“Should we stop using AI for our content? I heard Google is cracking down.” It is one of the most common worries business owners raise about content right now. So what is generative AI actually doing to search rankings? Nothing, if the content is good. Everything, if the content is lazy.
Short answer: no. Google is not cracking down on AI content. It never has. What it is cracking down on, and has been for years, is content that says nothing new, helps no one and exists only to fill a page. AI just made it absurdly easy to produce that kind of content at industrial scale. That is where the confusion lives.
The data backs that up clearly, and we will walk through it below. This article covers the research, explains the indexation gap, and shows the production pipeline we use at 21 Webs that keeps our content ranking.
Key Takeaways
- Google’s stated position is that it rewards quality regardless of production method. Its core updates and spam policies target low-quality content at scale, not AI authorship.
- In Ahrefs’ 2025 study of 600,000 top-ranking pages, 86.5 per cent contained some AI-generated content, 4.6 per cent were fully AI-generated and 81.9 per cent were hybrid.
- In Ahrefs’ 2026 study of 331,000 pages, 82.2 per cent of pages in the top three positions had under 50 per cent AI content. The top of the rankings skews towards human-heavy or lightly AI-assisted content.
- In the same 2026 study, indexation dropped from 49.3 per cent for low-AI pages to 40.4 per cent for very high-AI pages: a gap, but not a ban.
- The content that fails is interchangeable, has no firsthand experience, and adds nothing original. That describes most AI-only content, and plenty of human-only content too.
- The pipeline that works: AI drafts, humans edit, verify, add experience and publish.
Google Rewards Quality Regardless of Method. Its Updates Punish Thin Content at Scale.
Google has been explicit about this. Multiple times.
In February 2023, Google updated its guidance to state that AI-generated content is not against its guidelines, provided it is helpful, reliable and people-first. In its own words, the focus is on the quality of content, not how it is produced.
Every core update since then has reinforced the same principle:
The March 2024 core update came with a widened spam policy on “scaled content abuse”: content produced in bulk to manipulate rankings, “whether automation, humans or a combination are involved.” Google expected the changes to cut low-quality, unoriginal content in search results by 40 per cent, and later reported a 45 per cent reduction. Sites built on thousands of thin pages took the hit.
The helpful content system became part of core ranking in March 2024. There has been no separate “helpful content update” since. The same assessment of usefulness now runs inside every core update, including those Google rolled out in March 2025, June 2025, December 2025, March 2026 and May 2026. The signal: Google measures usefulness, not authorship.
The 2026 updates continued the pattern. Google has never announced an update that targets AI content as a category. Its updates target content that fails its quality standards, regardless of who or what produced it.
The distinction matters because it changes what you should be worried about. If you are producing AI-assisted content that is edited, verified, enriched with firsthand experience and genuinely useful, you have nothing to fear from any Google update. If you are publishing raw ChatGPT output at scale with no editing, you have everything to fear, not because it is AI, but because it is boring.
There is a reason the panic persists despite Google stating its position clearly and repeatedly. “Scaled content abuse” and “AI content” overlap so heavily in practice that the two look identical from the outside. When a site with 4,000 AI-generated pages gets wiped out, the visible story is “Google punished AI.” The actual story is “Google punished 4,000 thin pages,” and the AI part was a detail about how they were made. The update did not need to mention AI to end that site.
That is why the better question is not whether Google can tell. It is whether a stranger who read your page would learn something. Google’s systems are increasingly a proxy for that stranger.
We covered how AI search is reshaping SEO in a recent analysis. The content quality bar is the same bar, whether you are optimising for traditional search or AI-generated answers.
82.2 Per Cent of Top-Three Pages Have Under 50 Per Cent AI Content
Two Ahrefs studies sit behind the numbers in this article, and it is worth keeping them apart. The first, by Si Quan Ong and published in July 2025, scored 600,000 pages from the top 20 Google results for 100,000 keywords. The second, by Ryan Law and published on 27 July 2026, analysed 331,000 pages from Google’s top 10. Both used Ahrefs’ own AI content detector.
