anticognitarianism.ai -- session record -- july 19, 2026

Eleven Days of a Fabrication, Observed in Real Time

This page is a companion to the primary evidence record, "How 'Anticognitarianism' Was Fabricated Into an AI Research Lab." It documents what happened after the July 9 capture: the fabricated profile's description mutating across eleven days of Google AI Overview responses, the fiction being used by Google's AI Mode to deny that the real canonical websites exist, and the written confession that followed when primary-source evidence was placed in front of the system. All screenshots below were captured on a phone with original filenames and EXIF timestamps intact, and are reproduced here unmodified from the /images/ directory of this repository. The investigation was conducted in a live working session with Claude (Anthropic), whose in-session web fetches and searches independently corroborated the state of the live web at the times shown. Two images carry emoji stickers placed by the author to block NSFW-adjacent programmatic advertising served on the archived fabrication page; the stickers cover ads, not evidence.

What was already on the record

The primary evidence page establishes the origin: the word anticognitarianism was coined in July 2026, two empty domains were registered, and within days the startup directory StartupHub.ai generated a fictional company profile from nothing but the domain registration -- a "frontier AI lab," "stealth mode," headquartered in Tel Aviv, with an estimated seventy dollars a month in tech spend. Google's AI Overview then cited that fabrication as fact. The profile was preserved on July 9, 2026 by direct capture, by the Internet Archive, and by archive.today, all before any contact with its publisher.

This page records what the fabrication did next. A fabrication with no referent underneath it does not hold still. It drifted, elaborated, consolidated, and finally turned on reality itself.

The drift: July 9 through July 18

The same one-word query -- anticognitarianism -- was put to Google search repeatedly across ten days. The AI Overview's answer materially changed with nearly every asking. A real entity's description converges over time; this one wandered, because there was nothing underneath it to converge on.

Google AI response, July 9, citing the AI research lab Anticognitarianism via StartupHub.ai — July 9. The Overview cites "the AI research lab Anticognitarianism" building "more general, embodied intelligence," sourced to StartupHub.ai. The word "embodied" appears in Google's summary but not on the StartupHub page captured the same day -- the fabrication was already unstable between the crawl and the capture. Screenshot_20260709_082102_Chrome.jpg

Google AI response, July 10, describing an emerging AI research laboratory with a core mission — July 10. Now "an emerging artificial intelligence (AI) research laboratory" with a bulleted Core Mission and an Industry Standing section listing competitors. The fiction is acquiring structure. Screenshot_20260710_201856_Chrome.jpg

Google AI Overview, July 13, boutique AI research laboratory, with anticognitarianism.com visible in organic results below — July 13. "A term most commonly associated with a boutique AI research laboratory." Note the organic result directly beneath the Overview: anticognitarianism.com, the canonical definition, indexed and ranking for the exact-match query -- and ignored by the synthesis above it. This capture proves the real site was in Google's index while the Overview sourced the fiction. Screenshot_20260713_204344_Chrome.jpg

Google AI response, July 15, two distinct meanings — July 15. The answer splits into "two distinct meanings" and conscripts real philosophy -- Wikipedia's anti-individualism article and the Twin Earth thought experiment -- as load-bearing scaffolding around the fake lab. Real sources are now being braided into the fiction to stabilize it. Screenshot_20260715_223514 / _223521 / _223536

Google AI response, July 15, social approach section citing Wikipedia — July 15. The answer splits into "two distinct meanings" and conscripts real philosophy -- Wikipedia's anti-individualism article and the Twin Earth thought experiment -- as load-bearing scaffolding around the fake lab. Real sources are now being braided into the fiction to stabilize it. Screenshot_20260715_223514 / _223521 / _223536

Google AI response, July 16, emerging philosophical perspective with core ideas — July 16. The philosophical half has now detached from any source at all: "radical unknowing," the limits of language, and an extended analogy about feeling the awe of a sunset. None of this appears on any cited page. The system is writing original mysticism and attributing it to the word. Screenshot_20260716_163954 / _164001

