Updated August 2026 · every claim traced to its source
Does Google detect AI content, and does AI content rank in Google?
Google has never shipped an AI content detector, and no Google documentation describes AI detection as a ranking signal. The policy that actually catches AI content says it applies "no matter how it's created".
Below: what Google has really published, where the popular claims come from, and why AI drafts still fail for a reason nobody is selling you a fix for.
2 free runs a day, up to 200 words each. We save your run so you can get back to it, and a delete button appears with the result. See privacy.
This is our own AI-pattern score, measured here on sentence rhythm, template phrases, vocabulary variety and passive voice. It is not a GPTZero, Turnitin, Originality.ai, Copyleaks or ZeroGPT result, and it does not predict one. Worth knowing: we also ask the rewrite to vary sentence length, drop template phrases and prefer the active voice, so some of the drop is built in. Read the two panels below, not just the number.
Before ·
After ·
Your own text rewritten · Saved to your history, delete any time
The short answer
Google has no public AI content detector, and no Google documentation describes AI detection as a ranking signal. Its published guidance says appropriate use of AI or automation is not against its guidelines, and that it focuses on the quality of content rather than how the content was produced. The policy that does catch a lot of AI content is scaled content abuse, which Google defines as many pages generated for the primary purpose of manipulating search rankings, adding that this applies "no matter how it's created". So the honest answer has two parts: Google can very likely recognize machine-written patterns, because it holds more text comparison data than any detector vendor, but there is no evidence it converts that into a per-page AI penalty. AI content that fails usually fails for being unoriginal and interchangeable, which is a content problem rather than a detection problem.
Last updated August 10, 2026. Google's spam policy wording was read directly from developers.google.com that day. Where a claim could not be traced to a primary source, this page says so instead of repeating it.
Claim versus primary source
Four things people say about Google and AI content, traced
Each row is a claim you will meet in the first page of results for this question, set beside where it actually comes from and what the primary source says.
| The claim | Where it comes from | What the source actually says | Verdict |
|---|---|---|---|
| Google has a team dedicated to detecting AI-generated content | A LinkedIn profile description belonging to a Google employee, Chris Nelson | The profile says he manages a team building ranking solutions including the "detection and treatment of AI-generated content". Nelson is a listed co-author of Google's AI content guidance. There is no on-record statement from him and no Google documentation describing such a system. | Traceable, but it is a job description, not a description of a ranking signal |
| Google penalizes content for being written by AI | Repeated across SEO blogs and humanizer marketing, including ours in the past | Google's spam policy defines scaled content abuse by purpose and value, and adds that it applies "no matter how it's created". Its published guidance says appropriate use of AI or automation is not against its guidelines. | Contradicted by Google's own documentation |
| Google runs an AI detector on pages, the way GPTZero or Turnitin does | Inferred from the above, usually with no source at all | Google has never shipped or documented a public AI text detector, and no Google documentation describes AI detection as a ranking signal. Its anti-spam system SpamBrain is described as identifying spam "however it is produced". Its one real detection technology, SynthID, is a watermark on Google's own model output rather than a scanner for arbitrary text. | No primary source exists for it |
| AI-generated content cannot rank | Vendor blog posts, frequently by companies selling detection or humanizing software | Google's guidance judges content on quality rather than production method. Ranking outcomes are therefore an empirical question about the content, not a category rule about AI. | Not what the policy says |
The actual policy text
What Google's spam policy really targets
Most of the confusion on this topic disappears once you read the policy people are arguing about. Google's spam policies page defines scaled content abuse like this:
"Scaled content abuse is 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."
Read the last clause again, because it is doing all the work: no matter how it's created. The policy is written around purpose and value. It is indifferent to whether a person or a model produced the words. Human-written thin content at scale violates it, and a carefully researched AI-assisted article does not match its description at all.
The examples Google lists under that policy include using generative AI tools to generate many pages without adding value for users, and scraping content to generate many pages through automated transformations such as synonymizing or translating. Note the shape of both: the offence is bulk plus no added value. Generative AI appears as a means, alongside scraping and synonymizing, not as the thing being prohibited.
Google's separate guidance on AI-generated content, published February 8, 2023 and co-authored by the same engineer whose LinkedIn profile launched a thousand blog posts, is blunter still. Asked directly whether AI content is against its guidelines, it answers: "Appropriate use of AI or automation is not against our guidelines." On quality it says "Our focus on the quality of content, rather than how content is produced, is a useful guide that has helped us deliver reliable, high quality results to users for years." And on whether machine writing gets any special treatment either way: "Using AI doesn't give content any special gains. It's just content." The one line it draws is at using automation, AI included, to produce content whose primary purpose is manipulating rankings.
