In short
Yes, AI-written content can rank, and Google's stated policy since 2023 is that how content is produced does not matter; whether it is helpful, original and made for people does. Google does not run an "AI detector" and detectors in general are unreliable. What it does act on is the pattern AI makes easy: large volumes of generic pages that add nothing, which its spam policies name as scaled content abuse. AI content fails in search when it is unedited, unsourced and indistinguishable from a thousand other pages, not because a machine wrote it.
What Google actually says
Google's guidance, published in February 2023 and unchanged in substance since, is that using automation including AI to generate content is not against its policies, that it rewards high-quality content "however it is produced", and that using AI to manipulate rankings is a violation of its spam policies. In March 2024 it added a specific policy, scaled content abuse, covering the production of many pages for the purpose of manipulating rankings regardless of whether humans or machines made them.
So the honest answer to "does AI content rank" is that Google has explicitly said it can. The interesting question is why so much of it does not.
The detection myth
There is a widespread belief that Google runs every page through an AI detector and demotes what it flags. There is no evidence of this, Google has said it does not, and the technology would not support it: commercial AI detectors produce high false-positive rates on human writing, particularly from non-native speakers and from anyone who writes plainly, and they are trivially defeated by light editing. A search engine that penalised on that basis would be demoting its own best results at random.
What Google demonstrably does evaluate is the thing detection would be a proxy for: whether the page is useful, whether it says anything not already said, whether it demonstrates experience and expertise, and whether the site as a whole looks like it exists to serve readers or to harvest them. Those are the signals behind E-E-A-T and the Helpful Content system, and unedited AI output scores badly on all of them, not because of its origin but because of its content.
Why most AI content fails
It is the average of what already exists. A language model writes the most probable text given the prompt, which for a general topic means a competent summary of the current top ten results. Google has no reason to rank an eleventh version of the same answer. The page that ranks is the one that adds something: first-hand data, a specific case, an opinion with reasons, a step the others missed.
It has no experience in it. The E in E-E-A-T is experience. "We tested this", "our client saw", "when we tried it the second step failed" cannot be generated, only reported. Content without it reads as what it is.
It is unsourced or mis-sourced. Models invent citations, dates, and statistics with total confidence. A page carrying an invented figure is a page carrying a reason for a reader, and a quality rater, to distrust the whole site.
It comes in bulk. The economic case for AI content is volume, and volume is exactly what the scaled content abuse policy targets. We have direct experience: our own guides section once held 1,072 templated pages across 134 platforms with 18 of 25 sentences identical between siblings, and we consolidated it to 403 because it was precisely the pattern the policy describes.
What ranking AI-assisted content looks like
The distinction that matters is not AI versus human but assisted versus generated.
The expertise comes from a person. Someone who knows the subject decides what the page should say, what is wrong with the existing answers, and what specific experience or data it will contribute. That is the part that earns the ranking.
The model does the labour, not the thinking. Drafting from a detailed brief, restructuring, tightening, producing variants of a headline, summarising a source the author has actually read. All legitimate and all faster.
Every claim is checked. Numbers, names, dates, and quotations are verified against the source or cut. A byline means a person stands behind the page.
Volume is set by what you can stand behind, not by what you can generate. If the editorial process can genuinely review and improve two pieces a day, that is the publishing rate. The temptation to publish two hundred is exactly the trap.
Done this way, AI is a productivity tool and the output is indistinguishable in quality from good human writing because a human made it good. Done the other way, it is a content farm, and content farms lose regardless of the technology they use.
The disclosure question
Google does not require disclosure of AI use and has said so. Some publishers disclose as a matter of policy; some regulators and industries are moving toward requiring it for certain content types. The sensible position is the same as for any tool: a real author with real accountability is named, and if a reader would feel misled to learn how the piece was produced, that is a signal about the piece, not about the disclosure.
The practical test
Before publishing anything produced with AI help, ask: what does this page say that the top results do not? If the answer is nothing, it will not rank, and no amount of optimisation changes that. If the answer is a specific thing a person contributed, the tool that typed the rest is irrelevant.
If you are building a content programme and want it to use AI without turning into the pattern Google penalises, book a call; we run one ourselves and can show you the workflow.
Common questions
Does Google penalise AI-generated content?
No, not for being AI-generated. Google's policy since 2023 is that content is judged on quality and helpfulness however it is produced. What it does penalise, under its scaled content abuse policy, is producing large volumes of low-value pages to manipulate rankings, which AI makes easy but which applies equally to human-made pages.
Can Google detect ChatGPT content?
Google has said it does not use AI detection to rank pages, and the detectors that exist are unreliable, with high false-positive rates on ordinary human writing. What Google evaluates is whether the page is useful, original and shows experience. Unedited AI output tends to fail those tests on its merits, not because it was identified as machine-written.
Why doesn't my AI content rank?
Usually because it restates what the top results already say without adding first-hand experience, data, or a specific point of view, so Google has no reason to prefer it. Invented statistics and missing sources make it worse. The fix is editorial: a person with expertise deciding what the page contributes, and verifying every claim.
Do I have to disclose that content was written with AI?
Google does not require it. Some industries and jurisdictions are moving toward disclosure rules for particular content types, and some publishers disclose by policy. What matters for search and for trust is that a named, accountable author stands behind the page and that its claims are verified.
How much AI content is too much?
The limit is set by review capacity, not generation capacity. Publish only what a knowledgeable person has genuinely edited, checked and improved. If output is outpacing what your team can stand behind, you are producing the volume pattern Google's spam policies target, whatever the per-page quality looks like.
