Google penalises arbitrariness. The tool is irrelevant
Google’s stance on AI texts has been consistent for years, and it is remarkably straightforward. Their own guidelines say, in essence: how a piece of content is produced does not matter. What matters is whether it offers readers something of value or exists only to game a ranking. In March 2024, Google drew a hard line by introducing a rule against “scaled content abuse”, the mass publication of low-value pages.
The clever part of that rule is that it deliberately sidesteps the question of tooling. It applies equally to bulk content churned out by AI and to bulk content typed up by an army of poorly paid writers. The stated goal was to displace forty percent of low-quality results. The actual figure turned out to be forty-five. About a year later, in January 2025, Google followed up by instructing its quality reviewers explicitly: a page whose content is almost entirely AI-generated with little original contribution receives the lowest possible rating.
The 850,000-page website did not fall because AI wrote it. It fell because nobody had added anything that did not already exist a hundred times over elsewhere on the web. And the fall did not stop at Google: the domain disappearing from ChatGPT responses at the same time shows how tightly the two worlds are now connected. Failing in search often means failing in AI systems as well.
Why “slop” was the word of 2025
There is a word that describes the current state of the web with some precision. Merriam-Webster, one of the major American dictionaries, named “slop” its word of the year in December 2025, defining it as low-quality digital content, typically produced in bulk by artificial intelligence. Dictionaries react late, not early. When a term reaches that stage, the phenomenon has long since left the niche.
Lily Ray, who analyzes search at the agency Amsive, describes the consequence this way: generative AI has amplified the same old cycle, flooding the web with repetitive, low-value content and forcing Google into increasingly firm responses.
The flood raises the bar for everyone. When the web fills up with mediocrity, what genuinely helps becomes more valuable and more rare. Google’s updates follow exactly that logic. For a business working honestly, this is more of a good sign than a bad one.
I use AI too. That is not the point.
It would be dishonest to write this piece and pretend I never touch AI tools. I do. Not to produce texts on a conveyor belt, but to speed up research, compare phrasings, and pressure-test structures. The difference is not in the tool. It is in whether the result is something that would not exist, in this form, without this page and the thinking behind it.
That is also the line that determines whether using AI helps or harms your ranking. It does not run between “by hand” and “by AI”. It runs between care and arbitrariness.
| Workable use | Slop |
|---|---|
| AI for research, drafts, variants | AI as sole author, unreviewed |
| A person checks the facts and owns them | Facts taken over without verification |
| Your own angle, your own data, your own experience | Reformulated content that already exists elsewhere |
| As many pages as needed | As many pages as possible |
| Goal: the reader knows more afterwards | Goal: trick the search engine |
From August, it is not just Google watching
That line between care and arbitrariness also appears in law from 2 August 2026 onward. The EU AI Act will then require AI texts on topics of public interest to be labeled, with an exemption for anyone who reviews the content and takes responsibility for it. What is remarkable is how closely the two things align: what protects you from a Google downgrade is almost identical to what exempts you from the labeling requirement. Editorial control. The details are in the dedicated post on the AI Act.
Danny Sullivan, Google’s public-facing voice, summarizes the whole debate with disarming simplicity: SEO for AI is still SEO. Write for people, not for ranking systems, whether the system behind it is classic search or a language model.
If you want to use AI without putting your ranking at risk
The question is rarely whether to use AI, but how. Building content that holds up in search and AI services takes someone who knows that line and keeps to it under time pressure. That is where my work starts. Let’s talk.
Frequently asked questions
Does Google penalize texts written with AI?
No, not because of the tool. Google evaluates whether a text offers value to readers, regardless of how it was produced. What gets penalized is mass-produced, arbitrary content with no added value, and that applies equally to human-written filler and AI-written texts.
So why did an entire website get dropped from the index?
Because its content was produced in enormous quantity but without any original value. Google calls that “scaled content abuse”. The fact that the pages were AI-generated was not the reason; it was simply the method by which arbitrariness scaled into volume.
Can I safely use AI for my website copy?
Yes, as long as a person checks the facts, brings their own perspective, and takes responsibility for the result. Using AI as a tool for research, drafts, and variants is fine. It becomes risky when nobody is looking anymore and pages are published purely to add to the pile.
Do I need to label AI texts from August 2026?
Only in narrow cases. The AI Act requires labeling for AI texts on topics of public interest. Anyone who reviews and takes responsibility for the content is exempt. Ordinary marketing and product copy does not fall under this obligation at all. More on that in the dedicated post on the AI Act.
Can AI content also hurt my visibility in ChatGPT and Perplexity?
It can. In the documented case, the penalized site disappeared from ChatGPT responses shortly after the Google penalty. A source that is seen as low-value in classic search tends to be cited less often by AI services as a reliable reference.
How much AI content is actually in search results?
Analyses estimate the share in Google’s top results at just under one fifth, with a slight downward trend since Google has been enforcing more strictly. These figures come from automated AI-detection tools and should be read as an order of magnitude, not a precise measurement.