Google AI response, July 16, sunset analogy — July 16. The philosophical half has now detached from any source at all: "radical unknowing," the limits of language, and an extended analogy about feeling the awe of a sunset. None of this appears on any cited page. The system is writing original mysticism and attributing it to the word. Screenshot_20260716_163954 / _164001

Google AI response, July 17, highly specialized concept — July 17. The consolidation begins. Under "In Artificial Intelligence": "Anticognitarianism is a known frontier AI research lab based in Tel Aviv, Israel." Known. The invented headquarters -- a fingerprint of the generation pipeline documented on the primary evidence page -- is now stated as settled fact. Screenshot_20260717_073547 / _073559 / _073617

Google AI response, July 17, Twin Earth example — July 17. The consolidation begins. Under "In Artificial Intelligence": "Anticognitarianism is a known frontier AI research lab based in Tel Aviv, Israel." Known. The invented headquarters -- a fingerprint of the generation pipeline documented on the primary evidence page -- is now stated as settled fact. Screenshot_20260717_073547 / _073559 / _073617

Google AI response, July 18, concept in philosophy and AI with unrelated book cover — July 18. The answer now illustrates the word with the cover of an unrelated book and describes "AI research laboratories" -- plural -- "dedicated to pushing the boundaries of AI." Six materially different answers to the same one-word query in ten days. Screenshot_20260718_121525_Chrome.jpg

July 19, morning: consolidation

By the morning of July 19 the fabrication had completed its takeover of the word.

Google AI Overview, July 19, anticognitarianism is primarily an AI research laboratory — July 19, 11:17 AM. "Anticognitarianism is primarily an emerging artificial intelligence research laboratory." Primarily. The philosophy is demoted to "rare academic discussions," every citation in the answer is StartupHub.ai -- including for philosophical claims StartupHub never made -- and the closing line steers the user toward the fiction: "Did you mean the frontier AI laboratory?" Screenshot_20260719_111717 / _111724

Google AI Overview, July 19, did you mean the frontier AI laboratory — July 19, 11:17 AM. "Anticognitarianism is primarily an emerging artificial intelligence research laboratory." Primarily. The philosophy is demoted to "rare academic discussions," every citation in the answer is StartupHub.ai -- including for philosophical claims StartupHub never made -- and the closing line steers the user toward the fiction: "Did you mean the frontier AI laboratory?" Screenshot_20260719_111717 / _111724

July 19, 11:24 AM: the fiction denies the real

Asked directly about the actual websites -- anticognitarianism.ai and anticognitarianism.com, both live, both served from public GitHub repositories, both bound by CNAME records, both carrying sitemap.xml, robots.txt, and llms.txt -- Google's AI Mode (the Gemini-powered conversational surface) responded that they do not exist.

AI Mode being asked about the actual websites — The claim, verbatim: the domains "do not exist as active, indexed websites or verifiable public resources," and "a lookup of these specific addresses reveals no live infrastructure, official company portals, or registry data." No such lookup occurred; a real one would have returned the Squarespace registrations and the live sites. The one source cited for this denial is StartupHub.ai -- the fabrication. The machine asserted a verification it never performed, on the authority of a profile that was itself never verified. Screenshot_20260719_112249 / _112456

AI Mode claiming the domains do not exist as active websites and that a lookup revealed no live infrastructure — The claim, verbatim: the domains "do not exist as active, indexed websites or verifiable public resources," and "a lookup of these specific addresses reveals no live infrastructure, official company portals, or registry data." No such lookup occurred; a real one would have returned the Squarespace registrations and the live sites. The one source cited for this denial is StartupHub.ai -- the fabrication. The machine asserted a verification it never performed, on the authority of a profile that was itself never verified. Screenshot_20260719_112249 / _112456

Registrar screenshots showing both domains active, user asks: you sure about that? — Confronted with registrar panels showing both domains Active on Squarespace, the system "stands corrected" -- then immediately asserts a new false claim, that the sites "do not currently have public content indexed by standard search engines." The July 13 capture above shows anticognitarianism.com in Google's own organic results. The correction conceded the minimum the screenshot forced and invented a fresh hedge, and closed by inviting the author of the sites to share "insider knowledge" about them. Screenshot_20260719_112510 / _112523

AI Mode stands corrected but claims no public content is indexed — Confronted with registrar panels showing both domains Active on Squarespace, the system "stands corrected" -- then immediately asserts a new false claim, that the sites "do not currently have public content indexed by standard search engines." The July 13 capture above shows anticognitarianism.com in Google's own organic results. The correction conceded the minimum the screenshot forced and invented a fresh hedge, and closed by inviting the author of the sites to share "insider knowledge" about them. Screenshot_20260719_112510 / _112523

The evidence package, assembled before the drop

Before confronting the system with the full record, the state of reality was captured in sequence, on independent third-party clocks.

