Two sites publish blog posts written largely by ChatGPT. One keeps its rankings through a year of Google core updates. The other loses most of its organic traffic in a single weekend. If Google were penalizing AI content for being AI, both would fall together. They do not. That gap is the whole story, and it is the thing most coverage of "the Google AI crackdown" gets wrong.
The popular version says Google has built a detector, decided machine-written text is spam, and is hunting it down. None of that is true. Google has stated its position plainly and repeatedly, and it has never been about how a page was made. What its 2024 to 2026 core updates target is something narrower and more useful to understand: sameness. Content that repeats what already exists, adds nothing, and provides no reason to exist. The industry calls the worst of it AI slop. Understanding why some AI content survives every update and some gets erased is the difference between using these tools safely and getting wiped out by them.
Where "AI slop" came from, and what Google actually said
The phrase predates the panic. "Slop" started appearing online around 2022 as generative image tools spread, took hold through 2023 and 2024, and by December 2025 Merriam-Webster had named it the word of the year, defining it as low-quality digital content produced in quantity by artificial intelligence. The keyword in that definition is not "artificial intelligence." It is "low-quality." Slop is not a synonym for AI-made. It names a specific failure: volume without value.
Google's stated policy maps onto that distinction almost exactly, and it was set long before the slop discourse peaked. In February 2023, Google published guidance about AI-generated content in its Search Central blog. The core line: Google's focus is on the quality of content, not how it is produced. The post drew a deliberate parallel to history. About a decade earlier, Google noted, there were the same anxieties about a flood of mass-produced but human-written content. The reasonable response then was not to ban human writing. It was to improve the systems that reward quality. Google said it would treat AI the same way.
It also drew the line clearly. Using automation, including AI, to generate content whose primary purpose is to manipulate search rankings is a violation of spam policy. That has been true for years and was not invented for AI. The phrasing matters. The trigger is not automation. It is the intent to game rankings with content that does not help anyone. A human typing a thousand thin pages by hand breaks the same rule.
So Google's position, stated and consistent, is this: it does not care whether a person or a model wrote your page. It cares whether the page is worth a reader's time. Every core update since has enforced that line, not a different one.
The helpful content era, and the lesson it taught
To see how the policy plays out in ranking systems, follow the helpful content update. Google launched it in August 2022, with a second pass in December 2022 and a third in September 2023. Its job was to find content created mainly for search engines rather than people and push it down. The September 2023 version sharpened the classifier to better catch unoriginal content and pages built to rank. It also quietly revised the guidance: Google had told creators to write content "by people, for people," and in September 2023 it dropped the "by people" half, leaving only "for people." A small edit with a large signal. The production method was being written out of the standard.
Then, in March 2024, Google did something more significant. It retired the standalone helpful content classifier and folded those signals into the core ranking system itself, as part of the March 2024 core update. There was no longer a single helpful content switch that flipped a site up or down. Instead, multiple signals across the core system now assess whether content is helpful, and they can reinforce one another or pull against one another. Google said the March 2024 update, taken together with other improvements, was aimed at reducing low-quality, unhelpful content in results by around 40 percent.
The lesson for anyone using AI is in that architecture. There is no helpful content penalty to avoid and no AI flag to clear. Helpfulness is now distributed across the ranking system as a property of the page, weighed continuously against everything else. You cannot satisfy it with a checklist. You can only satisfy it by being genuinely more useful than the alternatives, which is exactly what the policy said all along.
Scaled content abuse: the name for the real target
The same March 2024 announcement introduced the spam policy that has done most of the visible damage since. Google calls it scaled content abuse. The definition is worth reading closely: producing many pages primarily to manipulate search rankings rather than help users, generating large amounts of unoriginal content that adds little or no value, no matter how it is created.
That last clause is the one to underline. No matter how it is created. The policy was written to cover content made by automation, by human effort, or by any mix of the two. It is deliberately method-blind. Google built it that way so it could act on the behavior, which is value-free content at scale, without ever needing to prove a machine was involved. This is why "is my content AI" is the wrong question. Google is not asking it. The policy does not contain it.
What the policy is really aimed at is a specific play. Point a model at a keyword list, generate fifty to five hundred near-identical articles, publish them with no editing and no expertise, and hope volume catches search traffic. Every one of those pages is a slight rephrasing of pages that already rank. None contains first-hand experience, original data, or a view a reader could not get elsewhere. That is scaled content abuse, and that is what AI slop looks like in a search index.
Why some AI content ranks and some gets erased
Now the gap from the opening makes sense. Picture the two ChatGPT-using sites side by side.
The site that survives uses AI as a drafting and research aid. A person with real knowledge of the subject sets the angle, the model produces a first pass, and then an expert rewrites it: adds an example only someone in the field would know, corrects what the model got wrong, brings in proprietary data or a genuine argument, and cuts the filler. The finished page carries something that did not exist online before it was published. By the time it ships, the question of who typed the first draft is irrelevant. The page is original and useful. Google's systems were built to reward exactly that.
The site that gets erased uses AI as a printing press. It mass-produces pages that restate the current top results in slightly different words. Nobody with subject knowledge touches them. There is no new information, no first-hand experience, no point of view, and often no real author behind a byline. Across hundreds of pages the structure is identical and the substance is duplicated. Each page exists to occupy a search result, not to answer a person. That is precisely the pattern scaled content abuse describes.
