key takeaways
Your reader has already read three articles on this before they reach yours.
You are not competing to be comprehensive. You are competing to be worth reading fourth, and that is what Google's own patent measures (information gain)
If you nail it you'll crack the top 3, then secure the 1st position in Google search results, AI overviews, and in AI mode.
Google now names the four attributes it judges quality on, and it updated that documentation on October 5, 2026.
The definition is finally specific enough to audit against.
Most pages fail on the same attribute, because most SEO content was built for a search engine alone. Now it has to show real human effort.
Google means four things by content quality, and it lists them by name in its documentation on creating helpful, reliable, people-first content.
EFFORT: "The extent to which human work went into creating the content or the systems powering it."
ORIGINALITY: "The extent to which the content offers unique, original information or perspectives that aren't already available on other websites."
TALENT OR SKILL: "Whether the content shows the talent, skill, or expertise necessary to provide a satisfying experience for visitors."
ACCURACY: For informational pages, content should be factually accurate. For topics affecting health, finance or safety, it must be highly accurate and consistent with expert consensus.
Duy Nguyen, a Search Quality Analyst at Google, presented these same four pillars at Search Central Live Deep Dive in Barcelona, with a one line instruction attached: the Search Quality Rater Guidelines, where E-E-A-T is defined, are public, so read them.
That page carried a last updated date of October 5, 2026 when we pulled it.
Google's separate guidance on generative AI content was updated on October 1.
Two documentation changes in five days is rare, both about what counts as quality work, which reads as deliberate rather than coincidental.
Effort counts as the human work that went into the page, not the length of it.
Google's wording covers the content or the systems powering it.
A tool you built, a dataset you compiled or a process you ran all qualify.
This is the attribute the industry keeps mistranslating into volume.
A 5,000 word guide assembled from the top ten results has almost no effort in it by this definition.
A 700 word page reporting something you measured yourself has a great deal of elbow grease.
The 2024 Google API documentation leak included an attribute named contentEffort, described as an LLM-based effort estimation for article pages.
Treat that as a leaked attribute name rather than a confirmed ranking factor.
Google has never acknowledged it.
The shape of it still matters: an effort estimate made by a language model judges how hard a page would be to reproduce, not how long it is.
The honest test is replication.
A simple sense check we use for content is “Could someone with AI replicate this page in one afternoon?”
That question leads straight into information gain.
The patent everyone cites is US11354342B2, "Contextual estimation of link information gain", assigned to Google LLC, filed October 18, 2018 and granted June 7, 2022.
The usual summary in SEO writing is that it scores how much new information a page adds to the index. That is close, and the advice it produces is sound.
The patent baseline is narrower, and the difference is where it gets interesting.
Read the patent itself, it scores a document's information gain relative to documents the user has already viewed.
The context is a session, a sequence, a reader who has already read three things about your topic before arriving at yours.
Google filed this in October 2018, and claims one describes an assistant picking a single document to read aloud. Seven years later that is Google’s AI Mode.
That is the mechanism behind the fourth-reader problem at the top of this page. You are competing against what that reader was already shown.
It also explains why the instinct to write 3,000 words covering everything keeps failing. Covering everything is what the documents before you already did, that information is already out there and the consensus is decided.
Whether a machine wrote your page is a separate question, and Google has now answered it directly.
AI-generated content is not automatically low quality, and Google says so in its own documentation.
Quality is assessed on the content, not the authorship, which is why your AI content watermark isn't what Google is looking for.
Low quality work comes from humans and machines alike, so the AI content versus human written content question is less relevant today.
What Google does flag is the failure mode.
Its guidance on generative AI content states that generative models "don't retrieve facts, but predict a likely sequence of words based on their training data.
Because of this, generative AI outputs may contain inaccuracies (also known as hallucinations)."
The instruction attached to it is unusually direct: "It is critical to manually factcheck and review all AI-generated content for accuracy and trustworthiness before publishing."
The October 2026 update went further than the article body, the fact-check instruction now covers title tags, meta descriptions, structured data and image alt text as well.
Most teams generating metadata at scale have not read that change.
Read that against the four attributes and it lands on accuracy, the fourth pillar.
Google is saying the accuracy attribute does not get a pass because a model wrote it.
Volume is where this gets expensive, and Google has been blunt about that too.
It does not matter how scaled content gets made, and Google has said so in plain words. Danny Sullivan, Google Search Liaison, put it this way: "we don't really care how you're doing this scaled content, whether it's AI, automation, or human beings."
His verdict on it was four words: "It's going to be an issue."
He also pointed out that none of this is new, noting that people were generating thousands of pages overnight back in 2005.
The tactic is old, you could do the same with human freelancers, now the costs are lower because of AI workflows.
Publishing more pages was once limited by how fast people could write, and that limit is gone, so the behavior it used to contain now happens at a scale the old systems were never built for.
The practical read is not "publish less". It is that page count stopped being evidence of anything.
A hundred pages that could each be reproduced in an afternoon are a hundred pages of commodity, and the industry has a name for what that competes against.
Content is commodity when anyone, or any model, could have produced it from publicly available information. It is non-commodity when it requires you to have done something, seen something, or measured something.
Google used real examples for this at Search Central Live.
For a running store, the commodity version is a listicle on choosing running shoes.
The non-commodity version is a breakdown of why one customer's shoes collapsed after 400 miles, and what their gait did to the foam.
For a real estate agent, the commodity version is tips for first-time homebuyers.
The non-commodity version is the inspection they waived and what they found when they crawled the sewer line themselves.
Simple test for your content, could this have come from anyone, or could it only have come from the person who experienced this?
You prove effort by publishing the things only you could publish: what you measured, what you tested, what happened when you did it, and what it cost.
Across our own campaigns, the pages that recover are consistently the ones where we added something nobody else had:
Two of those three added no new page at all, pruning and deepening both raise the average effort per page by working on what is already there.
Change the pages you already have before you write new ones:
None of that requires a new publishing schedule.
It requires effort, which is now the thing being measured by name, and the standard our SEO content writing is built on.
Effort, originality, talent or skill, and accuracy. Google lists all four in its documentation on creating helpful, reliable, people-first content, and a Search Quality Analyst presented the same four at Search Central Live in 2026. They sit alongside Google's E-E-A-T, which the Search Quality Rater Guidelines define separately.
Google assesses the content rather than the author. Its guidance on using generative AI content states that generative models predict likely word sequences rather than retrieving facts, so outputs may contain inaccuracies, and it instructs publishers to manually fact-check AI-generated content before publishing. We covered where the line falls in AI content versus human written content for SEO.
Length is not what Google describes as effort. Its definition covers the human work that went into the content or the systems powering it. A short page reporting original measurement carries more effort than a long page assembled from existing sources. Our long-form content guide works through what to write instead of writing more.
Information gain measures what a document adds beyond what a reader has already seen. The Google patent most often cited, US11354342B2, scores a document's information gain relative to documents the user has already viewed, which is narrower than the common description of adding value to the index.
Ask whether it could have been produced by anyone, or any model, from publicly available information. If the answer is yes, it is a commodity. Non-commodity content requires first-hand experience, original measurement, or access that others do not have. The same test decides whether a page is worth optimizing for AI search.