The Purr-fect Illusion: How a Fake Cat Standard Exposed the Low Bar of Evidence in AI Search Optimization

LONDON — In the fast-paced ecosystem of digital marketing, the line between genuine innovation and collective delusion is often razor-thin. For months, digital strategists have debated the efficacy of llms.txt, a proposed plain-text standard meant to guide artificial intelligence bots through website content. Its proponents point to a constellation of apparent "proofs"—crawlers fetching the file, search engines indexing it, and AI chat models validating its utility—as evidence of a paradigm shift in Generative Engine Optimization (GEO).

Enter Mark Williams-Cook, a technical SEO expert and industry commentator, who decided to test these foundational assumptions. Tired of what he describes as unverified anecdotes masquerading as empirical data, Williams-Cook engineered a brilliant piece of satire: cats.txt.

Formally defined as a text file where domain owners declare their office cats, their job descriptions, breeds, and a mandatory affection metric known as PurrLevel, the spoof standard was published on Williams-Cook’s blog and cross-pollinated onto LinkedIn.

To the astonishment of few seasoned skeptics and the amusement of many, cats.txt successfully cleared every single benchmark commonly cited to prove the validity of llms.txt. It was aggressively crawled by major AI bots, indexed seamlessly by Google, referenced by generative AI retrieval systems detailing entirely fictional felines, and enthusiastically endorsed by ChatGPT as a valuable ranking signal.

The experiment has since snowballed into an organic internet phenomenon complete with early adopters, third-party implementations, and live server loggers. Yet, beneath the humor lies a sobering critique: the digital marketing industry is currently utilizing a standard of evidence so profoundly low that it validates entirely fictitious data about non-existent office cats without missing a stride.

Main Facts: Deconstructing the cats.txt Phenomenon

The core of Williams-Cook’s experiment relies on exposing the logical fallacies inherent in modern SEO and GEO hype cycles. The initiative began with a straightforward observation: industry professionals were routinely marketing unverified techniques by pointing to basic web behaviors and mischaracterizing them as strategic validation.

To challenge this trend, Williams-Cook established cats.txt based on the following framework:

• The Specification: A formal, over-engineered draft proposing a root-directory text file to declare corporate felines, including telemetry data like PurrLevel (scored out of 10).
• The Seeding: Published initially via a personal blog post and a strategically placed LinkedIn article—a platform Williams-Cook notes LLMs hold in unaccountably high regard.
• The Community Adoption: Far from remaining an isolated joke, technical peers—such as prominent SEO specialist Dave Smart—implemented cats.txt on their own domains. Soon after, an independent developer launched catstxt.org, a fully operational, cleanly organized portal featuring a live-filtered server log viewer.

Most importantly, the file successfully triggered the exact four "proofs" regularly used to validate real technical optimization methods: bot traffic, indexation, conversational references, and direct LLM endorsement.

Chronology: From Irritation to a "Web Standard"

The trajectory of cats.txt from a conceptual grievance to an accidental global movement highlights how rapidly information—and misinformation—circulates within the digital ecosystem.

• Phase 1: Frustration and Conception. For months, Williams-Cook observed industry experts presenting standard web activities as definitive proof that llms.txt was transforming search discovery. Recognizing that counter-arguments were failing to penetrate the echo chamber, he resolved to run the same "proofs" on transparent nonsense.
• Phase 2: Launch and Seeding. Williams-Cook published the formal draft specification online. Understanding the algorithmic biases of the modern web, he deliberately utilized LinkedIn to broadcast the "missing standard for SEO and GEO."
• Phase 3: Viral Propagation. The SEO community recognized the satirical nature of the project and leaned in. Dave Smart added a cats.txt file to tamethebots.com, introducing fictional felines like "Odd"—a Tuxedo cat credited as a "Render Cat" with a specific PurrLevel. Simultaneously, independent engineers launched dedicated infrastructure supporting the spoof standard.
• Phase 4: The Empirical Audit. With the satirical standard live, Williams-Cook tested it against the four primary arguments used by vendors selling AI optimization packages. To the detriment of rigorous methodology, the cat-centric text file passed every single one.
• Phase 5: Public Exposure and Academic Critique. The experiment culminated in conference discussions, such as Martin Splitt’s public inquiry at Athens SEO regarding whether Williams-Cook intended to maintain a "monopoly" on the standard or open it up to the Internet Engineering Task Force (IETF).

