Pew Research Center checked a sample of 10,000 English-language webpages in July 2026 and found that 10% of them showed significant signs of AI authorship. That is one real number from one real study, and it already tells you something important: whoever asks how much of the internet is AI written is really asking several different questions at once, and each one comes back with a different answer.
If you search for how much content is AI generated, you will run into numbers all over the map: 10%, 35%, 50%, 74%. None of them is wrong. They are just measuring different things. A broad snapshot of the whole web looks nothing like a snapshot of what got published last quarter, and a snapshot of Google's top results looks nothing like a snapshot of long posts on LinkedIn. This piece walks through what each study actually measured, so you can use the right number for the question you're actually asking.
How Much Content Is AI Generated Across the Whole Web
Start with the biggest, broadest measure: every kind of page, old and new, sitting on the open web right now. Pew Research Center pulled 490,000 English-language webpage texts from Common Crawl, spread across 49 crawls collected between January 2021 and July 2026. In the most recent snapshot, from July 2026, 10% of the 10,000 sampled pages showed significant signs of AI authorship.
That 10% is a stock number. It includes years of pages written by hand long before generative AI existed, alongside pages published last month. Pew also broke the number down by domain type: about one in ten .com pages showed signs of AI authorship, compared with 4.6% of .org pages and roughly 1% of .edu and .gov pages. Only about 10% to 15% of pages in a crawl sample carry a usable publication date, so any number that depends on knowing when a page was written is working from a smaller, dated slice of the full sample.
Here's the number that starts to explain the gap with the bigger headlines you've probably seen: among pages published after ChatGPT's November 2022 release, more than a third showed signs of AI authorship in that same July 2026 sample. Same study, same detection method, much higher number, because the denominator changed from "every page on the web" to "pages published since generative AI went mainstream."
What Newly Published Articles Look Like
That shift from the whole web to new output is exactly where the widely quoted "half the internet" figure comes from, and it deserves a more careful look than it usually gets.
Graphite, a content marketing company, ran a study called "AI Now Writes as Many Online Articles as Humans," published in May 2026. It pulled 55,400 randomly selected URLs from Common Crawl, kept only English-language articles and listicles with article schema markup, at least 100 words long, published between January 2020 and March 2026, then used three separate AI detectors to score each one. An article counted as primarily AI-generated when the combined AI and AI-assisted share of its text outweighed the human-written share.
The result: in Q1 2026, 49.9% of eligible articles in the sample were primarily AI-generated, against 50.1% human-written. Graphite's numbers show the share climbing fast after ChatGPT launched, reaching 49.6% in Q1 2025 and peaking at 50.9% in Q4 2025, before settling back to roughly 50% and staying there. An earlier Graphite study, published in October 2025 with a smaller 43,000-URL sample and a single detection method, put the crossover point at November 2024, the month AI-generated articles first outnumbered human-written ones in that sample, at 39% after twelve months of tracking.
Read that carefully and it says something specific: about half of the eligible English-language articles Graphite sampled from recent publishing output were primarily AI-generated. It does not say half of everything online was written by AI, and it does not mean every one of those articles had zero human involvement. Many were AI-drafted and human-edited, which the primarily-AI classification still counts on the AI side once the AI share crosses 50%.
Why 35% and 74% Aren't Contradicting Each Other
Two more studies land at very different numbers, and it's worth seeing why before you pick one to quote.
An arXiv paper from April 2026, "The Impact of AI-Generated Text on the Internet," sampled publicly accessible websites from the Internet Archive across 33 monthly intervals between August 2022 and May 2025. It found that by the first half of 2025, roughly 35% of newly published websites were classified as AI-generated or AI-assisted. That figure counts whole websites, not individual articles, and it folds AI-assisted content in with fully AI-generated content, which is one reason it sits close to Pew's 35%-plus figure for post-ChatGPT pages rather than Graphite's roughly 50% article number.
