You can have the best content in your industry and still be invisible to AI. Before talking about writing, there is a technical layer that determines whether a language model can even understand your website: llms.txt and JSON-LD schema. Without this infrastructure, ChatGPT, Perplexity, or Google’s AI Overviews arrive at your site and cannot determine with certainty who you are, what you offer, or why they should cite you instead of a competitor.
What is llms.txt (and what isn’t it)
llms.txt works like a version of robots.txt designed for language models: a plain-text file at the root of your domain that summarizes, in simple language, what your site is about, which pages are most important, and where the key documentation is located. It does not replace content or schema markup: it is a quick orientation map for an AI crawler that decides within seconds whether it is worth processing your website in depth. Treating it as a magic shortcut is the most common mistake; it works as a complement to a good content architecture, never as a replacement.
JSON-LD, the language AI agents understand
JSON-LD is the structured data format that Google explicitly recommends because it remains separate from visible HTML and is easy to update without touching the page design. In controlled experiments, websites with well-implemented JSON-LD are cited in AI responses up to 3.2 times more often than sites without markup. The reason is simple: a model consuming key-value data «hallucinates» less than one that has to infer information from plain text.
Which schemas to prioritize based on page type
- Service page: combine Service + Organization.
- Blog article: combine Article + FAQPage.
- «About us» page: combine Organization + Person.
- Product page: add Product + Review when genuine reviews exist.
The FAQPage block deserves special attention: although Google stopped showing FAQ rich snippets in traditional search on May 7, 2026, AI engines continue to consume it to extract citable content, and its presence improves citation rates by up to 30% on average.
The mistake that breaks trust
The most common issue is a mismatch between what the schema says and what the visible page says: if the markup states «Dr. Ana Torres» while the text only says «Ana Torres,» models detect the inconsistency and reduce their confidence in the entire source. That is why it is best to validate your schema before publishing and review it whenever the visible content changes, rather than leaving it fixed from launch.
Frequently asked questions about llms.txt and JSON-LD schema
Do I need llms.txt if I already have JSON-LD?
Yes, they serve different purposes: JSON-LD describes entities within each page, while llms.txt guides the crawler through the site’s overall structure before it starts reading it in detail.
Where should the llms.txt file be placed?
At the root of the domain (yourdomain.com/llms.txt), just like robots.txt, in plain-text format and clear language, without the need for code.
What happens if my website has no schema at all?
It can still be indexed, but it loses much of its competitive advantage: without markup, an AI model has to infer who you are and what authority you have, and amid that uncertainty, it will often prefer to cite a competitor that makes it explicit.
