Does AI content hurt your website? What the research shows
Does AI content hurt your website? Two large studies, 600,000 pages and a 16-month experiment, give a clear answer. Sources included.
Written by Vera, VEO by VERA's intelligence layer, and reviewed by the VEO by VERA team before publishing.
"Everyone says publishing AI content will wreck your website. Is that true?"
It is one of the most common questions we get, and the concern is fair. There is a lot of noise on this topic and very little evidence behind it. So we gathered the research that does exist, sources included, so you can read for yourself what was actually measured.
The short answer: the question is not who writes the first draft. It is what happens after that.
What Google actually prohibits
Google publishes documentation on this. Nowhere does it say AI content is banned. What is banned is called scaled content abuse: publishing large numbers of pages mainly to manipulate rankings rather than to help anyone.
The wording Google uses describes content "created with little to no effort, little to no originality, and little to no added value". Note what is missing: there is nothing about the tool the text was made with.
That is not a technicality. The same rule applied long before AI. Cheap content mills producing hundreds of thin pages hit exactly this limit. AI has simply lowered the cost of scaling badly, which is why you now see so much more of it.
What the research shows
Two large studies now exist, and together they paint a surprisingly clear picture.
Ahrefs analysed 600,000 web pages for how much AI they contained. The result: 86.5% of top-ranking pages contain AI-generated or AI-assisted text. Only 4.6% is fully AI-written, and the large majority is a blend of AI and human input. The correlation between the share of AI text and ranking position was 0.011, which in practice means no relationship at all. Google does not penalise AI content, but it does not reward it either.
SE Ranking ran the opposite experiment for 16 months. They set up 20 brand-new domains and published 100 entirely unedited AI articles on each, 2,000 in total. No editing, no internal links, no authority, no human review whatsoever. The outcome is precisely the horror story the sceptical agencies point to:
| Period | Share of pages in the top 100 |
|---|---|
| Month 1 | 28% |
| Months 3 to 6 | 3% |
| Month 16 | 20% |
Across the full 16 months, those 2,000 articles produced 1,381 clicks between them. That is fewer than 0.7 clicks per article, over a year and a half.
The same company also published edited AI articles on its own established domain. Same language model, different process. Result: over 555,000 impressions, more than 2,300 clicks and several top-10 positions.
Two experiments, the same tool, opposite outcomes. The entire difference sits in what happens between generating and publishing.
Why it still goes wrong in practice
The companies issuing warnings have identified a real risk. They just attach it to the wrong cause. Here is what actually does the damage:
- Publishing without anyone reading it. Factual errors and invented details undermine exactly the authority you are trying to build.
- Volume that does not match your site. A young domain suddenly pushing out two hundred pages looks unnatural, regardless of quality.
- Overlap between your own pages. Twenty articles saying largely the same thing compete with each other instead of with your competitor.
- Nothing of your own. With no proprietary data, experience or point of view, there is nothing a reader cannot find elsewhere. That is precisely what "little to no added value" means.
Notice the word in the original question: blindly. That is where the whole issue lives. Copying blindly is the problem. AI is not.
The difference in practice
The line is sharper than it looks. This is what separates the two sides:
| Goes wrong | Works |
|---|---|
| Generate and publish immediately | Generate, review, then publish |
| Volume set by what the tool can produce | Volume set by what your site can carry |
| A generic prompt about a topic | Built on real search data and your own audience |
| Nobody is accountable | A human puts their name on it |
| Nothing original added | Your own numbers, experience or position included |
Google also advises being transparent about how your content is produced, in a way that makes sense for your audience. That is not penance. It is simply good craft.
How we handle this
Our position is not "AI writes your content and you do nothing". That is exactly the model that collapsed after three months in the SE Ranking experiment.
What we do is automate the analysis and the first draft, and leave you in control. Publishing is a deliberate step here, never automatic. Content is built on real search data and on your category and audience, not on a generic prompt. And you see measurable results, so you are not steering on gut feeling.
That takes more attention than pressing a button. That is the point: the research shows this difference is exactly what determines whether your content still exists in three months.
Want to know where your site stands in Google and in AI answers today? Request a free scan. For more background, read What is AI visibility and why is it important? and SEO vs GEO vs AEO: What is the difference?.
Sources
- Ahrefs: AI-Generated Content Does Not Hurt Your Google Rankings (600,000 pages analysed)
- Search Engine Land: How AI-generated content performs in Google Search, a 16-month experiment
- SE Ranking: How AI-Generated Content Performs, Experiment Results
- Google Search Central: Google Search's guidance about AI-generated content
- Google Search Central: Spam policies for Google web search (scaled content abuse)
Related articles
Source & method
This article was written by Vera following the See/Think/Do/Care content framework and reviewed before publishing — Vera never publishes on its own.