# 皮尤研究：ChatGPT 发布后超三分之一新网页疑似 AI 生成

- 来源：TechCrunch：AI（RSS）
- 作者：Sarah Perez
- 发布时间：2026-08-21 01:18
- AIHOT 分数：50
- AIHOT 链接：https://aihot.virxact.com/items/cmt1sj0k606j3roovbyba61j7
- 原文链接：https://techcrunch.com/2026/08/20/a-third-of-webpages-published-since-chatgpts-launch-show-signs-of-ai-authorship-study-finds

## AI 摘要

皮尤研究中心新研究发现，ChatGPT 发布后发布的网页中，超过三分之一（35%）显示出明显的 AI 创作或编辑痕迹。该结论基于对 Common Crawl 中近 50 万英文网页的分析，其中 2026 年 7 月随机抽样的 1 万网页中约 10% 有显著 AI 痕迹。.com 域名的 AI 创作比例约为 .edu 或 .gov 域名的 10 倍。

## 正文

Over one-third of webpages published after the release of ChatGPT show signs of being written by AI, according to a new study from Pew Research released on Thursday. The report corroborates other studies that detail how much of the web’s newer webpages are now either written by or “substantially edited” by AI, the firm says.

It also arrives shortly after internet infrastructure provider Cloudflare reported that bot web traffic had overtaken human web traffic — a milestone that was reached sooner than the company had estimated.

Pew’s data, however, is focused not on who or what is browsing the web, but on what is being browsed. And apparently, much of it is bots reading webpages written by other bots.

To compile the report, Pew said it used the Common Crawl web archive to collect nearly half a million English-language webpages from the past five or so years, starting a couple of years before ChatGPT’s November 2022 release. Pew then used Open Pangram’s technology to detect how many were likely written or heavily edited by AI.

In a random sample of 10,000 webpages collected in July 2026, around 10% showed “significant signs of AI authorship,” Pew said.

However, Pew pointed out that a random sample like this would inevitably include older webpages — ones published before AI writing tools even existed, and therefore couldn’t have been AI-written.

% of webpages showing significant AI editing or authorship Image Credits:Pew Research

To get a better sense of how much of the web is now being written by AI, Pew filtered out the older webpages and focused only on those published after the release of ChatGPT.

In this same snapshot with the older pages removed, signs of AI authorship were found in over one-third (35%) of the pages.

ScreenshotImage Credits:Pew Research

Drilling down into domains themselves, Pew found that URLs with a .com domain showed signs of AI authorship at around 10 times the rate of a .edu or .gov domain (both of which were around 1% AI-authored. In addition, .org domains had only a 4.6% rate of AI authorship.

Of course, this analysis is not perfect. Pangram, like other AI detection tools, can misclassify pages as AI-written when they weren’t. But at scale, the data is likely at least directionally correct.

In addition, Pew found that other supposed tells of AI authorship had also increased over the years, like the use of em dashes, Oxford commas, and phrasing like “it’s not X, it’s Y,” among other things.