By Rhodri Williams, founder of Roddy Digital. Published 27 October 2026. Last updated 27 October 2026.
Quick answer
AI-written content is bad for SEO when it is generic, unverified and published at volume. Google's guidance, last updated on 1 October 2026, does not penalise content for being produced with AI. It penalises pages that add little value, whoever or whatever wrote them. The question to ask is whether your page says something a reader cannot get elsewhere.
Introduction
Most business owners we speak to have heard two opposite stories about AI content. One says Google can detect it and will bury your site. The other says you can publish fifty articles a week and watch the traffic climb. Both stories are sold by people with something to sell, and both are wrong in ways that cost money.
We run an AI-first agency. Every article we publish for a client starts with research and structure built with AI, and every one of them is then rewritten, checked and signed off by a person who knows the business. So we have a direct interest in getting this right, and a direct view of what happens when businesses get it wrong.
This article sets out what Google has actually said, where AI-assisted articles lose rankings in practice, and the five checks we run before anything goes live.
What does Google actually say about AI content in 2026?
Google has said the same thing for three years, and its position has hardened only against volume, never against the tool.
In February 2023, Google Search Central published its guidance on AI-generated content. The line most people quote is that Google's systems reward "high-quality content, however it is produced". The line fewer people quote is the condition attached: using automation, including AI, with the "primary purpose of manipulating ranking in search results" breaks its spam policies (Google Search Central, February 2023).
The current guidance page, Google Search's guidance on using generative AI content on your website, carries a last-updated date of 1 October 2026. It says generative AI "can be particularly useful when researching a topic, and to add structure to original content". It also says that using AI to generate many pages without adding value "may violate" the scaled content abuse policy, and it reminds publishers that generative models "don't retrieve facts, but predict a likely sequence of words based on their training data". The practical instruction is plain: fact-check and review before publishing.
The Search Quality Rater Guidelines were updated in January 2025 to name the problem directly. The lowest quality rating now covers "Using automated tools (generative AI or otherwise) as a low-effort way to produce many pages that add little-to-no value for website visitors" (Search Engine Journal, January 2025). Raters do not set rankings, but the guidelines show what Google is training its systems to find.
Read those three documents together and the policy is clear. Google is hunting for pages that exist to fill a keyword rather than to help a reader. AI makes those pages cheap to produce, which is why the enforcement has sharpened. The tool itself has never been the offence.
Why do so many AI-written articles still lose?
Because they are the same article. If a hundred businesses ask the same model the same question, they get a hundred versions of the same answer, with the same structure, the same hedged phrasing and the same absence of anything specific. Google does not need an AI detector to rank those pages poorly. It only needs to notice that none of them adds anything to what it already has.
In our audits, AI-assisted articles that have fallen out of the rankings share four traits:
| Trait | What it looks like on the page | Why it costs rankings |
|---|---|---|
| No first-hand knowledge | Advice that could apply to any business in any town | Nothing for Google or an AI engine to cite that is not already indexed a thousand times |
| No numbers | "Many businesses find" and "studies show" with no study named | Fails the specificity test that helpful content systems reward |
| Unchecked facts | A regulation that changed in 2024, a tool that no longer exists | Reader trust drops, dwell time drops, the page is quietly demoted |
| Volume over depth | Thirty 600-word posts in a month on near-identical topics | Reads as scaled content abuse, the one policy that does carry a penalty |
Third-party tracking by Originality.AI, reported widely during 2025, suggested that something like a fifth of top-20 results carried AI-assisted content by the middle of that year (ICODA analysis of Originality.AI data). Treat that figure as directional rather than precise. What it shows is that AI-assisted pages rank every day. The ones that do are the ones somebody edited into something worth reading.
When is AI content good for SEO?
AI content works when AI does the part it is good at and a person does the part it cannot. In practice that means the model handles research synthesis, outline structure, first drafts of tables and FAQ phrasing, and the person supplies the opinion, the examples, the numbers from real work and the final edit.
