TLDR: Everyone is talking about llm.txt like it's a cheat code for AI visibility. It isn't. Three independent studies show zero correlation between having one and getting cited by AI tools. Google confirmed they don't use it. Only 7.4% of Fortune 500 companies even bother. It was built as a developer documentation tool, not a business visibility hack. If you want AI models to actually recommend your business, invest in quality content, proper site structure, and real industry authority. There are no shortcuts.
It started with a simple idea. In September 2024, Jeremy Howard of fast.ai proposed a humble text file called llm.txt. The goal was straightforward: give large language models a clean, markdown-formatted version of developer documentation so they could "read" it more easily. It was a tool built by an engineer, for engineers, to solve a very specific engineering problem.
Fast forward to today, and you’d think it was the second coming of the Sitemap.
I've watched as LinkedIn "thought leaders" and marketing agencies pivot their entire Generative Engine Optimization (GEO) strategies around this single file. The narrative is enticing: "Just drop this file on your server and suddenly ChatGPT will know everything about your business." It’s the ultimate low-effort, high-reward promise that the tech world loves to swallow whole.
There is just one problem. It doesn’t actually work that way.
The Irony of the Afterglow llm.txt
Before I get into the data that dismantles the hype, I should probably point something out. If you go to afterglow.engineering/llm.txt, you will find one.
Yes, Afterglow has implemented the very thing I'm about to tell you isn't a magic bullet.
The irony isn't lost on me. I put it there because, as an engineer, I like clean data structures. If a developer wants to point an LLM at Afterglow’s site to understand the technical capabilities on offer, that file makes it slightly more convenient. It’s a polite gesture. But I didn't do it because I thought it would skyrocket Afterglow to the top of Perplexity's search results.
Honesty is a core value here. And the honest truth is that for 99% of businesses, an llm.txt file is currently doing exactly nothing for your AI visibility.
What the Data Actually Says
In the tech world, hype travels at the speed of light while data moves at the speed of a rigorous peer review. Now that there are a few months of real-world usage data available, the numbers are starting to trickle in. They aren't pretty for the llm.txt enthusiasts.
Let’s look at the SE Ranking study from November 2025. They analyzed over 300,000 domains to see if there was any connection between having an llm.txt file and appearing in AI-generated responses from tools like Google’s Gemini or OpenAI’s SearchGPT.
The result? Zero correlation.
Not "a little bit of a boost." Not "it helps in certain niches." Zero. Sites with the file were no more likely to be cited than sites without it.
Then there is the crawl data. OtterlyAI released a report showing that only 0.1% of AI bot requests are actually looking for llm.txt files. The results showed that while a handful of developers were finding the files useful for specific documentation tasks, the vast majority of AI agents were simply ignoring them. The bots aren't even checking the front door; they're still climbing through the windows by scraping your HTML and reading your structured data.
It’s a classic case of what I call "Technological Hope." People want there to be a simple fix for a complex problem. They want to believe that by adding one file, they can bypass the years of work it takes to build real digital presence. But the web has always been more complicated than that, and AI-driven search is no different.
The models are essentially trying to replicate how a human researcher would evaluate your business. A human wouldn't just look for a summary file; they’d look at your case studies, your reviews, your technical white papers, and your industry contributions. The AI is doing the same thing, just at a scale most can't easily fathom. If you try to shortcut that process with a text file, you're missing the forest for the trees.
If that wasn't enough, the big players have been refreshingly blunt. In December 2025, John Mueller and Gary Illyes from the Google Search team confirmed that Google does not use llm.txt for ranking or for informing their AI overviews. They already have sophisticated ways of understanding your site. They don't need a text file to do it for them.
The Fortune 500 Reality Check
Despite the noise in the "AI Strategy" circles, the biggest companies in the world aren't buying it yet. A study by ProGEO.ai in March 2026 found that only 7.4% of Fortune 500 companies have implemented an llm.txt file.
When you see that low of an adoption rate among the companies with the largest R&D budgets on the planet, it’s usually for one of two reasons: either they know something you don't, or they’re waiting for actual proof of value. In this case, it’s both.
Most businesses are treating llm.txt like it’s a new version of the robots.txt file or a mandatory SEO tag. But remember Jeremy Howard’s original intent: it was for developer documentation.
If you run a SaaS company with a complex API, llm.txt is great. It helps developers use your tool. If you run a law firm, a manufacturing plant, or a luxury e-commerce brand, an LLM doesn't need a special text file to understand what you do. It needs high-quality, authoritative content.
Why the Hype Persists
Why is everyone still talking about it if the data is so bleak?
Because it’s easy.
Real AI visibility—actually getting your brand mentioned as a trusted source in an LLM’s response—is hard. It requires building genuine topical authority. It requires having your work cited by other authoritative sources. It requires technical excellence in how your site is built.
Selling a "magic file" is much easier than selling a long-term engineering and content strategy. It’s the "Get Rich Quick" scheme of the AI era.
Think about the incentives here. If you're a marketing agency, telling a client they need to completely overhaul their technical architecture and invest in six months of expert-led content development is a tough sell. But telling them they can spend an hour on a new "AI Visibility File"? That's an easy win for your monthly reporting.
The problem is, an easy win that produces zero results isn't a win at all. It’s just busywork. This has happened before with the meta keywords tag. People spent years stuffing those tags with every keyword imaginable, only for Google to eventually come out and say they’d been ignoring them for a decade. The llm.txt file feels like history repeating itself, just with a more modern acronym.
At Afterglow, the focus is always on what actually moves the needle. The pattern is clear: the companies that win in new technical environments are the ones that focus on the core fundamentals, not the latest gimmick.
What Actually Works for AI Visibility
If llm.txt isn't the answer, what is?
At Afterglow, I spend a lot of time looking at how models actually retrieve and process information. The "secret" isn't a secret at all; it’s just the fundamentals of the modern web, executed with more precision.
Content quality is the only true currency. LLMs are essentially sophisticated pattern matchers. If your content is thin, repetitive, or sounds like it was written by a 2023-era chatbot, the models will treat it as low-value noise. They are looking for unique insights, data-backed claims, and clear, structured thinking.
Semantic HTML and structured data are your real best friends. Instead of hoping a bot finds a separate text file, make your actual website easy to read. Use proper header hierarchies. Use Schema.org markup to tell the bots exactly what your products are, who your authors are, and what your business does. This is information they are already programmed to consume.
Topical authority and citations are what move the needle. AI models don't just "read" the internet; they "understand" relationships. If reputable industry publications link to you and cite your expertise, the models learn that you are a trusted source. You can't hack this with a text file. You earn it by doing good work and sharing it.
The No-BS Takeaway
Should you add an llm.txt file to your site?
Sure, why not? It takes ten minutes, and it doesn't hurt anything. I did it for Afterglow. But don't do it because you think it’s a strategy.
If you’re a business owner or a marketing director, don't let anyone tell you that this file is the key to your AI future. It’s a distraction from the real work.
The future of "searching" is changing. The world is moving from blue links to synthesized answers. Getting into those answers requires a website that is technically flawless, content that is genuinely useful, and a brand that people actually trust.
Stop looking for the magic bullet. Start building something worth finding.
That’s how you win in the AI era. No text file required.