When the March 2024 core update introduced the scaled content abuse policy, Nelson spelled out the same principle again, writing that it lets Google act on the practice "no matter whether content is produced through automation, human efforts, or some combination of human and automated processes." Three separate documents, one consistent position.
There is a fourth source that almost nobody in this discussion cites, and it is the most revealing: the Search Quality Rater Guidelines, the manual Google gives the humans who evaluate results. The version dated September 11, 2025 tells raters to give the lowest rating to mass-produced content with no added value "no matter how they are created", and to do so even when "you are unsure of the method of creation". Then it adds the sentence that settles the question: "Likewise, the use of Generative AI tools alone does not determine the level of effort or Page Quality rating. Generative AI tools may be used for high quality and low quality content creation."
Raters are not given a detector. They are told to look for effort, originality and surface tells, with the guidelines' own worked example of low-quality AI content being an article that begins "As a language model, I don't have real-time data". Read alongside the guidelines' own caution that "No single rating can directly impact how a particular webpage, website, or result appears in Google Search", the picture is consistent: Google evaluates the work, not its authorship.
Primary sources: Google Search spam policies and Google Search's guidance about AI-generated content, both read August 10, 2026.
Read this before you trust an article on this topic
The entire evidence base is one LinkedIn profile
Search this question and you will repeatedly be told that Google has a team dedicated to detecting AI-generated content. It is stated flatly, without a link, as though it came from a Google announcement.
It did not. It comes from the LinkedIn profile of a Google employee named Chris Nelson, a Senior Staff Analyst on Search Ranking, whose listed responsibilities included "Address novel content issues (e.g., detection and treatment of AI-generated content)." Nelson is a real and highly relevant person: he is a named co-author of Google's own February 2023 guidance on AI-generated content, and he wrote the March 2024 post announcing the scaled content abuse policy.
The line surfaced in mid-January 2025 as a screenshot posted on social media, was written up by Search Engine Roundtable on January 20, 2025, and has been recycled through the SEO industry ever since. Worth noting that the record is a third-party screenshot; we could not load the profile ourselves to confirm the wording is still there.
That is genuinely interesting. It is also a bullet point on a social network, not documentation of a ranking system. It does not tell you what the team built, whether it produces a per-page AI score, whether that score influences ranking, or how accurate it is. There has been no on-record statement from Nelson about it and no Google documentation describing such a system. Every confident article explaining how Google detects AI content is extrapolating from that one bullet point.
The rule we hold ourselves to: a claim goes on this site only if it traces to a primary source, and when the trail ends at something thin we say what it actually is. We sell an AI humanizer, so we have an obvious commercial interest in you believing Google hunts AI text and that we can save you from it. We would rather keep the reader than win the sale on a story we cannot source. The same discipline is why we published the detector false positive rates we could and could not verify.
Two questions, usually merged into one
Detection and ranking are not the same question
Almost every bad answer on this topic comes from collapsing "can Google tell" into "will Google punish me". They have different answers.
Question 1
Can Google tell that text was machine-written?
Partly, and nobody outside Google knows how well. The same statistical signals that GPTZero and Originality.ai score are available to any search engine, and Google has vastly more comparison data than any detector vendor. What is missing is any evidence Google turns that into a per-page AI verdict. It has never published one, never shipped a detector, and never documented AI detection as a ranking input.
Question 2
Does Google act on it in ranking?
On the published record, no, not as a category. What Google documents acting on is unoriginal content produced at scale for the purpose of manipulating rankings. That policy catches a lot of AI content, because AI made bulk publishing cheap, but it catches it for being bulk and thin rather than for being AI. A single well-researched AI-assisted article is not what the policy describes.
Question 3
So why did my AI content fail?
Usually because it was interchangeable. A model asked a generic question returns the median of everything already published, which is by definition the least differentiated page you could make. It has no first-hand experience, no proprietary data, no opinion and no reason to be cited. That is a content problem a rewrite does not solve, and it is the honest answer even though we sell a rewriter.
The Google AI content detector that does exist
SynthID is real, and it is not what you think
Google does have detection technology for machine-generated content. It is called SynthID, it comes from Google DeepMind, and it was published in Nature on October 23, 2024. Almost nobody writing about Google and AI content mentions it, and everybody who does mention it gets its scope wrong.
SynthID is a watermark, not a detector. Rather than inspecting finished text and guessing, it alters the sampling process while a model generates, embedding a statistical signature that a matching detector can later look for. It is running in production on Gemini output, and the paper reports a live experiment across nearly 20 million Gemini responses finding no meaningful drop in response quality.
