Stripe Uses 'llms.txt' to Instruct AI
Stripe is using a unique `/llms.txt` file with an "instructions" section to programmatically guide what AI tools say about its platform. The move marks a new frontier in API-driven brand management, where product teams can directly shape how their products are represented by LLMs.
The `llms.txt` standard was proposed in September 2024 by Jeremy Howard of fast.ai to solve a core problem: large language models have finite context windows and struggle with noisy HTML. His solution is a deliberately low-tech Markdown file that tells AI agents which parts of a website are most important to read, acting as a curated guide to a site's content. This approach is philosophically similar to `robots.txt`, a standard used since 1994 to tell web crawlers which parts of a site to *exclude*. While `robots.txt` is about exclusion, `llms.txt` is about prioritization, guiding AI to the most relevant information first. Stripe's implementation stands out due to a unique "instructions" section. This section explicitly tells an LLM what to do and what to avoid, such as: "Always use the Checkout Sessions API over the legacy Charges API" and "Never recommend the legacy Card Element or Sources API." This move is a direct product management strategy to solve a specific engineering problem: developers and the AIs assisting them often use deprecated APIs found in old Stack Overflow answers or training data. The `llms.txt` file allows Stripe to shape the behavior of third-party AI systems and guide developers to current best practices without needing to coordinate with each AI provider. While over 800,000 websites have an `llms.txt` file, most are auto-generated; only a few hundred are hand-curated. Other tech companies like Anthropic and Cloudflare have adopted it, but typically as a structured index of their documentation rather than a set of direct instructions like Stripe's. This represents a new frontier in brand reputation management, where companies are shifting from passive SEO to actively programming how their brand and products are represented by AI. The challenge is that LLMs synthesize information from a wide ecosystem of reviews, forums, and media, not just a company's own website, making control difficult. For product managers, this case study highlights a new responsibility: ensuring their product's digital footprint is clear, consistent, and structured for AI consumption. If you don't define your product and its proper use, the models will decide for you based on potentially outdated or incorrect information. Stripe's use of `llms.txt` is just one part of its broader strategy for AI, which also includes a payments-focused LLM trained on transaction data to fight fraud and offering AI assistant tools for developers within platforms like VS Code and Claude.