The articles that rank for our clients follow that split exactly. A pest control owner in Cardiff tells us which seven entry points his team finds most on surveys; the model helps structure the explanation around them. A client-side project manager explains which decisions are already fixed by the time a client appoints him; the model helps turn that into a stage-by-stage list. Neither article could have been written by a model alone, because the thing that makes them rank is not in any training data.
That is also what answer engines reward. ChatGPT, Gemini and Perplexity cite pages that contain a specific, attributable claim. A page with none is invisible to them regardless of who wrote it.
The five checks we run before any AI-assisted article publishes
This is the editorial test we apply at Roddy Digital. It takes fifteen minutes and it is the difference between content that compounds and content that gets demoted.
- The swap test. Could this article sit on a competitor's site with the logo changed? If yes, it has no reason to rank. Add the thing only this business knows.
- The number test. Is there at least one specific figure, date or named source a reader could check? "Many businesses" is a sign the model was guessing.
- The fact test. Has a person verified every regulation, price, tool name and statistic against a current primary source? Generative models predict words; they do not look things up.
- The voice test. Does a sentence from the article sound like the business owner on the phone? Hedged, symmetrical AI phrasing is the fastest way to tell a reader nobody cared.
- The reason test. Would this page exist if Google did not? If the only reason to publish is to occupy a keyword, it is the kind of page the scaled content abuse policy was written for.
An article that passes all five is helpful content in Google's sense, and it does not matter which tool typed the first draft.
Should you disclose that AI helped write it?
Google's guidance suggests explaining how content was created where it helps the reader, and it requires labelling for AI-generated product titles, descriptions and images in Google Merchant Center. For a standard article, the honest position is a named human author who stands behind every claim, with a visible published and updated date. That matters more to readers and to answer engines than a disclosure line, because a real name with real expertise is what the systems are built to find.
Key takeaways
- Google's guidance, updated 1 October 2026, contains no penalty for AI-written content. The penalty is for pages that add little value at scale.
- Most AI-assisted articles lose because they are generic, unverified and interchangeable, which is a quality failure the tool makes cheaper to commit.
- AI content ranks when a model structures and a person supplies the knowledge, the numbers and the final edit.
- Run the five checks (swap, number, fact, voice, reason) before anything publishes.
- A named, credentialed author matters more than an AI disclosure line.
Frequently asked questions
Does Google penalise AI-generated content?
No. Google's guidance states that it rewards high-quality content however it is produced. What it does penalise is using automation, including AI, to generate many pages with the primary purpose of manipulating rankings. That falls under its scaled content abuse spam policy, which applies whether the pages were written by a model or a person.
Can Google detect AI-written content?
Google has not said it uses AI detection to rank pages, and its guidance focuses on the value of the content rather than its origin. In practice it does not need to detect the tool. Generic, repetitive, unverified pages share the same traits whatever produced them, and those traits are what its helpful content systems are built to find.
Is it safe to use ChatGPT for blog posts?
Yes, provided a person who knows the subject edits and checks the result. Use the model for research, structure and first drafts, then add first-hand knowledge, specific numbers and verified sources. Publishing raw output at volume is the pattern Google's scaled content abuse policy targets, and it rarely ranks for long.
How much AI content is too much?
Volume on its own is not the problem. Thirty thin articles a month on near-identical topics is too much; four deep, verified articles that each say something new is fine. Judge each page by whether it adds something a reader could not find elsewhere, and keep every page to that standard.
Do AI search engines cite AI-written content?
They cite pages that contain a specific, attributable claim and a credible named author. Whether a model helped write the page makes no difference to them. A page with no original number, example or opinion is unlikely to be cited by ChatGPT, Gemini or Perplexity whoever wrote it.
Build content that compounds
If you are publishing with AI and not sure whether it is helping or hurting, we will audit your last twenty articles against the five checks and show you which ones to fix, which to merge and which to leave alone. That is part of how we build practical AI systems for small businesses, and it sits alongside the work in our AI and automation services. If you are earlier in the journey, start with where to actually start with AI and automation and how to get found on Google.
About the author
Rhodri Williams is the founder of Roddy Digital, an AI-first agency in Cardiff that designs and builds the platforms, automations and websites ambitious businesses run on. Every Roddy Digital article is drafted with AI and finished by a person who knows the business it is written for.
