An audit of 23 AI-generated wiki sites found invented redeem codes on 18 of them

The canonical declaration sat in one site's layout. Every page that site served inherited it, and it named a .com address the author had planned to buy and never registered, so DNS returned NXDOMAIN ^[7]. Every page was telling Google that the real copy of itself lived at an address that does not resolve. The author called that "a self-destruction switch" ^[8].

Three other sites had no canonical tag at all. An earlier fix script had injected metadata into client components and broken their builds, and the failures were silent for weeks while the team assumed Vercel was just slow ^[9]. Three sites served stale content for days after a regex-based cleanup script mangled JSX, and the static validation script in the pipeline could not catch what only a build would catch ^[10]. The remediation gates on the build before every push ^[13].

The sitemaps leaked too. 64 URLs listed in them redirected, because the sitemap generator and the server config disagreed about trailing slashes, and each entry spent a crawl request on a 308 ^[11]. Pages marked noindex were listed as well. The author's description of that envelope: "Please index me" and "don't index me" in the same envelope ^[12].

Search Console's "crawled, not indexed" bucket held 16 to 19 pages through late August, went to 107 on August 29, then stopped at 116 for three weeks ^[4]. That is 88 pages in one day ^[3], and no 88 sites launched that day. The author reads the jump as Google finishing its sample of the network and applying one verdict across the whole of it ^[6]. 116 is 38 percent of the 302 known URLs ^[2]. The sites Google never sampled stayed in "discovered, never crawled" with January 1, 1970 as the crawl date, the epoch default ^[5].

The 78 percent in the post's subtitle ^[17] is a site count. The author checked code-list pages against public code trackers and found fabricated code tables on 18 of 23 sites, roughly 63 invented code strings ^[15]; 18 of 23 is 78.3 percent ^[1]. Across the affected sites that is about three and a half invented codes each ^[4]. The patterns repeat: developer names used as codes, game title prefixes used as codes, "SKYASSASSIN" for a game titled [SKY ASSASSIN] Jujutsu Shenanigans ^[16]. The title's other loose figure is its timescale, which says 45 days where the window given, August 12 to September 10, is 30 ^[5].

The post-mortem cannot assign the Google outcome to the fabrication. The NXDOMAIN canonical, the three broken builds and the 64 redirecting sitemap URLs were live in the same window as the invented tables ^{[7][9][11][15]}. Bing indexed at least seven of the sites from the same pages and the same content ^[3], which the author takes as ruling out unindexable content; the post does not identify which seven, so there is no checking whether they were the sites whose canonicals resolved. The measurement that would separate the layers is recovery after the cleanup, and the author reports it changed nothing immediately ^[13]. "Technical health is a threshold, not a ranking factor," he wrote ^[14].