If you have run into the claim that Google measures how well a page compresses and uses that as a ranking signal, here is the short answer to is compressibility a ranking factor: no, not as something publicly confirmed. Compressibility is a real, documented technique for spotting repetitive spam. It is not confirmed as something Google currently uses to rank pages. The evidence for current Google use is weak, while the evidence that the technique exists and was studied is solid. And the patent people usually point to when they raise this version of Google compression spam detection is not even a Google patent. It belongs to Microsoft.
That mix, a real idea with shaky evidence for the specific claim people are making about it, is worth walking through slowly, because the pieces get mixed together a lot in SEO discussions.
Where the compressibility ranking claim came from
The idea has two real documents behind it, both from 2006, and neither one is from Google.
The first is a patent application called "Using content analysis to detect spam web pages," publication number US20060184500A1. It was filed on 11 February 2005 and published on 17 August 2006. The inventors are Marc Najork, Dennis Fetterly, Mark Manasse and Alexandros Ntoulas, and the filing lists Microsoft Corp. The current assignee on record is Microsoft Technology Licensing LLC. The application was later granted in 2011. You can find it on Google Patents, which is just a hosting and search service for patent records from any country or company. Being listed there does not make something a Google patent, the same way a book showing up in a library catalog does not make the library the author.
So the safe way to describe it is "a Microsoft patent application published in 2006 describes using the compressibility of visible text as a clue to web spam." Calling it "Google's compression patent" is not accurate, and it is one of the most common mistakes in how this idea gets repeated online.
The second document is a research paper from the same four authors, "Detecting spam web pages through content analysis," published on 23 May 2006 in the proceedings of the WWW '06 conference. This paper used data from the MSN Search crawler, not Google's. It is genuinely useful as empirical evidence that compression ratios correlated with spam in that dataset, but it tells you nothing directly about what Google does.
This is also where a lot of compressibility SEO discussion gets its start, because it is one of the few ideas in that space with an actual patent and an actual study behind it, not just a forum theory. The idea resurfaced in 2024 when a large set of Google's internal documentation leaked. Mike King published an analysis of that leak on 27 May 2024, and it mentioned a module called CompressedQualitySignals. That name sounds like confirmation of the old compressibility theory, and a lot of people treated it that way. We will get to why that connection does not hold up later in this piece, but keep this order of events straight: the patent and paper came first, in 2006, and they are a much older public record than anything in the 2024 leak. This was not discovered by the leak. It was rediscovered.
What the patent actually describes
The patent's method centers on a measure it calls zipRatio. The formula is straightforward: take the byte size of a page's visible text before compression, and divide it by the byte size of that same text after compression. The patent's description specifically uses visible text, not the full HTML including markup.
Here is an illustrative example, just to make the math concrete and not a claim about any real page. If a chunk of text is 10,000 bytes before compression and 2,500 bytes after, the ratio is 4.0. That means the compressed version kept only 25% of the original size, a 75% reduction. A ratio of 2.0 means the compressed version is half the original size.
The patent discusses a threshold, described as something like 2.0, where it says the probability of web spam increases sharply as the ratio rises past that point. But it is careful to qualify this: identifying a page as spam based on the ratio is described as pending other evaluations, not a standalone verdict. The patent also uses language like "such as" and "e.g." around that 2.0 figure, which signals it is an illustrative example in the patent text, not a fixed rule.
What that threshold does not establish is a current Google boundary, a universal line between good and bad writing, a rule that every page above 2.0 gets penalized, or any specific ranking weight. It also does not match the 4.0 figure from the research paper, and the two should not be blended together into one supposed rule, because they come from different documents measuring things slightly differently.
It is also worth being precise about what this technique is and is not. It is not asking an AI model to read a page and judge whether it is well written. It is a mechanical compression process, the same kind of thing GZIP does when it shrinks a file for storage or transfer, applied to a page's text to see how much repetition it contains.
What the research supports, and what it does not
The 2006 research paper is the clearest empirical evidence in this whole discussion, and it is worth laying out exactly what it found rather than the rounded-off version that circulates.
The authors worked from a crawl of 105,484,446 pages gathered by MSN Search. Out of that, they manually reviewed a sample of 17,168 English-language pages. Of those, 2,364 pages, or 13.8%, were labeled spam, and 14,804 pages, or 86.2%, were labeled non-spam. One detail that often gets left out: the crawler already applied some spam-filtering heuristics before this sample was drawn, so it is not a clean random sample of the whole web.
The central finding is this: 70% of all sampled pages with a compression ratio of at least 4.0 were judged to be spam. That is a real, specific, dated statistic, and it is the single most useful number in this whole topic. But it needs its scope attached every time you use it. It describes the spam share within a narrow high-ratio subset of this one historical sample. It is not an overall detection rate, it is not recall, and it is not a probability that a modern page with that ratio would be penalized by Google today.