Here is how the top-ranking pages broke down by production method in the 2025 study:
- 13.5 per cent were purely human-written (zero AI detected)
- 81.9 per cent were a blend of AI and human content
- 4.6 per cent were classified as entirely AI-generated
Among the blended pages, AI usage varied:
- Minimal AI (1 to 10 per cent): 13.8 per cent of pages
- Moderate AI (11 to 40 per cent): 40 per cent of pages
- Substantial AI (41 to 70 per cent): 20.3 per cent of pages
- Dominant AI (71 to 99 per cent): 7.8 per cent of pages
The 2026 study’s headline finding: 82.2 per cent of pages ranking in the top three positions contained under 50 per cent AI content. The top of the rankings skews towards human-heavy or lightly AI-assisted content.
But the data also shows that 5.3 per cent of pages in positions one through three are 100 per cent AI-generated, and 9 per cent are 80 per cent or more AI. Fully AI-written pages can and do rank at the very top. They are a minority, but they are not excluded.
That 5.3 per cent is the number to keep in your head when someone tells you AI content cannot rank. It can. The honest reading of the study is not “use AI” or “avoid AI,” it is that the production method is not what is being measured. The blend is dominant because most good content is produced by a process that uses AI somewhere, not because a particular ratio earns a bonus.
The average AI content level rises only slightly across page one: from 27.1 per cent at position one to 30.9 per cent at position ten. The gradient exists, but it is gentle.
A gentle gradient is actually the most interesting result in the study, because it is the opposite of what a penalty looks like. If Google were penalising AI content, the top positions would be nearly AI-free and the percentage would climb sharply as you moved down the page. Instead it drifts by less than four points across ten positions. That is noise, not a signal, and it fits the near-zero correlation Ahrefs found in its 2025 study.
Ryan Law’s conclusion: Google penalises bad content, not AI content, though the two overlap often enough to cause real-world confusion.
The Indexation Gap: 49 Per Cent vs 40 Per Cent
Indexation is where the AI signal shows up most clearly.
The 2026 Ahrefs study measured indexation rates by AI content level:
- Low AI content (under 20 per cent): 49.28 per cent indexed
- Moderate AI content (20 to 50 per cent): 43.38 per cent indexed
- High AI content (50 to 80 per cent): 40.72 per cent indexed
- Very high AI content (80 per cent plus): 40.35 per cent indexed
The drop from 49 per cent to 40 per cent is meaningful. Pages with heavy AI content are about 18 per cent less likely to be indexed than pages with minimal AI content.
But 40 per cent of heavily AI pages are still indexed. That is not a penalty. That is not a ban. It is a gap that likely reflects the quality distribution: heavily AI content is more likely to be thin, undifferentiated and unhelpful, which are all signals that independently reduce indexation probability.
Look at the shape of that curve rather than the size of the drop. The fall from the low-AI bucket to the moderate bucket is real. But between the high-AI bucket and the very high-AI bucket, the rate barely moves at all, from 40.72 per cent to 40.35 per cent. If Google were scaling a penalty against AI content, that line would keep falling as AI percentage rose. Instead it flattens. A penalty has a slope. This has a floor.
That matters for how you act on the finding. The realistic risk is not that a well-written AI-assisted page gets filtered. It is that a raw, unedited page gets filtered, and the difference between those two is editing rather than authorship.
For context, a separate 2023 Ahrefs study of around 14 billion pages found that 96.55 per cent of them get no traffic from Google at all. The indexation gap for AI content is a rounding error compared to the broader quality problem on the open web.
The practical takeaway: if your AI-assisted content is well-edited, original and useful, it gets indexed. If it is raw output published without review, it might not. The same is true of poorly written human content.
86.5 Per Cent of Top Pages Use AI Somewhere in Their Workflow
This is the number that should end the “should I use AI for content” debate.
Ahrefs’ 2025 study found that 86.5 per cent of top-ranking pages contain some level of AI-generated content. Only 13.5 per cent were purely human-written.
AI is not a fringe practice in content production. It is the dominant workflow. The question is not whether to use AI. It is how to use it without producing the boring, interchangeable content that fails.
Read that against the earlier number and the strategy becomes obvious. If 86.5 per cent of top pages use AI somewhere, and only 4.6 per cent are fully AI-generated, then AI is almost always present and almost never alone. The winning configuration is not a question of whether, it is a question of what the human contributes alongside it.
The same 2025 study measured the correlation between AI content percentage and ranking position at 0.011, which Ahrefs described as effectively zero. Google’s algorithm appears indifferent to whether content is AI-made or human-made.
What the algorithm is not indifferent to:
- Content depth. Does the page answer the question comprehensively, or does it skim the surface?