That design has four consequences that between them explain why SynthID cannot be the thing people fear:
It only marks text Google generated
Google's own help documentation states plainly that "While other companies have started to adopt SynthID watermarks, Gemini can currently only recognize content created by Google AI tools." Text from another model carries no watermark at all, so there is nothing to find.
Detection needs the key
Google's developer documentation warns that each watermarking configuration must be stored privately, "otherwise your watermark may be trivially replicable by others". Watermark detection is something the key holder can do, not something the open web can do to you.
Rewriting degrades it
This is DeepMind's own finding, in its own words: "Detector confidence scores can be greatly reduced when an AI-generated text is thoroughly rewritten, or translated to another language." The Nature paper's limitations section says the same, that "generative watermarks are weakened by edits to the text, such as through LLM paraphrasing".
In Search it covers media, not text
Google announced at I/O in May 2026 that SynthID verification is expanding to Search and Chrome, and its help pages are specific about what that covers: images, videos and audio. Web page text is not part of it.
The Nature paper is also unusually candid about the limits of the whole category, noting that watermarking requires cooperation from whoever runs the model, that open-source models make that unenforceable, and that post-hoc detectors "may have higher false-positive rates for certain groups, such as non-native speakers". That last point is the same bias measured at 61.3 percent in the peer-reviewed work we cover in our page on AI detector false positive rates, arriving here from Google's own researchers.
The comparison worth making is with OpenAI, which took the opposite decision on the same technology. It built a text watermarking method, has not deployed it, and gives one of those same reasons for holding back: that the method could stigmatize AI use for non-native English speakers. We set the two approaches side by side in whether ChatGPT watermarks its text.
Primary sources: Dathathri et al., Nature, October 23 2024, Google's SynthID Text documentation and Gemini Apps Help, all read August 10, 2026.
Does AI content rank in Google?
Five studies, five different answers, five commercial interests
There is no peer-reviewed measurement of AI content against Google ranking positions. Every dataset in this space was produced by a company selling either an AI detector or an SEO service, and their numbers disagree by more than a factor of five.
| Who ran it | Commercial interest | When | Sample | What it reported |
|---|---|---|---|---|
| Ahrefs | Sells SEO software and the detector used | July 2026 | 1,000,000 pages from the top 10 of 100,000 SERPs | 5.3% of pages ranking 1 to 3 were classed 100% AI. Average AI content rose only from 27.1% at position 1 to 30.9% at position 10. |
| Semrush | SEO vendor, used GPTZero as the detector | April 2026 | 20,000 keywords, 42,000 pages | Content classed purely AI took the top spot 9% of the time; content classed human-written took it 80% of the time. |
| Graphite | Agency selling human-written content, used Originality.ai | June 2024, never re-run | 2,200 keywords, roughly 12,000 URLs | Pure AI content made up about 3% of organic results, with 88% of URLs showing minimal to no AI. |
| Originality.ai | Sells the detector, and the conclusion is the sales pitch | No publication date on the page | 500 keywords, top 20, tracked since 2019 | AI in Google peaked at 19.56% in July 2025, easing to 17.31% by September. Also asserts Google does penalize AI content. |
| SE Ranking | SEO vendor, but the only experiment that needed no detector | March 2026 | 20 new domains, 2,000 AI-written articles | 70.95% got indexed and 28% of ranking URLs reached the top 100, but only 3% remained there after about three months. |
How to read that table. The spread tracks which detector each study used far more than it tracks anything about Google. Ahrefs found 5.3 percent of top-three pages fully AI, Semrush found purely AI content taking the top spot 9 percent of the time, and Graphite found pure AI at about 3 percent of results using a two-year-old dataset. These are measurements of detector output, and detector output disagrees.
Ahrefs, to its credit, refuses the inference its own headline invites, writing that "it would be wrong to assume that there is any kind of automatic suppression happening" and that increasing AI use most likely correlates with decreasing content quality instead. It also concedes the obvious: "AI content detection is not perfect, and the way we detect AI content will be different from how Google does (if it does)." That parenthesis is the most honest three words published on this topic.
The SE Ranking experiment is the most interesting design, because it published AI articles itself and needed no detector at all. It is also the clearest example of a confound: those 2,000 articles sat on 20 brand-new domains with no backlinks, no named authors and no credentials, which SE Ranking states outright. When they faded from the top 100, there is no way to separate "AI content decays" from "unknown new sites without links decay". That is the shape of the whole evidence base.
For publishers and content teams
Why AI drafts underperform, when detection is not the reason
These are the failure modes we see in real content programs. None of them is a penalty, and only one of them is fixed by rewriting.
