I nonetheless have loads of questions on how Google treats AI-generated content material.
On the one hand, Google says they don’t care whether or not AI is used for content material creation, so long as the content material isn’t designed “with the first function of manipulating rating in search outcomes.”
Automation has lengthy been utilized in publishing to create helpful content material. AI can help with and generate helpful content material in thrilling new methods.
However, individuals like Lily Ray and Glenn Gabe have shared many examples of so-called “mount AI” site visitors graphs: web sites that scaled content material creation with AI, noticed a heroic spike in rankings and site visitors… after which watched their natural efficiency tank a couple of months after publication.
An instance of a “Mount AI” site visitors graph from Lily’s article.
Many entrepreneurs at the moment are unwilling to make use of generative AI in any capability for worry Google will punish them. However is Google punishing these web sites for utilizing AI content material? Or is their AI use incidental to the truth that they’re making spammy, scaled content material?
We got down to reply 4 questions that may shed some gentle on the subject:
- Are there any absolutely AI-generated pages in top-ranking positions?
- How frequent is AI-generated content material all through the highest 10?
- Does Google index content material with a really excessive probability of being AI?
- Does the natural efficiency of AI-generated content material tank inside a couple of months?
Primarily based on this analysis, I consider that Google isn’t in opposition to AI content material; it’s in opposition to dangerous content material, however confusion arises as a result of AI content material and dangerous content material overlap a major quantity of the time.
- 5.3% of top-ranking (positions 1–3) pages are 100% AI-generated, and 9% are ≥80% AI content material—so absolutely AI-written pages can and do rank on the very prime.
- Pages with underneath 50% AI content material account for 82.2% of top-3 rankings, which means closely AI-generated content material continues to be a minority on the prime.
- Each place within the prime 10 accommodates a significant share of closely AI-generated pages: between 8.4% (place 1) and 11.7% (place 10) of pages have ≥80% AI content material.
- Common AI content material degree rises solely barely from place 1 (27.1%) to place 10 (30.9%), and the median tells the same story, climbing from 17.1% to 19.5% over the identical vary.
- Indexation price drops from 49.28% for low-AI-content pages to 40.35% for very-high-AI-content pages: a significant however removed from disqualifying hole.
- Even within the very-high-AI-content bucket, 40% of pages had been nonetheless listed, displaying no binary block on AI-generated content material getting into the index.
- Low and reasonable AI-content pages acquired 2–3x the natural impressions of excessive or very-high AI-content pages, the starkest efficiency hole discovered within the research.
- Excessive and very-high AI-content pages confirmed comparable, steady impressions over time with no precipitous drop-off.
This analysis makes use of Ahrefs AI content material detector, which Ahrefs clients can use by way of Web site Audit and the High pages and Web page examine experiences in Web site Explorer.
AI content material ranges point out the estimated proportion of AI-generated textual content on the web page, starting from low (<20%) to very excessive (≥80%). For better accuracy, AI detection solely triggers on pages not less than 350 phrases in size.
Importantly, AI content material detection isn’t excellent, and the way in which we detect AI content material will likely be completely different from how Google does (if it does).
In the identical means that an LLM makes use of chance to generate textual content, AI detectors use chance to categorise it. Any such analysis is, for my part, the easiest way to make use of AI content material detection: large-scale datasets and directional developments, and never fixating on particular person articles and their AI content material.
We analyzed 1,000,000 pages pulled from the highest 10 positions in 100,000 SERPs in June 2026. Round 300,000 of those pages had been out there in our crawler database, of which 150,000 contained sufficient web page content material for AI detection.
We discovered that almost all of top-ranking pages are predominantly human-written. 54.7%—over half—have lower than 20% AI content material. Collectively, pages with underneath 50% AI content material account for 82.2% of top-3 rankings.
However though our knowledge means that greater AI use is correlated with decrease rating positions, it additionally means that it is doable for absolutely AI-generated content material to rank in positions 1-3.