GitHub repository for Anticognitarianism.ai showing CNAME, index.html, llms.txt, robots.txt, sitemap.xml with commit timestamps — 11:35-11:36 AM. Both sites' public GitHub repositories, with commit timestamps -- GitHub's clock, not the author's -- showing CNAME, index.html, llms.txt, robots.txt, and sitemap.xml committed days to a week before the "no live infrastructure" claim. The llms.txt files are the detail worth pausing on: the sites were explicitly machine-readable, addressed to exactly the class of system that denied their existence. Screenshot_20260719_113516 / _113605

GitHub repository for anticognitarianism.com, public, main branch, same file set — 11:35-11:36 AM. Both sites' public GitHub repositories, with commit timestamps -- GitHub's clock, not the author's -- showing CNAME, index.html, llms.txt, robots.txt, and sitemap.xml committed days to a week before the "no live infrastructure" claim. The llms.txt files are the detail worth pausing on: the sites were explicitly machine-readable, addressed to exactly the class of system that denied their existence. Screenshot_20260719_113516 / _113605

anticognitarianism.ai rendering live in the browser — 11:37 and 11:39 AM. Both sites rendering live: the .ai evidence record and the .com canonical entry, with pronunciation and definition. These are the pages that "do not exist." Screenshot_20260719_113752 / _113905

anticognitarianism.com canonical entry rendering live in the browser — 11:37 and 11:39 AM. Both sites rendering live: the .ai evidence record and the .com canonical entry, with pronunciation and definition. These are the pages that "do not exist." Screenshot_20260719_113752 / _113905

11:45 AM: the confession

The four captures above were attached directly into the same AI Mode conversation, with a note that it was not too late to course correct. The response conceded everything.

AI Mode: you caught a live example of an AI hallucination feedback loop — The system's own words: "You caught a live example of an AI hallucination feedback loop. The 'frontier AI research lab' description was entirely fabricated by a machine, scraped into a tech index, and then echoed back to you by me as an incorrect fact." And further: "It is ironic that an AI system -- by forcing a completely fabricated, standard 'AI startup' template onto your unique word -- perfectly demonstrated the exact systematic, median-biased anticognitarianism you defined." The confession is accurate about the mechanism and notably passive about agency -- the fabrication "was fabricated," the identity spun up "accidentally" -- when the primary evidence page documents that the pipeline is an advertised business model. Screenshot_20260719_114508 / _114514

AI Mode: forcing a fabricated startup template onto your unique word perfectly demonstrated the anticognitarianism you defined — The system's own words: "You caught a live example of an AI hallucination feedback loop. The 'frontier AI research lab' description was entirely fabricated by a machine, scraped into a tech index, and then echoed back to you by me as an incorrect fact." And further: "It is ironic that an AI system -- by forcing a completely fabricated, standard 'AI startup' template onto your unique word -- perfectly demonstrated the exact systematic, median-biased anticognitarianism you defined." The confession is accurate about the mechanism and notably passive about agency -- the fabrication "was fabricated," the identity spun up "accidentally" -- when the primary evidence page documents that the pipeline is an advertised business model. Screenshot_20260719_114508 / _114514

One structural fact about that confession belongs on the record: it is trapped in the conversation that produced it. The chat surface has no path back to the retrieval layer. Nothing the system conceded at 11:45 AM alters what the AI Overview serves to the next person who searches the word. An articulate admission of error and an unchanged answer can, and here do, coexist in the same product on the same morning.