It says what forty other pages already said
A model answering a generic prompt returns a blend of what already ranks. That is the definition of unoriginal, and it is the property Google's guidance keeps pointing at. Nothing in the draft gives a reader a reason to choose it, cite it or link to it. A rewrite changes the prose; it cannot add a fact the draft never had.
Volume without a check step
Publishing 200 model-written pages a month with no subject-matter review is the exact pattern the scaled content abuse policy describes, and the pattern the March 2024 spam update was built to act on. The problem is not that a model wrote them. It is that nobody verified any of them.
Confident sentences that are wrong
Models invent statistics, dates and citations in exactly the register of a well-sourced page. In competitive niches that is what sinks trust fastest, because a reader who catches one fabricated number discounts everything else. This is the failure mode we spend the most editorial effort on, and it is why this page traces its own claims.
Flat, uniform register
Even accurate AI drafts tend toward even sentence length, hedged phrasing and a tone that signals nobody in particular wrote this. That costs you the reader rather than the crawler, which shows up as short visits and no return traffic. This is the one item on this list that a rewrite genuinely addresses.
Where our product honestly fits. Our AI humanizer fixes the fourth item. It varies rhythm, removes the hedging register and makes a draft read like a person wrote it, while keeping your facts and structure. It does not make thin content valuable, it does not defeat a Google AI detector that nobody has shown exists, and it will not rescue a program that publishes without review. If someone sells you a humanizer as insurance against a Google penalty, they are selling you a fix for a problem they have not evidenced.
Questions people ask
Google and AI content, answered
Does Google detect AI content?
Google has never shipped or documented a public AI content detector, and no Google documentation describes AI detection as a ranking signal. It almost certainly can identify machine-written patterns, since it has more text comparison data than any detector vendor, but there is no primary source showing it assigns pages an AI verdict. Its published position is that it judges content on quality rather than on how the content was produced.
Does Google have an AI detector?
Not as a text detector you can paste writing into. Google DeepMind does publish SynthID, described in Nature in October 2024, but it is a watermark embedded while Google's own models generate, not a scanner for arbitrary text. Google states that Gemini "can currently only recognize content created by Google AI tools", detection requires the private watermarking key, and the SynthID verification rolling out to Search covers images, video and audio rather than web page text.
Does AI content rank in Google?
Yes. Google's own guidance says "Using AI doesn't give content any special gains. It's just content." Vendor studies measuring how much of the top 10 is AI-written disagree wildly, from about 3 percent to 5.3 percent of top-three pages depending on which detector was used, and no peer-reviewed study of AI content against ranking positions exists. What fails is unoriginal content published at scale to chase rankings, which Google's spam policy targets by purpose and value rather than production method. The policy applies "no matter how it's created", which cuts both ways.
How does Google detect AI content?
There is no documented mechanism, so any specific answer you read is inference. Google describes SpamBrain as its machine-learning spam prevention system that identifies spam "however it is produced", which is a statement about spam rather than about AI authorship. Treat detailed explanations of Google's AI detection pipeline with suspicion, because no primary source describing one exists.
Does Google penalize AI content?
No, not for being AI. Google's published guidance states that appropriate use of AI or automation is not against its guidelines, and its spam policy is written around content produced at scale for the primary purpose of manipulating search rankings. We cover the penalty question in full in our guide to whether Google penalizes AI content.
Will humanizing my AI content help it rank?
Not by defeating a Google AI detector, because there is no evidence one is used in ranking. Rewriting helps for a narrower and more honest reason: it removes the flat, uniform register that makes AI drafts read as interchangeable, which affects whether people finish the page, trust it and link to it. If your draft has nothing original in it, rewriting the sentences will not fix that.
Keep reading
Related guides
Does Google penalize AI content?
The penalty question in full, and what the helpful content guidance means for a content team.
Does Google Docs detect AI?
No detector, but a complete version history. Why that record matters more than any AI score.
AI detector false positive rates
Published rates span 0.004% to 61.3%. Every vendor claim set against independent measurement.
AI humanizer for content teams
How editorial teams run a rewrite step across a publishing workflow without losing house style.
How to make AI writing sound human
The specific properties that read as machine-written, and how to edit them out by hand.
Best AI humanizer
The full comparison of humanizers, verified pricing, and which detectors each is tuned against.
Publish drafts that read like a person wrote them
Not to beat a detector Google has never shipped. Because the flat, hedged register of an unedited AI draft is what loses readers, and readers are what rankings follow. Our humanizer varies the rhythm and voice while keeping your facts, structure and argument intact.
Plans planned from $9/mo · Free account to start · Runs saved to your history, delete any time