- Originality. Does the page say something the reader has not already seen on ten other pages?
- Firsthand experience. Does the author or business have direct experience with the topic?
- User engagement. Do readers stay on the page, scroll through it and click further into the site?
AI can produce content that meets all four criteria, if a human edits it, adds experience, verifies claims and removes the generic filler that LLMs default to. AI cannot meet these criteria on its own, because it has no firsthand experience and it defaults to the statistical average of everything it was trained on.
That last phrase explains the whole problem. A language model produces the most likely version of a sentence. If you ask about a topic a hundred businesses write about, you get the answer a hundred businesses gave, because that is what “most likely” means. It is fluent, it is correct, and it is indistinguishable from the competition. Rank is a comparative measure, so a page that is identical to its competitors has no reason to be chosen over them.
“Firsthand experience” is the criterion you can actually win on. You have customers, jobs, quotes, mistakes and outcomes that no model has access to. That is genuine, defensible content, and it is the part you should be protecting most carefully. It is also the only input in the whole process that a competitor cannot simply copy, because they do not have your customers, your jobs or your history.
It is also the cheapest improvement available to most businesses, and the most commonly skipped. Adding firsthand detail does not require a bigger budget or a new tool. It requires the person who knows the work to spend twenty minutes putting something real into the draft, which is a scheduling problem rather than a content problem.
What "Boring" Actually Means in Google's Algorithm
Google does not use the word “boring.” But its quality signals measure exactly that.
Here is what Google’s guidance on helpful, people-first content asks you to check, translated into plain language:
Does the content provide substantial value beyond what other pages offer? If your AI-generated blog post says the same thing as the top ten existing results, it adds nothing. Google has no reason to rank it.
Does the content demonstrate firsthand experience or expertise? A page about “best accounting software for Australian small business” written by someone who has never used accounting software fails this test, whether written by a human or an AI.
Was the content created primarily for search engines rather than people? A 2,000-word article stuffed with keywords but offering no insight beyond what the SERP snippet already provides is search-engine-first content. Google deprioritises it.
Does the site have a clear purpose and identity? A site that publishes 500 AI-generated articles across 50 unrelated topics looks like a content farm. Google treats it as one.
The pattern in every core update since 2023 is the same: sites that published high-volume, low-quality content, whether AI or human, lost rankings. Sites that published lower-volume, high-quality content, whether AI-assisted or fully human, held or gained.
The word that captures what Google penalises is not “AI.” It is “interchangeable.” If your content could have been written by anyone (or anything) and says nothing that is not already on page one, it will not rank regardless of how it was produced.
The four tests above are worth running on your own pages, and the fourth one is the most revealing. Swap your own domain out of the question and ask whether the page could sit on a competitor’s site without anyone noticing. If it could, the problem is not your production method, it is that you have not put anything of your own into it.
This is also why “more content” is a losing answer to weak rankings. Volume multiplies whatever your pages already are. If they are original, volume compounds an advantage. If they are interchangeable, volume just gives Google more examples of the same problem.
AI Answers Recommend AI-Assisted, Human-Verified Content
Here is an irony worth noting.
Ask ChatGPT, Claude or Gemini about the best approach to AI content creation and the answer you will usually get is the same: use AI to draft, then have a human edit, verify facts, add firsthand experience and ensure originality.
The AI tools themselves do not recommend publishing raw AI output. They recommend the hybrid workflow that the Ahrefs data shows is already dominant among top-ranking pages.
There is a straightforward explanation for that consensus, and it is not modesty. These systems are trained on a large body of expert writing about content and SEO, and that body of writing has converged on the same conclusion for the same reason the search data has. The advice is the average of what people who do this well have said, and what they have said, over and over, is that the human part is where the value is.
The consistency of that answer across every tool is itself informative.
This matters because AI-generated search answers (Google AI Overviews, ChatGPT’s browsing mode, Perplexity) cite the same quality signals that traditional search uses. A page that is well-structured, authoritative and original gets cited by AI answers. A page that is generic, surface-level and interchangeable does not.
For Australian businesses investing in AI SEO, the implication is clear: the same content quality that ranks in traditional search is the same content quality that gets cited in AI-generated answers. There is no shortcut.
We also analysed how AI traffic converts differently and what that means for conversion rate optimisation. The traffic you earn through quality content, whether from traditional search or AI citation, converts. The traffic you try to game does not.
