I’m not sure that your test is conclusive. It would be if the only way that OpenAI indexed new websites was to find them linked from other places, but I think they have other methods as well.
Basically Cloudflare put up a website with a robots.txt file that banned Perplexity’s crawler, and Perplexity crawled it anyway. The article was focussed on the rudeness of the crawler, but there is something else interesting here too. About their test domains they said, “These domains were newly purchased and had not yet been indexed by any search engine nor made publicly accessible in any discoverable way.” I think that this would mean they didn’t do anything to have other websites link to them.
One way that you could find websites that aren’t linked to by anyone would be to use something like zmap, which can scan the whole ipv4 address space in about 45 minutes with a good internet connection. You would tell it to send packets to ports commonly used by https servers, and then after a scan you would have a list of all homepages, whether or not they are linked to from anyone. In fact, you’d even get ones that don’t have a domain name, but are just IP addresses. It is harder to scan ipv6 because it’s so much bigger, but most things are still ipv4 I believe, and I think there are still supposed to be ways to do it. I think there are blocks of IP addresses that are more likely to be used. There is a search engine called Shodan that let’s you search for servers, and they do scans like this. Scanning in this fashion might be illegal in some places, but I think it’s legal in America, where Shodan and Perplexity and OpenAI all are.
So that’s one way you could do it. There might be another way too though. I own a couple domains that I haven’t done anything with yet, and I used Ahrefs’ backlink checker thing on them, and even though I hadn’t done anything they were actually linked to by some websites that were like “List of newly registered domains for this month”. I don’t think these people found my domains by scanning, because I didn’t have servers running so a scan shouldn’t have picked them up unless Cloudflare had some response, but I don’t think they did. They may have gotten the information somehow from the registrar, or maybe from a higher up like ICANN. It’s public information what is registered, and they might have gotten a list somehow.
Unlinked domains can definitely be found in a lot of ways, but like I show in the article there was literally no fetching of the page except for Googlebot. So even if the hostname was leaked somehow the contents of the page require fetching the page, which was only done by Google. Also like I show in the article the content that ChatGPT knows identically matches what's in a Google search snippet, down to where a word-break is.
Oh, yes you’re right. I didn’t read the whole thing, which was clearly a mistake. I apologize. Good job with the article, and thank you. It’s pretty funny that they only get the excerpts.
I’m not sure that your test is conclusive. It would be if the only way that OpenAI indexed new websites was to find them linked from other places, but I think they have other methods as well.
This article from Cloudflare is about the behavior of Perplexity’s crawler: https://blog.cloudflare.com/perplexity-is-using-stealth-unde...
Basically Cloudflare put up a website with a robots.txt file that banned Perplexity’s crawler, and Perplexity crawled it anyway. The article was focussed on the rudeness of the crawler, but there is something else interesting here too. About their test domains they said, “These domains were newly purchased and had not yet been indexed by any search engine nor made publicly accessible in any discoverable way.” I think that this would mean they didn’t do anything to have other websites link to them.
One way that you could find websites that aren’t linked to by anyone would be to use something like zmap, which can scan the whole ipv4 address space in about 45 minutes with a good internet connection. You would tell it to send packets to ports commonly used by https servers, and then after a scan you would have a list of all homepages, whether or not they are linked to from anyone. In fact, you’d even get ones that don’t have a domain name, but are just IP addresses. It is harder to scan ipv6 because it’s so much bigger, but most things are still ipv4 I believe, and I think there are still supposed to be ways to do it. I think there are blocks of IP addresses that are more likely to be used. There is a search engine called Shodan that let’s you search for servers, and they do scans like this. Scanning in this fashion might be illegal in some places, but I think it’s legal in America, where Shodan and Perplexity and OpenAI all are.
So that’s one way you could do it. There might be another way too though. I own a couple domains that I haven’t done anything with yet, and I used Ahrefs’ backlink checker thing on them, and even though I hadn’t done anything they were actually linked to by some websites that were like “List of newly registered domains for this month”. I don’t think these people found my domains by scanning, because I didn’t have servers running so a scan shouldn’t have picked them up unless Cloudflare had some response, but I don’t think they did. They may have gotten the information somehow from the registrar, or maybe from a higher up like ICANN. It’s public information what is registered, and they might have gotten a list somehow.
Unlinked domains can definitely be found in a lot of ways, but like I show in the article there was literally no fetching of the page except for Googlebot. So even if the hostname was leaked somehow the contents of the page require fetching the page, which was only done by Google. Also like I show in the article the content that ChatGPT knows identically matches what's in a Google search snippet, down to where a word-break is.
Oh, yes you’re right. I didn’t read the whole thing, which was clearly a mistake. I apologize. Good job with the article, and thank you. It’s pretty funny that they only get the excerpts.
No worries & thanks! Yea I didn't expect to get them with the excerpts too, that was a surprise bonus.