The paper also shows why you should not stretch this number further than it goes. In the higher ranges, the authors report an average spam probability of 72% for ratios of 4.2 and above, but only about 1.5% of pages in the sample actually fell into that range. A strong-looking association in a tiny slice of pages is a real finding, but it covers very little of the web. The paper also found that 4.8% of pages actually grew in size when compressed. These were mostly very short pages, where the overhead of the compression format itself outweighs any savings. That is a reminder that a compression ratio reflects the mechanics of the measurement as much as it reflects the content.
None of this amounts to a controlled experiment on Google's rankings. The paper does not show that cutting repetition causes a ranking improvement, does not show that Google adopted this exact method, and does not give you a present-day penalty threshold or a benchmark you could apply to AI-generated writing, a use case the paper predates entirely.
Why CompressedQualitySignals is not proof
This is the part where the 2024 leak gets misread most often, so it is worth slowing down here.
The leaked documentation includes a module named CompressedQualitySignals. Shaun Anderson's write-up on this, published 5 October 2025, quotes its description as "a message containing per doc signals that are compressed and included in Mustang and TeraGoogle." Read that description closely. It is about data being compressed for storage inside Google's own systems. It is not, by itself, a statement that a page's compression ratio is one of the signals being measured.
There is a real difference between two things that sound similar. One is taking a bundle of quality-related data Google already has about a page, and compressing that bundle to save space. The other is taking a page's own content and measuring how compressible that content is, then using the result as a spam signal. The existence of the first does not prove the second. Compressing a spreadsheet full of scores is a different act than scoring an article by how well it compresses.
Anderson's piece does draw a conceptual line between the module name and the older compression-based spam research, and that connection is reasonable commentary. But it is commentary, not confirmation. The material available does not name a specific field that measures page compressibility inside that module, does not give a formula for it, and does not give a present-day threshold or ranking weight.
It is also worth noting what the leak discussion says more broadly. Mike King's own analysis is clear that the documents he reviewed do not expose Google's actual scoring functions. A separate assessment of the leak, published by Juliet John on 10 June 2024, makes the same point: the documentation does not show how these attributes get used in ranking, or confirm that every listed element is currently active. Google has also warned publicly against drawing firm conclusions from out-of-context or incomplete internal documentation. That is a general caution about the whole leak, not a denial aimed specifically at compressibility.
The verdict for content teams
Here is the honest summary. A compression-based spam-detection method exists and is documented in a real patent. Compression ratios correlated with spam in a specific 2006 research sample. Neither of those facts establishes that Google currently uses this method in production, that it applies a specific threshold to change rankings, or that the leaked CompressedQualitySignals module proves anything about page-level compressibility. And publishers should not try to make their writing harder to compress. That is not a real target, and chasing it would make your content worse for the actual reader in front of you.
If you run a content team at any kind of volume, the practical move here is not a new technical project. Do not add a compression-ratio target to your editorial brief. There is no current Google number to hit, so there is nothing to optimize against. What is worth doing is the plain editorial judgment this points back to: cut repetition when it adds nothing for the reader. That is good writing, not a defense against an unconfirmed algorithm. Ask yourself, when two sections of a piece look similar, whether they are actually saying something different or just restating the same point in new words. Do not go the other direction either. Adding filler, unusual phrasing, or unnecessary complexity just to lower a compression score is trying to game a proxy you cannot even confirm is being measured, and it tends to make an article worse to read.
Two things to leave alone entirely. Do not touch your server's delivery compression (GZIP or Brotli) over this. That is about how efficiently your pages get transferred to a browser, a completely separate question from whether a page's text content is repetitive, and nothing here suggests disabling it helps you. And do not treat this as a method for spotting AI-written content. The patent and the 2006 study were about spam detection in general, built years before anyone was writing with AI tools, and nothing in them supports using compressibility as a sign of machine authorship.
If you want one sentence to carry away, it is this one. On the question of Google compression spam detection as an active ranking mechanism, compressibility is a documented spam-detection technique, but current Google use of it is unconfirmed. Treat compressibility SEO as a useful lens for thinking about redundant content, not as a term with an official Google definition. The question in the title deserves a real answer, and a real answer includes saying plainly what is not known yet.
What would change this verdict
The grade on this would move if something specific showed up that has not shown up yet. A direct statement from Google naming content compressibility as an active search-quality input would do it. So would production-level documentation that clearly defines a page-compressibility field and explains how it gets used. A credible, reproducible experiment that isolates compressibility while holding other changes constant would help too, though even a correlation in rankings would not fully prove the mechanism on its own, since changing how repetitive a page is tends to change its readability and substance at the same time.
It is just as possible that a specific, credible disclosure could show this approach has been retired. Nothing found in researching this piece establishes that either. The honest position, for now, is that the mechanism is understandable and the question of its current use at Google is open.
The next time someone raises the question "is compressibility a ranking factor" and treats "Google's compression spam patent" as settled fact, you now know enough to ask which document they mean, and whether it says what they claim.