The Human-Edited AI Pipeline That Works
At 21 Webs, we use AI in our content production. We also edit everything a human would edit, and we add what AI cannot.
Here is the pipeline:
Step 1: Research and brief (human) A human identifies the target keyword, analyses search intent, reviews competing content and writes a content brief that defines what the page needs to say and what original angle it brings.
Step 2: Draft (AI-assisted) AI generates a first draft based on the brief. This draft is a starting point, not a finished product. It covers the structure, the subtopics and the general argument.
Step 3: Edit and enrich (human) A human editor rewrites for voice, adds firsthand experience, verifies every statistic, removes generic filler, inserts original analysis and ensures the page says something that no other page on the topic says. This is where the value is created.
Step 4: Technical SEO (human and tools) Schema markup, internal linking, meta descriptions, alt text and structured headings are added or refined. This step determines whether the page is machine-readable and citable.
Step 5: Review and publish (human) A final review for accuracy, tone, compliance and brand voice. The page publishes only after a human signs off.
The AI saves time on Step 2. The human creates all the value in Steps 1, 3, 4 and 5. The result is content that passes every quality test, ranks in traditional search, gets cited by AI answers and reflects the business’s actual expertise.
Step 1 is the one most people skip, and it is the one that decides the outcome. A model given a vague instruction produces a vague page, and no amount of editing later recovers an angle that was never chosen. The brief is where you decide what the page argues. Everything downstream is execution.
Step 3 is where the hours actually go, and it is worth being honest about the trade. People assume AI cuts content production time dramatically. It cuts the drafting time dramatically. What it does not do is remove the editing, because a draft that has not been edited is not a finished page. What AI changes is the arithmetic: you spend less time typing and more time deciding, which is a better use of the person who knows the subject.
If you skip Steps 1, 3, 4 and 5 and publish Step 2 raw, you produce the boring, interchangeable content that Google deprioritises. Not because it is AI, but because it is not good enough.
Important FAQs
Is AI content penalised by Google?
Can I publish AI content and rank?
What is the difference between AI content and AI slop?
Should I tell Google my content is AI-written?
The Bottom Line
The data is clear and it has been clear for two years.
Google does not penalise AI content. It penalises boring content. The two overlap often enough to create confusion, but they are not the same thing.
In Ahrefs’ 2025 study, 86.5 per cent of top-ranking pages used AI somewhere in their workflow. In its 2026 study, 82.2 per cent of top-three pages had under 50 per cent AI content. The sweet spot is AI-assisted, human-edited content that adds original insight, firsthand experience and genuine value.
If you are avoiding AI because you think Google will penalise you, you are solving the wrong problem. If you are publishing raw AI output at scale because you think volume wins, you are creating the wrong content.
Want AI-assisted content that is still worth reading?
The businesses that rank in 2026 use AI to move faster and humans to make the work worth reading. That is how we run AI SEO for our clients. Talk to our team about yours.
Sources
[1] Ahrefs, “Google Doesn’t Punish AI Content; It Punishes Bad Content (331k Pages Studied)”, Ryan Law, 27 July 2026. https://ahrefs.com/blog/google-doesnt-punish-ai-content/
[2] Ahrefs, “AI-Generated Content Does Not Hurt Your Google Rankings (600,000 Pages Analyzed)”, Si Quan Ong, 7 July 2025. 600,000 pages from the top 20 results for 100,000 keywords. https://ahrefs.com/blog/ai-generated-content-does-not-hurt-your-google-rankings/
[3] Ahrefs, search traffic study, Tim Soulo, 1 December 2023. Around 14 billion pages. https://ahrefs.com/blog/search-traffic-study/
[4] Google Search Central, “Google Search’s guidance about AI-generated content”, February 2023. https://developers.google.com/search/blog/2023/02/google-search-and-ai-content
[5] Google, Elizabeth Tucker, March 2024 Search update on low-quality content and new spam policies, 5 March 2024. https://blog.google/products/search/google-search-update-march-2024/
[6] Google Search Central, helpful content system (“In March 2024, it evolved and became part of our core ranking systems”). https://developers.google.com/search/updates/helpful-content-update
[7] Google Search Status Dashboard, ranking update history 2024 to 2026. https://status.search.google.com/products/rGHU1u87FJnkP6W2GwMi/history
