Search “opinions on AI-generated content for SEO” and you get a wall of confident, contradictory takes. One post says AI content is fine and Google has said so directly. The next shows a site that published hundreds of AI articles and lost almost all its traffic in a single update. Both are describing real things.
This article lays out the actual range of views, with sources you can check: what Google has documented, what the cautious camp has seen happen, what the “it’s just a tool” camp reports, and what the studies do and do not prove. The evidence on both sides is thinner and messier than most headlines admit, and that is worth saying up front.
The question itself is usually framed wrong. Google’s systems do not detect or penalise AI authorship. They target content that is low value, unoriginal, or mass-produced to manipulate rankings. AI makes that kind of content cheaper to produce at scale, but it does not require it, and plenty of AI-assisted work never goes near that line.
Key Takeaways
- Google does not penalise content for being AI-generated. Its guidance since February 2023 has been to reward “high-quality content, however it is produced,” judged on helpfulness, originality, and E-E-A-T.
- Scaled content abuse is a spam-policy violation, tightened in March 2024. The trigger is many pages made mainly to rank and not to help users, “no matter how it’s created.”
- The skeptic camp has real examples. Sites that mass-published unedited AI content have been hit hard by core and spam updates, and one analysis found every site deindexed with a manual action in March 2024 showed signs of AI content.
- The pragmatist camp also has data. A 2026 Ahrefs study of roughly 150,000 pages found only a gentle inverse correlation between AI content and rankings, and concluded Google is judging quality, not origin.
- Most of the “evidence” on both sides is weak. AI detectors are unreliable, SERP studies are correlational, and single-site stories cannot isolate AI as the cause.
- AI-assisted, human-edited, evidence-grounded content is a different thing from raw model output published unreviewed. The risk sits almost entirely with the second one, and rises for reviews, YMYL topics, and anything scaled without editorial investment.
Why “Is AI Content Good or Bad for SEO” Is the Wrong Question
The debate gets stuck because people are arguing about the tool when the disagreement is really about the output. “AI content” covers a first draft that a subject expert then rewrites and fact-checks, and it also covers 1,800 pages spun out of a competitor’s sitemap overnight. Those are not the same product, and Google’s systems do not treat them the same way.
A more useful framing: Google’s ranking and spam systems are built to find low-value and manipulative content, whoever or whatever produced it. AI lowers the cost of producing that kind of content at volume, which is why the two topics keep getting tangled together. But the failure mode being penalised is the low value and the manipulation, not the model.
Google’s Documented Position
Google has been unusually explicit about this. Its February 2023 guidance on AI-generated content says the company focuses on “rewarding high-quality content, however it is produced.” The FAQ on that page is blunt:
- “Appropriate use of AI or automation is not against our guidelines.”
- “Using AI doesn’t give content any special gains. It’s just content. If it is useful, helpful, original, and satisfies aspects of E-E-A-T, it might do well in Search. If it doesn’t, it might not.”
- “If you see AI as an inexpensive, easy way to game search engine rankings, then no.”
That last line is the hinge. Google’s position is not “AI is fine” without qualification. It is “AI is fine if the goal is to help people, and it is spam if the goal is to manipulate rankings.” That distinction runs through all of its documentation.
Scaled content abuse: the policy that does apply
In March 2024, Google added three spam policies and rewrote one. The relevant one is scaled content abuse, defined as content “generated for the primary purpose of manipulating search rankings and not helping users,” typically “large amounts of unoriginal content that provides little to no value to users, no matter how it’s created.” The listed examples include “using generative AI tools or other similar tools to generate many pages without adding value for users.”
Google’s own explanation of the change is worth quoting. It says the new policy “builds on our previous spam policy about automatically-generated content, ensuring that we can take action on scaled content abuse as needed, no matter whether content is produced through automation, human efforts, or some combination.” In the March 2024 announcement it added that its “long-standing spam policy has been that use of automation, including generative AI, is spam if the primary purpose is manipulating ranking.”
The same update also involved core ranking changes aimed at unoriginal, unhelpful content. Google said it expected the combined effort to “reduce low-quality, unoriginal content in search results by 40%.” For a real, documented example of what that enforcement looks like from the receiving end, and how a site recovers, see our job board spam-update recovery case study.
The E-E-A-T angle: AI has no first-hand experience
Google’s helpful content guidance frames quality around Experience, Expertise, Authoritativeness, and Trustworthiness, and states that “of these aspects, trust is most important.” The first E, Experience, is the one AI structurally cannot supply. Google describes it as expertise “that comes from having actually used a product or service, or visiting a place.”