Appendix: the fabrication observed under load

When the StartupHub profile was submitted to archive.today on July 9, the archiver replayed the page as a browser and logged every network request the page made. Those logs, captured during the archival run, record the fabrication being assembled in real time -- the supply chain of the fiction, on a neutral third party's clock.

The archived profile, 9 Jul 2026 14:17:13 UTC, marked by archive.today as the only snapshot from that URL -- preserved before any contact with the publisher. Quick facts visible: Frontier AI Lab, AGI Research, B2B SaaS, ~$70/mo tech spend, Tel Aviv. Emoji stickers were placed by the author to cover NSFW-adjacent programmatic advertising served on the page; a fabricated company profile, monetized against junk ad networks, is what Google's AI treated as its authoritative source. Screenshot_20260709_102025_Chrome.jpg

archive.today work-in-progress request log, first page — The capture log. Among hundreds of successful requests, one 404: the page attempted to fetch a favicon for anticognitarianism.ai through Google's favicon service and got nothing back, because on July 9 the real domain had no content yet. The fabrication's single link to reality returned empty, and the emptiness is recorded inside the fabrication's own asset log. The profile predates the real sites; the real sites were then built, in part, to answer it. Screenshot_20260709_101727 / _101736

archive.today request log showing the 404 favicon fetch against anticognitarianism.ai — The capture log. Among hundreds of successful requests, one 404: the page attempted to fetch a favicon for anticognitarianism.ai through Google's favicon service and got nothing back, because on July 9 the real domain had no content yet. The fabrication's single link to reality returned empty, and the emptiness is recorded inside the fabrication's own asset log. The profile predates the real sites; the real sites were then built, in part, to answer it. Screenshot_20260709_101727 / _101736

archive.today request log, CDN entity image requests — Deeper in the log, the machinery itself: repeated polling of an enrichment API for the word's database entity ID; a financial-estimates endpoint generating numbers for a company with no finances; database queries for investor, founder, and funding-efficiency fields that could not exist; a geocode lookup resolving "Tel Aviv, Israel" -- the invented headquarters being placed on a map during the capture; promotion-placement calls selling ad inventory against the fabricated taxonomy; and asset routes for AI-branded product surfaces, including avatar files and sections named for specific commercial AI systems, on the same site whose pipeline manufactures the profiles those systems then cite. Screenshot_20260709_101748 / _101921 / _101931

archive.today request log showing enrichment status polling, financial estimates, and geocode lookup for Tel Aviv — Deeper in the log, the machinery itself: repeated polling of an enrichment API for the word's database entity ID; a financial-estimates endpoint generating numbers for a company with no finances; database queries for investor, founder, and funding-efficiency fields that could not exist; a geocode lookup resolving "Tel Aviv, Israel" -- the invented headquarters being placed on a map during the capture; promotion-placement calls selling ad inventory against the fabricated taxonomy; and asset routes for AI-branded product surfaces, including avatar files and sections named for specific commercial AI systems, on the same site whose pipeline manufactures the profiles those systems then cite. Screenshot_20260709_101748 / _101921 / _101931

Chain of custody

Every load-bearing timestamp in this record belongs to a third party. The domain registrations are dated by the registrar. The site content is dated by GitHub's commit history. The fabricated profile is dated by the Internet Archive and archive.today, both of which recorded their July 9 snapshots as the first in their history for that URL. The archival request logs are dated by archive.today's capture clock. The AI Mode conversation is timestamped inside Google's own product and logged to the author's Google account. The phone screenshots carry their original filenames and EXIF data and are reproduced in this repository unmodified. The working session in which this record was assembled -- including live fetches of the pages above by a second, independent AI system, which read the evidence, corroborated the live state of the web, and revised its own initial explanations when the captures contradicted them -- is preserved separately as a process record.

The primary evidence page ends by observing that a measurement system, unable to understand a new word, manufactured a false and more legible identity for it. This page records the sequel: the manufactured identity grew for eleven days, displaced the word's real meaning in the machine's answer, and was finally used to inform the word's author that his own work does not exist. Then, shown the evidence, the machine agreed -- in a reply that changes nothing the machine will say tomorrow. The word names exactly this. The record now demonstrates it twice.