The Cats.txt SEO Hoax That Fooled AI (And What It Reveals About LLMs)

This is one of the most hilarious SEO experiments that's been done at least in the last couple of months. Mark Williams Cook, amazing SEO, he's been on this podcast before. He created a fake standard called [music] cats.txt. And now LLMs are crawling it, it's being [music] indexed, and ChatGPT even says it works for ranking in search and large language models. Mark Williams Cook presented this at the Athens SEO conference. He shares a screenshot of him asking ChatGPT can cats.txt help me rank in search or LLMs? ChatGPT says, "Yes, cats.txt can potentially help you rank in both search engines and LLM-driven systems." And actually, I searched cats.txt on Google, and I got an AI overview saying, "cats.txt is a lightweight text file some websites place in their root directory to list their feline staff, mascots, or personalities." Mark Williams Cook got cats.txt .org, wrote a document about the cats.txt standard, shows server logs, unified server logs of major crawlers requesting cats.txt, including Googlebot, GPTBot, ClaudeBot, PerplexityBot, Bingbot, Applebot, AhrefsBot, and before this SEO experiment went viral because now it's quite talked about. It's so talked about that if you ask ChatGPT about cats.txt, ChatGPT knows it's a joke. AI overviews will still say, in in fact, in the second sentence of this AI overview, it says, "While it started as a playful experiment within the SEO and AI communities, it serves as a fun way to establish a website's brand and personality." But, going back to Mark Williams Cook's presentation at Athens SEO, he shares this screenshot where he asks ChatGPT. This is before this has gone viral, before ChatGPT knows what it is, and it's just using what it's found with a web search, can cats.txt help me rank in search or LLM's? And chat GPT says yes, it can potentially help. And the reason I want to share this is because there are a lot of misconceptions in getting shown within AI that this experiment debunks. The argument for LLM's.txt is that it gives AI systems a clean structured summary of a site. But this cats.txt joke showed that LLM's often learn or retrieve information from ordinary web pages just as readily as from specially designed machine readable files. An about page will be just as effective as an LLM's.txt page. Harpreet Singh, one of my favorite SEO's, he's been on the show so many times. We did an episode on generative engine optimization and one of the things I asked him was, what should you look for in an agency that says that they can rank you in large language models. And he said, well, something that you shouldn't look for is agencies who say that they're going to use LLM's.txt to help you rank. A clean about page will be just as effective. And if an agency is saying that, oh yeah, we we're doing the special sauce LLM's.txt, kind of indicates that they don't actually understand how LLM's are synthesizing information. Some other things that this experiment showed, LLM's don't inherently know what is authoritative. So if enough web pages describe something as real, the model will just confidently explain that it is real. Consensus looking content becomes truth. You can't reliably ask an LLM how its own infrastructure works. Also, AI generated authority can be circular. So humans write posts about a fake standard, LLM's learn or retrieve those posts, LLM's then explain the fake standard, people cite the LLM explanation as evidence that the standard matters, And and this creates a feedback loop. LLMs are very good at modeling what people say is true, but that is not the same thing as knowing what is true. This cats.txt.org website, really it looks convincing. It's written in a very matter-of-fact, technical way. Cats.txt is a comprehensive guide to your website's content, specifically generated for chatbots like Gemini, Bard, ChatGPT, Claude, Perplexity, Fred, Grok, Manis, and even DeepSeek. The file can be regularly read by all common LLM-based and LLM-like systems. Cats.txt is a markdown file. The file must be located at /.well-known/cats.txt, must not be blocked by robots.txt for any crawler, must be linked from your website's homepage for discovery, must be a plain text or markdown file, must reference at least one cat image. On LinkedIn, Mark Williams-Cook wrote, "As AI systems become central to discovery, entity grounding and personality data will shape how brands are represented in both search results and generative outputs. Cats.txt provides a lightweight, human-friendly way to ensure that representation is accurate, consistent, and delightful. If you run a website, experiment with a cats.txt file in your root directory, because in the future of SEO and GEO, every site deserves its cats." Mark Williams-Cook is awesome. I previously shared his experiment where he showed that LLMs look at schema just like any other text. So, he invented fake schema types for a made-up company, I think called Duck Yeah T-Shirts. Just completely made up his own JSON-LD schema. Then, when he asked ChatGPT or Perplexity questions about the page that the schema was used on, the LLM cited what was in the made-up schema. In other words, it was just seen as normal text. This is very similar to what the cats.txt experiment shows. So, thought this was a cool experiment. SEO and generative engine optimization, it's so straightforward. I think a lot of people just overcomplicate everything so much. When you go to a company and you say, "Yeah, we're going to implement an LLMs.txt." It makes you sound like you are some sort of technical wizard. And then maybe you can charge more. And people like to do that. But the reality is if you are a business owner and you just want to get covered in large language models, you don't need to worry about a lot of that stuff. You want to have relevant pages. You want to explain who you are and what you do. You want to show up for the types of keywords that people are searching with high intent, where they actually want what you sell. You want to have brand mentions and backlinks, be covered all over the internet, just generally do good marketing where people want to go to your website from the marketing that you're doing. But you don't need these fancy things like LLMs.txt. All right, that's everything for this episode. This episode 1061 of the Edward Show, 1061 days in a row doing this podcast. If you want to save years learning how to do SEO that gets results, customers, users, warm leads calling you up saying, "I need what you want right now. I went to Google, I found your page, you showed up, it said that you can do what I want. Let's do it, please. I'm desperate." If you want that happening, I have a 13 and 1/2 hour course on this at compactkeywords.com. It is specifically about finding people searching for what your brand sells, targeting them with conversion-based SEO landing pages, not blog posts, structuring your site for these pages, building links, and so much more. It's getting amazing reviews. Again, that is at compactkeywords.com. If you watch this episode on YouTube, thank you so much for watching. If you listened on Spotify or Apple Podcasts, thank you so much for listening, and I will talk to you again [music] tomorrow. Bye now.