For a lot of queries this does not matter. A definition of a marketing term or a comparison of file formats does not need lived experience. But for product reviews, medical or financial advice, travel recommendations, and hands-on how-to content, readers and raters expect evidence that a person did the thing. A model can describe a hiking trail convincingly without anyone having walked it, and that is exactly the gap the Experience signal is meant to catch. Google also asks publishers to consider “Who, How, and Why” a piece of content was made, and to add an AI or automation disclosure “for content where someone might think, How was this created?”
The Skeptic Camp: AI Content Is Derivative by Construction
A large group of publishers and SEOs argue that, whatever the policy says, AI content tends to lose. Their case rests on three things.
Documented losses after updates. After the March 2024 core and spam updates, a wave of sites lost most of their visibility or were deindexed entirely, and a visible share of them were AI content operations. Originality.ai, a company that sells an AI-detection tool, analysed sites hit with manual actions and reported that every one showed signs of AI-generated content, with about half publishing 90 to 100 percent AI content. Treat the exact figures with care, since the source has a commercial interest and “signs of AI” comes from a detector rather than proof, but the direction is consistent with a lot of independent reports.
High-profile flameouts. The most-cited example is the “SEO heist” of late 2023, in which an operator exported a competitor’s sitemap, used AI to generate roughly 1,800 articles on the same topics, and publicly claimed to have “stolen” millions of visits. Follow-up reporting found the traffic collapsed within months, from around 610,000 monthly visits to under 200,000, as the pages lost rankings. Separately, CNET’s 2022 to 2023 experiment publishing AI-written finance explainers under a staff byline ended with corrections on more than half the articles, some substantial, and a lasting hit to the outlet’s credibility.
The originality argument. The structural criticism is that a language model produces a statistically likely synthesis of what already exists. Run at scale against competitive topics, that tends to yield content which is fluent but adds nothing new: no original data, no test results, no point of view, no entities or relationships that were not already well covered. Google’s quality systems are increasingly tuned to reward exactly those things, which is part of why entity coverage and genuine topical depth have become louder signals.
The Pragmatist Camp: It Is a Tool, and the Workflow Decides
The other well-populated camp uses AI daily and reports normal or good results. Their argument is that “AI content” as a category is meaningless, and that what matters is where AI sits in the process.
In this view, AI is used for drafting, research synthesis, outlining, restructuring, and first-pass expansion, while humans keep ownership of the claims, the fact-checking, the original input, and the expert review. Content produced this way is, in their experience, indistinguishable to Google from fully hand-written content, because on the signals that matter it is the same: accurate, specific, genuinely useful, and clearly attributable to a real publisher.
There is data behind this position too. A 2026 Ahrefs study looked at roughly 150,000 ranking pages that had enough text to test, classified them with its own AI detector, and found only “a gentle inverse correlation” between AI content level and ranking position. Pages under 50 percent AI held about 82 percent of top-three positions, but heavily AI-generated pages still appeared throughout the top results. The author’s stated conclusion was that Google “is relying on the same old hallmarks of content quality that it always has. It’s just that AI-generated content is usually lower quality than human-generated content.” The study also flags its own limitation: its detector is imperfect and works differently from whatever Google may or may not do.
Read together, the two camps are closer than they sound. Both agree low-quality content loses. They disagree about how often AI-assisted content manages to clear the quality bar in practice, and about how much editorial effort that takes.
What the Evidence Actually Shows, and What It Doesn’t
It is worth being honest that most public evidence in this debate is weak, on both sides.
- AI detectors are unreliable. They produce false positives on human writing, especially non-native English and older or formulaic text, and false negatives on lightly edited AI output. Any study whose core measurement is “this page looks AI-generated” inherits that error.
- SERP studies are correlational. Finding that lower-AI pages rank better on average does not isolate AI as the cause. Sites that invest in editing, original research, and expert authorship also tend to use less raw AI output, and those investments are plausible causes of the ranking difference.
- Single-site stories cannot separate variables. A site that mass-published AI content and then lost traffic also usually had thin content, aggressive scaling, weak authority, and a business model built on arbitrage. Any of those could carry the blame.
- There is no clean controlled experiment at scale. Nobody has published the same large content set in AI and human versions on comparable domains and tracked them for a year. The small experiments that exist have tiny samples and heavy caveats.