Totally or near-fully AI-generated pages are a small minority, however they definitely exist on this pattern: 9% of top-ranking pages are ≥80% AI content material, and 5.3% are 100% AI-generated.
Right here’s how the info breaks out:
- 5.3% of top-ranking pages returned an AI content material degree of 100%.
- 9.0% of top-ranking pages returned an AI content material degree of ≥80%.
- 8.8% of top-ranking pages returned an AI content material degree of 50-80%.
- 27.5% of top-ranking pages returned an AI content material degree of 20-50%.
- 54.7% of top-ranking pages returned an AI content material degree of <20%.
However what about the remainder of the SERP?
Wanting on the distribution of AI-generated content material throughout the highest 10 positions, extremely AI-generated content material will get rarer as you progress greater within the SERP.
Place 1 has the best share of low-AI-content pages and the bottom share of very-high-AI-content pages:
However there’s a light gradient: the prevalence of AI-generated content material will get progressively rarer transferring from place 10 to 1, in comparatively small increments.
Each rating place, together with place 1, nonetheless accommodates a significant share of closely AI-generated pages: between 8-12% containing ≥80% AI content material.
| Place | Very Excessive (≥80%) | Excessive (50-80%) | Average (20-50%) | Low (20%) |
|---|---|---|---|---|
| 1 | 8.4% | 8.1% | 27.5% | 55.9% |
| 2 | 9.0% | 8.8% | 27.6% | 54.6% |
| 3 | 9.6% | 9.5% | 27.4% | 53.5% |
| 4 | 9.8% | 10.0% | 26.9% | 53.3% |
| 5 | 10.0% | 10.1% | 26.9% | 53.0% |
| 6 | 10.7% | 10.2% | 26.5% | 52.6% |
| 7 | 10.2% | 10.1% | 27.7% | 52.0% |
| 8 | 11.1% | 9.8% | 28.0% | 51.0% |
| 9 | 11.3% | 10.2% | 28.1% | 50.3% |
| 10 | 11.7% | 9.5% | 27.7% | 51.1% |
We additionally calculated the typical and median AI content material degree throughout a pattern of 15,000 pages for every place. It’s the same story: AI content material turns into extra frequent additional down the SERP, however nonetheless seems in greater positions:
| SERP Place | Common AI Content material Stage | Median AI Content material Stage |
|---|---|---|
| 1 | 27.1% | 17.1% |
| 2 | 28.1% | 17.7% |
| 3 | 28.9% | 18.2% |
| 4 | 29.4% | 18.3% |
| 5 | 29.6% | 18.3% |
| 6 | 30.1% | 18.6% |
| 7 | 29.8% | 18.9% |
| 8 | 30.7% | 19.4% |
| 9 | 31.2% | 19.9% |
| 10 | 30.9% | 19.5% |
Analysing rating pages introduces a sort of choice bias: we’re solely evaluating profitable URLs that Google has already deemed match for indexing. However does AI use scale back the possibility of a URL being listed within the first place?
We appeared on the indexation charges for buckets of pages with completely different AI content material ranges. We sampled 1,000,000 pages from our crawler database, 1 web page per area. Round 100,000 pages contained sufficient content material for AI detection. We counted a web page as listed if it had any of the next traits:
- At the least one identifiable natural key phrase rating, or
- At the least one GSC impression since January 2026, or
- An actual URL hit for a Google web site: search.
If we bucket pages by their estimated AI content material degree, we will see that greater AI content material ranges correlate with decrease indexation charges:
Right here’s how the info breaks out by bucket:
- Pages with low AI content material degree had an indexation price of 49.28%
- Pages with reasonable AI content material degree had a mean indexation price of 43.38%
- Pages with excessive AI content material degree had a mean indexation price of 40.72%
- Pages with very excessive AI content material degree had a mean indexation price of 40.35%
Whereas the indexation price is decrease for content material with excessive or very excessive AI content material ranges, a full 40% of those pages had been nonetheless listed—not 1,000,000 miles away from the 49% indexation price of low AI content material.
Lastly, we wished to see how the natural efficiency of AI-generated pages trended over time relative to non-AI-generated pages.
