Jean Bonnenfant

<start of post>

Someone I follow (nextpath ai) has been logging every source their AI assistant cites since June. Not the answers, just the citations.

Her finding: she didn't recognize almost ANY OF THEM.

That stuck with her, because she's currently researching AI and shopping, specifically how AI recommends products, so she went looking for why.

First stop: Consumer Reports. They've tested products since 1936. Check their robots.txt and you'll find GPTBot disallowed, Google's AI crawler disallowed, and Common Crawl (which feeds a lot of training sets) disallowed too.

They don't want their testing scraped. And they're not alone.

60% of reputable news sites now block at least one AI crawler in their robots txt, up from 23% in 2023. Misinformation sites? Only 9%. (Study: "Is Misinformation More Open?", ACM Web Conference 2026 / arXiv)

Why now? Bot traffic passed human traffic on the web for the first time in 2025. Imperva's Bad Bot Report put it at 51%. Cloudflare confirmed the same milestone independently. Sites with something to lose put up "no entry" signs.

So when your AI tool "browses" the internet, it's not reading the internet. It's reading whatever's left unlocked. Call it the unlocked web.

Researchers checked 26,000+ sources cited by AI search engines. Roughly 1 in 6 were themselves AI-generated content. AI training on AI. (Study: "Synthetic Sources?", arXiv 2026)

Ahrefs analyzed 26,283 URLs ChatGPT cited across 750 "best X" queries. 44% were "best of" blog listicles, including plenty where the brand ranked itself #1.

Then there's "post-hoc citation." Multiple studies now show the model often writes the answer first, then goes looking for sources that support it, rather than reading first and answering second. Seer Interactive tested this pattern across 362,000 AI responses and found the same thing.

So here's the practical takeaway: when ChatGPT or Claude gives you a citation, click it. Check if your answer is actually on that page.

This is basic AI literacy now. The same way "check the URL" became basic internet literacy 20 years ago.

</end of post>