What the evidence does support, consistently, is narrower: content quality predicts ranking outcomes, and mass-produced unoriginal content, however it is made, is being cleared out by core and spam updates. AI is strongly associated with that second category because it makes the category cheap to enter, not because Google has an AI filter.
A Practical Framework: When AI-Assisted Content Is Fine, and When It’s a Real Risk
Instead of a yes or no, run the specific page through five questions. The more answers that land on the higher-risk side, the more editorial investment the page needs before it is safe to publish.
- What kind of query is it? Commodity explainers, definitions, and structured comparisons are lower risk. Reviews, first-hand accounts, and opinion pieces are higher risk, because the value is the experience.
- Is it YMYL? For health, finance, legal, and safety topics, a wrong sentence can cause real harm, raters hold the bar higher, and unedited AI text is a poor fit.
- Does it need original data or first-hand use? If the topic is already well covered and synthesis genuinely helps a reader, AI drafting is reasonable. If readers expect testing, numbers, screenshots, or lived experience, AI alone will produce a thinner page than what already ranks.
- Is a qualified person editing and fact-checking? A named editor who verifies every claim and adds specific detail changes the output category. Publishing model output largely as generated does not.
- Is it being scaled without editorial investment? Steady volume where review scales with output is normal production. Hundreds of pages fast, mainly to capture search traffic, is the pattern that scaled content abuse describes.
If you want the positive version of this, our guide on how to get into Google AI Overview covers what genuinely helpful, extractable content looks like, and the same qualities that earn AI Overview citations are what keep AI-assisted content on the safe side of this framework.
So, Does AI-Generated Content Work for SEO?
Here is the defensible position, consistent with the sources above. Google’s systems do not detect or penalise AI authorship. They detect low-value content and scaled, manipulative content, and they demote it regardless of how it was produced. AI makes that kind of content dramatically cheaper to mass-produce, which is why AI content sites keep showing up in the casualty lists, but nothing about using AI forces you into that category.
AI-assisted content that is edited by someone who knows the subject, grounded in real evidence or original input, and published at a pace that lets review keep up, performs like any other good content. Raw AI output published at volume to chase rankings performs like any other thin content, which is to say it may spike and then lose most of it in the next relevant update. The tool is not the variable. The editorial standard is.
One note on where we stand, since this is a live question the whole industry is still working through: in South Asia Digital’s own content and SEO work, AI is treated as a drafting and research aid inside a human-edited, source-verified process, not a publish button. That is a methodology choice, and it is the one the evidence here supports.
If your actual concern is not Google rankings but whether AI assistants mention your brand, that is a separate problem with its own signals, covered in why your SaaS isn’t showing up in ChatGPT answers.
Frequently Asked Questions
Does Google penalise AI-generated content?
Not for being AI-generated. Google’s documented position is that it rewards high-quality content however it is produced, and judges it on helpfulness, originality, and E-E-A-T. What it does act on is scaled content abuse, meaning many pages made mainly to rank and not to help users, which is a spam-policy violation no matter how the pages are created.
Can AI-generated content rank on the first page of Google?
Yes. Studies of ranking pages consistently find AI-assisted and even heavily AI-generated pages throughout the top results. Fully AI-generated pages are less common at position one, which most analysts attribute to quality rather than an AI filter. Pages that are edited, fact-checked, and add original value rank normally.
Why did some sites lose all their traffic after publishing AI content?
In the documented cases, those sites were mass-publishing unoriginal, unedited content at high volume, mainly to capture search traffic. That matches the scaled content abuse policy and the core-update focus on unhelpful content. AI made the scale possible, but the low value and the manipulation are what triggered the demotion.
Is AI content a bigger risk for some topics than others?
Yes. It is riskier for reviews and experience-led content, where readers expect first-hand use, and for YMYL topics like health and finance, where accuracy standards are higher and a wrong claim can cause harm. It is lower risk for commodity explainers, definitions, and structured comparisons that do not depend on lived experience.
Do I need to disclose that content was made with AI?
Google recommends an AI or automation disclosure for content where a reader would reasonably wonder how it was created, and recommends against listing AI itself as the author. It does not require a disclosure on every page. Accurate authorship information from a real person or organisation matters more for trust.
Is AI-assisted content the same as AI-generated content for SEO purposes?
In practice they behave very differently. AI-assisted content that a qualified person edits, fact-checks, and enriches with original input tends to perform like hand-written content. Raw AI output published with little review carries most of the risk in this debate. Google’s systems judge the finished page, so the editorial process is what separates the two outcomes.