To do that, we used GSC knowledge from reside URLs, matched these URLs with our crawler database, filtered out pages with too little textual content for AI detection, after which bucketed the URLs by their estimated AI content material degree. This left us with the next samples:
- Low AI content material degree: 28,643 pages
- Average AI content material degree: 23,498 pages
- Excessive AI content material degree: 12,911 pages
- Very Excessive AI content material degree: 15,809 pages
For every bucket, we used GSC knowledge to plot the change in impressions over time, beginning with one panel of internet sites from June 2025 to January 2026…
…and a second panel of internet sites from January 2026 to June 2026:
Natural efficiency is clearly influenced by many, many components, and there are few concrete conclusions we will draw from this restricted time window, however I believed the info was attention-grabbing for a couple of causes:
- Typically, AI content material use appears negatively correlated with impressions.
- Low and reasonable AI degree content material acquired 2–3x the impressions of excessive or very excessive AI-generated content material.
- Excessive and really excessive AI-generated content material each acquired comparable impressions throughout each samples.
- There aren’t any apparent precipitous drops in impressions for extremely AI-generated content material.
There are various the reason why AI-generated content material would possibly earn fewer impressions, so it could be incorrect to imagine that there’s any sort of automated suppression taking place. Websites that lean closely on AI content material might are typically newer or lower-authority within the first place; inversely, websites with good natural efficiency could also be disincentivized to publish AI content material.
Most certainly, for my part, is that growing use of AI correlates with reducing content material high quality. I feel that is the important thing to understanding how Google treats AI content material.
All through this analysis, AI use appears to be correlated with decrease efficiency in search: decrease impressions, decrease rating place, and decrease indexation price.
However this doesn’t inform your entire image. AI-generated content material is present in vital quantities in each rating place, together with the highest 1 to three. A big quantity of AI-generated content material is efficiently listed. Impressions to AI-generated content material remained comparatively steady over time.
There aren’t any apparent onerous cutoffs suggesting a binary AI classifier is stopping AI-generated pages from rating extremely or making it into the index. As an alternative, we see a gradual gradient: efficiency usually worsens with growing AI use… in all probability as a result of content material high quality usually worsens with growing AI use.
As I’ve mentioned earlier than, when corporations interact in an AI content material technique, they’re typically partaking in a method that could be very completely different from conventional content material creation and website positioning—completely different and worse in some ways. Primary AI content material typically:
- Repeats frequent data with out including any new info to the SERP.
- Fails to incorporate inner and exterior hyperlinks, photographs, visible curiosity, and first-person experiences.
- Depends on tedious academic-style writing.
- Incorporates apparent errors and inaccuracies.
These are all issues that would markedly worsen the natural efficiency of any piece of content material, however they’re particularly frequent in most AI-generated content material. As Dan Taylor defined in his nice SEJ article just lately, publishing this content material at scale is the proper recipe in your very personal “Mount AI” scenario, as Google “burst-crawls” new low-quality, AI-generated pages however dumps them from the index quickly after:
Google would possibly initially burst-crawl the brand new setup out of curiosity. However, if the area lacks the baseline authority to maintain that scale, Google will throttle its useful resource allocation. Simply because Google provides you the assets to index your pages initially, it doesn’t imply it’ll grant them to you indefinitely.
I don’t suppose Google is making an attempt to punish AI-generated content material; I feel it’s counting on the identical previous hallmarks of content material high quality that it at all times has. It’s simply that AI-generated content material is normally decrease high quality than human-generated content material.
Crucially, these high quality issues generally co-occur with AI use, however they don’t seem to be assured by AI use—as proven by the not-insignificant variety of listed and top-ranking AI-generated pages. I do know from firsthand expertise that it is vitally doable to create detailed, attention-grabbing, well-researched content material utilizing AI technology—the caveat being that most individuals don’t. But.
I don’t consider entrepreneurs and SEOs ought to be frightened of AI content material creation—however they need to be frightened of spam and scaled content material.
