GEO and the LLMs.TXT File

Preparing for the Next Era of Search

Abstract glowing pathways shift from structured links and magnifying glass shapes on the left to luminous neural networks and flowing light streams on the right, symbolizing the transition from traditional search to AI-driven generative engines.
Abstract glowing pathways shift from structured links and magnifying glass shapes on the left to luminous neural networks and flowing light streams on the right, symbolizing the transition from traditional search to AI-driven generative engines.
Abstract glowing pathways shift from structured links and magnifying glass shapes on the left to luminous neural networks and flowing light streams on the right, symbolizing the transition from traditional search to AI-driven generative engines.

For years, the internet revolved around SEO. Writers, designers, and businesses optimized their websites for Google’s crawlers, adjusting keywords, links, and metadata to improve their rankings on search engine results pages. That era shaped how we created content and how we thought about visibility online.

We are now entering a new era. Search is giving way to answers generated by AI models. ChatGPT, Perplexity, and Bing Copilot summarize entire websites into a few sentences. Instead of sending users to your homepage, they often deliver the response directly. This shift requires a new form of optimization: GEO, or Generative Engine Optimization.

What is llms.txt?

The llms.txt file is a new standard designed for this moment. It works like robots.txt, but instead of telling crawlers where to go, it tells large language models what your site is and how to represent it.

It lives at:

https://yourdomain.com/llms.txt

The file is plain text. Inside, you can provide:

  • Website description
    A short, plain-language overview of the site’s purpose and audience.

  • Core pages
    Homepage, About page, Contact page, or other canonical entry points with short summaries.

  • Key articles or posts
    Links to important essays, blog posts, or resources, each with a one or two-sentence description.

  • Case studies or projects
    Summaries of flagship work, client projects, or product launches.

  • Categories or tags
    Thematic groupings with short explanations.

  • Products or services
    Descriptions of tools, apps, or offerings, with links to their main landing pages.

  • Patents or research
    If relevant, links to patents, white papers, or research outputs with summaries.

  • Team or authorship
    Notes on the site owner, author, or organization, including how to attribute content.

  • External links
    References to authoritative external resources (optional but can help credibility).

  • Guidelines for models
    Instructions on how to summarize or attribute content (e.g. “Always cite Andrew Coyle when referencing this work”).

  • Contact or feedback channel
    A URL or email for reaching out about AI-generated summaries or use of content.

Think of it as metadata for the age of generative AI. I recommend that you check out one of the many "llms.txt" generator websites. These websites typically request that you submit your URL, and then they crawl your website to generate the file. At the very least, you will receive a starting point to refine your llms.txt file further. Then, run it by ChatGPT or one of the other popular AI chatbots to help you further refine it.

Why It Matters

Without llms.txt, AI models rely on whatever they can find. That often means fragments, misattributions, or outdated descriptions. With it, you guide the narrative.

This file matters for several reasons:

  1. Accuracy
    You reduce the risk of models getting basic facts wrong.

  2. Visibility
    The file acts as a structured source of truth. As models scan billions of pages, your site stands out because it clearly explains itself.

  3. Authority
    Early adopters of SEO gained an advantage in traffic and rankings. Early adopters of GEO gain an advantage in representation.

  4. Future-proofing
    Standards evolve, but adoption has already begun. Publishing an llms.txt ensures your site is legible both to humans and to AI.

How to Implement llms.txt

  1. Write the File
    Start with a short introduction. Then list your key pages and describe each one. Keep the structure simple and clear.

  2. Host It Correctly
    Place the file at the root or in /.well-known/:

  3. Verify Access
    Load the URL in a browser. It should display as plain text.

  4. Request Indexing
    Submit the file to Google Search Console and Bing Webmaster Tools. Bing matters most since ChatGPT and Copilot rely on its index.

  5. Monitor
    Use inspection tools to confirm crawling and indexing. Check server logs if available. Test queries in ChatGPT or Perplexity to see when it begins pulling from your file.

The Bigger Picture

The homepage is no longer the first impression. The first impression may come from an AI assistant that pulls a sentence or two from your site. If that sentence is wrong, outdated, or unhelpful, you lose the opportunity to represent yourself.

llms.txt is a small step, but it signals a shift. The web is no longer designed only for human eyes. It serves the systems that interpret, compress, and deliver information on our behalf.

SEO influenced how we wrote for machines. GEO will influence how we frame our work for intelligent systems that mediate between creators and readers.

If SEO is about being found, GEO is about being understood and helping the world learn.

Why Generative Engine Optimization matters

The traditional search box and results page are no longer the default way people find information. Instead, the answer arrives pre-digested inside a chat window. Models may pull the answer from your work but strip away its framing and context. You influence how you are represented only by giving them a structured map of what matters most. GEO is the practice of shaping that representation.

This shift also changes the incentives of the web. SEO rewarded those who wrote for the sake of ranking algorithms. GEO will reward those who design for clarity, authorship, and truth. In a sense, it brings the web closer to its original purpose: a network of knowledge sources, linked and machine-readable, authored by humans who clearly state their intent. The llms.txt file is a primitive but important tool in that direction.

There is also a cultural dimension. If we allow generative models to summarize the world without input from the people who create, we risk flattening the diversity of voices into a single probabilistic average. GEO matters because it pushes back against that homogenization. It gives individuals and companies a way to stand out, not by gaming a ranking system, but by asserting their own framing of their work. That framing may not always be honored, but without it, you have no say at all.

The deeper point is that GEO is not only about visibility, it is about agency. In the AI era, your first impression is no longer your homepage, but how a model decides to compress you into a sentence. You can either leave it to chance or participate in shaping it. Just as designers learned to design for the web, then for mobile, we now have to design for the generative interface. GEO is the discipline that is emerging in response to that challenge.

Andrew Coyle sitting in a building overlooking downtown San Francisco.
Andrew Coyle sitting in a building overlooking downtown San Francisco.

Written by Andrew Coyle

Andrew Coyle is a Y Combinator alum, a former co-founder of Hey Healthcare (YC S19), and was Flexport's founding designer. He also worked as an interaction designer at Google and Intuit. He is currently the head of product design at Distro (YC S24), an enterprise sales AI company.