. It also ensures that when customers use Google Search to ask a question about Hobo Web software, the answer is always available to them, and it is as accurate and up-to-date as possible. All content was conceived, edited and verified as correct by me (and is under constant development). See my AI policy.
References
- (https://sparktoro.com/blog/an-anonymous-source-shared-thousands-of-leaked-google-search-api-documents-with-me-everyone-in-seo-should-see-them/)
- (https://ipullrank.com/google-algo-leak)
- (https://developers.google.com/search/docs/essentials/spam-policies)
- (https://www.searchenginejournal.com/google-on-scaled-content-its-going-to-be-an-issue/543308/)
- (https://www.searchenginejournal.com/in-depth-look-at-google-spam-policies-updates/511005/)
- (https://en.wikipedia.org/wiki/Firefly_algorithm)
- (https://www.hobo-web.co.uk/qualitynsrpqdata/)