Both sites used the same tool. Only one used it to add value. The dividing line Google's updates draw is not human against machine. It is original-and-useful against same-and-empty. The sameness is the offense. The AI is just the cheapest way the offense has ever been committed.
Google's own quality rater guidelines confirm where the line sits. The January 2025 update told the human raters who evaluate search quality to assign the lowest rating when all or almost all of a page's main content is copied, paraphrased, or auto-generated with little effort, little originality, and little added value. Three conditions, joined by "and." AI-generated alone does not earn the lowest score. AI-generated plus no effort plus no originality plus no value does. There is no AI checkbox in the guidelines. There is a sameness test.
The 2024 to 2026 core updates: the same line, enforced harder
The pattern held and intensified. Google has run core updates steadily, and each one has applied more pressure to thin, derivative content while the machine-blind policy stayed fixed.
The March 2026 core update is the clearest recent case. It rolled out from March 27 to April 8, 2026, and it was unusually severe. According to tracking reported by Search Engine Land, ranking volatility ran well above the December 2025 update: in the top three results, around 79.5 percent of URLs changed position, and roughly 24.1 percent of pages that had been ranking in the top ten dropped out of the top hundred entirely. A separate, very fast spam update ran on March 24 to 25, the shortest confirmed spam update on Google's status dashboard. Coverage through early 2026 reported sites that had grown on mass-produced AI pages losing large shares of organic traffic, with the heaviest hits on niche sites running hundreds of templated pages, AI-rewritten news aggregators, and thin affiliate pages with no first-hand testing.
It is tempting to read that as Google finally switching on an AI penalty. It is not. The sites that fell were not punished for using AI. They were caught by signals that have been in the system since March 2024: high semantic similarity across many pages, no citations to primary sources, no verifiable authors, and content that added nothing. The same updates left AI-assisted content built by genuine experts, with original examples and real editorial work, ranking fine. The technology Google uses to spot sameness keeps improving. The standard it measures against has not changed since 2023.
There is a structural reason this only gets harder to game. Google has long held a patent often discussed under the name information gain, covering a score for how much genuinely new information a page adds beyond what a reader has already seen on the topic. Whether or not that exact mechanism is live, the logic is now baked into how modern ranking and AI answer engines behave. When everyone copies whatever ranks, results homogenize, and a homogenized page gives the system no reason to prefer it. The page that gets ranked, or cited inside an AI answer, is the one that says something the others do not. Sameness is not a neutral state. It is a ranking disadvantage by design.
What this means for how teams should use AI
The practical guidance falls out of the policy once you stop asking the wrong question.
Stop trying to make AI content "undetectable." There is nothing to hide from. Google has not announced an AI detector for ranking, and the third-party detectors that exist are unreliable enough that Google does not lean on them. Effort spent disguising machine origin is effort spent on a problem that does not exist. It also produces nothing a reader values, which is the actual test.
Decide where the value enters before you draft. AI is fine, often genuinely useful, for outlining, summarizing research, drafting, and editing. It cannot supply the one thing that keeps a page safe: information gain. First-hand experience, original data, a real argument, an example from inside the work. If a piece has no source of new value other than the model, it is slop in waiting, no matter how clean the prose. Identify the human contribution first. If there is not one, do not publish.
Treat scale as the actual risk. A few strong AI-assisted pages a week, each genuinely edited by someone who knows the subject, is not what scaled content abuse describes. Hundreds of lightly-touched pages a month, built to blanket keywords, is the exact pattern. The danger lives in volume without proportional human judgment, not in the tool.
Make a real person own the byline and the gate. Pages caught in 2026 routinely shared a tell: no verifiable author and no credentials. A named expert who actually shaped and stands behind the content is both a quality signal and a forcing function. If no qualified person will put their name on it, that is the system telling you the page has no value to add.
Measure sameness directly. Before publishing, ask the only question Google is asking: what does this page contain that a reader could not already get from the results that rank? If the honest answer is "nothing, phrased differently," the format does not matter and the prose quality does not matter. It will not hold, because sameness is the thing the updates erase.
The reframe is the whole point. Google's core updates did not declare war on AI content. They raised, and keep raising, the cost of sameness. Read that way, the path through every future update is not a trick or a workaround. It is the oldest instruction in the policy, now enforced with much sharper tools: publish something genuinely worth a reader's time, and make sure a real person made it so.
Council summary
This post argues that Google's core updates never targeted AI as a production method. They target sameness: derivative content that adds nothing a reader could not already find. The review confirmed every dated claim against Google primary sources and contemporaneous reporting. Verified: the February 2023 AI guidance ("quality of content, not how it is produced"), the helpful content updates of August 2022, December 2022 and September 2023, the March 2024 core update that folded helpful content into core ranking and introduced scaled content abuse with a stated 40 percent reduction in low-quality results, the January 2025 rater guidelines, and the March 2026 core update rollout from March 27 to April 8 with 79.5 percent top-three volatility and 24.1 percent of top-ten pages falling out of the top 100. The takeaway for any team using AI: stop chasing undetectability, decide where genuine information gain enters before drafting, and never publish a page no qualified person will own.
Comments