Supporting Data: The Four Faulty Proofs

To understand why cats.txt succeeded in fooling algorithmic systems, one must examine the four pillars of "evidence" routinely cited by practitioners selling AI search optimization services.

1. "The AI Bots Crawl It!"

• The Claim: Vendors argue that because server logs show PerplexityBot, GPTBot, ClaudeBot, and Googlebot actively fetching llms.txt, the files must be actively utilized in decision-making processes.
• The Reality: Crawlers fetch virtually any publicly accessible text file left in a root directory. A bot visiting a server is no more an endorsement of the file’s contents than a postal worker touching a gate is an endorsement of household trash. True to form, Williams-Cook’s server logs quickly filled with requests from major AI labs diligently indexing feline job descriptions.

2. "It Was Indexed by Google!"

• The Claim: Indexation is frequently cited as proof of strategic importance. Why would a search engine index data unless it mattered?
• The Reality: Google has indexed plain text files for decades. Indexation merely confirms that a URL exists and contains strings of characters; it does not validate truth, utility, or sanity. Google Search Console readily offers indexing data for British Shorthairs working as "GUI Purrfectionists."

3. "ChatGPT Retained Specific File Data!"

• The Claim: If a generative model cites a fact found exclusively within an optimization file, analysts claim it proves the file is being read as a trusted source.
• The Reality: This behavior is standard Retrieval-Augmented Generation (RAG). The AI runs an internal search, lands on an indexed web page (such as Dave Smart’s cats.txt), and reads the text just as it would any other URL. Consequently, Google’s AI Overviews solemnly reported that non-existent cats chase digital cursors and stash pixels on virtual carpets based solely on scraped text files.

4. "ChatGPT Confirms It Helps!"

• The Claim: Asking an LLM if an optimization file works yields an affirmative response, which marketers treat as validation from the system itself.
• The Reality: This highlights the AI Convergence Problem. Language models do not perform empirical reasoning when asked about emerging trends; instead, they compute a running average of what has already been written across the web. Because the internet quickly became saturated with enthusiastic discussions about cats.txt, ChatGPT dutifully echoed those sentiments back, detailing how feline-based telemetry files could improve machine trust and citation accuracy.

Official Responses and Industry Context

While independent studies continue to challenge the tangible benefits of AI-specific text files, major search engine representatives have been characteristically direct.

Ahrefs previously analyzed over 100,000 domains utilizing optimization text files, concluding that crawlers largely ignore them in practice. Subsequent industry analyses across 300,000 domains similarly reported no measurable citation advantage for websites implementing the standard.

Google’s John Mueller addressed the reality of server log behaviors with notable bluntness in public statements:

"FWIW no AI system currently uses llms.txt, […] It’s super-obvious if you look at your server logs. The consumer LLMs / chatbots (the ones that SEOs want traffic from) will fetch your pages — for training and grounding, but none of them fetch the llms.txt file."

Despite these warnings, agencies continue to package GEO tactics into client invoices, relying on speculative enthusiasm rather than reproducible data.

Implications: The Real Cost of Hype

The cats.txt experiment is ultimately more than an elaborate piece of technical trolling. It exposes a profound vulnerability within the modern digital marketing industry: the displacement of rigorous scientific testing by algorithmic confirmation bias.

Every capital expenditure and labor hour dedicated to chasing unverified optimization rituals represents resources diverted away from foundational, proven digital strategies. True optimization—the core mandate of Search Engine Optimization—relies on cumulative advantages achieved through verifiable, user-focused improvements over time. It does not rely on chasing transient text files that happen to be scraped by automated bots.

As the digital landscape evolves to accommodate generative engines, the bar for professional evidence must rise. Marketers can—and likely will—continue to deploy experimental files on the off chance that formal support materializes in the future. However, selling these speculative practices as scientifically validated optimization levers is an intellectual failure.

The office cats of the internet were entirely transparent about being works of fiction. It remains an open question whether the rest of the AI optimization industry will show the same level of honesty.