Ahrefs crawled 900,000 webpages it identified as newly created in April 2025 and reported that 74.2% of them contained AI-generated content. That's the highest number in this whole set, and the reason is right there in the wording: "contained AI-generated content" is a much lower bar than "was primarily written by AI." A page with one AI-written paragraph and the rest written by a person still counts toward that 74.2%.
So the honest way to hold these together is this: a study asking "did AI touch this page at all" will always land higher than a study asking "was AI responsible for most of this page's words," and a study of whole websites will score differently than a study of individual articles. None of these numbers is more correct than the others. They're answering different versions of the percentage of AI generated content online question, and mixing them up in one sentence is where most of the confusing headlines come from. Any single piece of ai content prevalence data only means something once you know which of these questions it was actually answering.
The same caution applies to how each study reached its number in the first place. Every figure here comes from an automated detector or a probabilistic classifier scoring text for likely AI involvement, not a confirmed record of who wrote what. That's why the careful phrasing in these studies is "showed signs of," "detected as," or "classified as," rather than "was proven to be." Treat any single ai content prevalence data point as an estimate with a margin of error, not a fact carved in stone.
How Much AI Content Shows Up in Google's Results
Originality.AI has been tracking a narrower slice since 2019: the text on the top 20 Google results for a fixed set of 500 informational keywords. In September 2025, 17.31% of those sampled results were detected as AI content, up from 2.27% back in February 2019. The number has bounced between roughly 16% and 20% through most of 2025, touching a high of 19.56% in July.
This is a useful number if what you actually care about is what shows up on page one of Google, but it isn't a stand-in for the wider web. It only covers 500 keyword searches, only the top 20 results per keyword, and it says nothing about YouTube results, Reddit threads, or anything outside a standard organic listing.
What Percentage of Social Content Is AI-Generated
Social platforms tell a different story again, and the story changes a lot depending on how long the post is. Pangram scanned more than a million opt-in posts across LinkedIn, Medium, Substack, X, and Reddit between April and June 2026, limited to posts over 50 words, and found an overall average AI rate of 13.8%. Once you filter down to long-form content, over 250 words, that average jumps to 25.72% fully AI-generated.
The platform breakdown is where it gets interesting for anyone thinking about their own content. Among long-form posts, 41% on LinkedIn were fully AI-generated, compared with 31% on Medium, 29% on X, 13% on Reddit, and 10% on Substack. Reddit stood out as the most human of the group: in a sample of 266,678 Reddit replies, just 1.7% were flagged as AI, against 98.1% human. So a single "percentage of social media" number hides more than it reveals. Format and platform both matter more than any single average does.
What These Numbers Still Can't Tell You
None of the studies here measure whether AI-written content is any good. If you're hunting for ai slop statistics, meaning some measure of low-quality, repetitive, or genuinely unhelpful AI content, you won't find one in this research. What these studies measure is AI involvement, detected or classified, not usefulness, accuracy, or reader value. A page can be entirely AI-written and genuinely helpful, or entirely human-written and terrible, and none of the percentages above would tell you which, so treat any ai slop statistics you see floating around social media with the same caution.
They also don't cover images, video, audio, or synthetic media of any kind. Every figure in this piece comes from text analysis on webpages or text-heavy social posts. And most of the strongest studies are limited to English-language, publicly accessible content, so paywalled pages, other languages, and private platforms are underrepresented or missing entirely.
What This Means for Your Own Content
The honest takeaway is that "AI-generated" has already become the normal state for a big share of new publishing, somewhere close to half of new English-language articles by the best current estimate, while the broader stock of existing web pages still looks mostly human-written. Neither fact should change what you actually do day to day: write things that are accurate, useful, and clearly grounded in real expertise, whether AI helped draft them or not.
What's worth tracking is not the industry-wide percentage but your own visibility inside it. If AI tools are increasingly the thing people ask instead of searching, the real question for your team isn't how much of the internet is AI written, it's whether your brand shows up when someone asks an AI engine about your category. DeepSmith's AI Search Visibility module tracks mention and citation rates across the major AI engines on a schedule you set, so you can see whether you're actually showing up where your buyers are asking, rather than guessing from an industry-wide statistic that was never about your brand in the first place.
