Why LLM-only pages aren’t the answer to AI search — What SEOs should do instead
Search Engine Land’s recent piece, “Why LLM-only pages aren’t the answer to AI search,” outlines why machine-only content formats like llms.txt files, .md copies, and /ai pages rarely produce the AI citations teams expect. The article — written by Wasim Kagzi and published Jan 20, 2026 — examines real-world tests and large-scale analyses that show unique, useful content matters more than special file formats.
What the data says
The Search Engine Land article summarizes targeted experiments and an industry study that together make a clear point: AI systems cite the content that is most useful and unique, not the format. One experiment covering nearly 18,000 citations found llms.txt files accounted for just 0.03% of citations, markdown copies received none, and JSON metadata performed only when it contained information not available elsewhere. At scale, SE Ranking’s analysis of 300,000 domains found only 10.13% adoption of llms.txt and showed the file’s presence did not improve citation prediction — removing it made their XGBoost model more accurate.
Why format-focused tactics fall short
Two reasons explain the poor performance of LLM-only formats. First, many AI models have already been trained on standard HTML pages and are adept at parsing normal web content. As the article quotes Google’s John Mueller: “LLMs have trained on – read and parsed – normal web pages since the beginning. Why would they want to see a page that no user sees?” That blunt observation gets to the heart of the issue: building pages only machines will see creates a mismatch with the signals AI models use.
Second, these alternative formats only help when they contain unique, actionable data. Both the targeted tests and SE Ranking’s analysis showed that the file type itself did not drive citations — only content that existed nowhere else on the site did. In other words, it’s the information, not the container.
Practical implications for SEO teams
If your team is debating whether to create /ai pages, add .md mirrors, or deploy llms.txt files, consider three priorities instead:
- Focus on unique value: Prioritize content that provides facts, structured details, or tooling information that competitors do not have. SE Ranking’s conclusion was direct: “Our analysis of 300,000 domains shows that LLMs.txt doesn’t impact how AI systems see or cite your content today.” Use special files only when they expose truly unique datasets (for example, JSON product feeds that contain attributes not published elsewhere).
- Make critical content crawlable and simple: Reduce heavy client-side rendering on pages containing essential information. As Search Engine Land notes, “Excluding JS, which still seems hard for many of these systems,” remains a real technical barrier for AI parsing. Server-rendered, well-structured HTML and clear semantic markup make content accessible both to humans and crawlers.
- Use structured data where supported: When platforms publish official specifications (for example, OpenAI’s product feed guidelines), follow them. Structured feeds and schema.org markup help downstream systems understand and attribute content correctly — but only when the fields add information beyond your visible pages.
How to test and measure impact
Don’t treat llms.txt or /ai pages as a set-and-forget solution. If you experiment, measure domain-level citation frequency and, crucially, ensure the special format contains unique answers. Use server logs to confirm whether major bots retrieve the special files, compare citation counts over time, and validate whether any increase correlates to the file or to other changes (site architecture, content depth, or structured data).
What to say to stakeholders
When stakeholders ask for a quick AI-visibility boost via machine-only pages, explain that the current evidence favors durable, user-focused improvements. Quote the Search Engine Land piece and the SE Ranking findings to set expectations. For example, as Wasim Kagzi reported: “LLMs have trained on – read and parsed – normal web pages since the beginning,” calling into question the need for separate, non-human pages. Complement that with SE Ranking’s empirical takeaway: “LLMs.txt doesn’t impact how AI systems see or cite your content today.”
Bottom line
LLM-specific files and shadow content are low-cost experiments, but they are not a substitute for clear, unique content and sound site engineering. Invest in pages that serve users and include the structured, unique information AI systems need to cite your site. Until platforms publish formal requirements for AI indexing, the best strategy remains the same: build pages that humans value and machines can parse.
Attribution: This article summarizes and analyzes Wasim Kagzi’s “Why LLM-only pages aren’t the answer to AI search” (Search Engine Land, Jan 20, 2026) and SE Ranking’s research on LLMs.txt (Nov 7, 2025). Direct quote from the original: “LLMs have trained on – read and parsed – normal web pages since the beginning,” — John Mueller, quoted in Wasim Kagzi. Additional quote from SE Ranking (Yulia): “Our analysis of 300,000 domains shows that LLMs.txt doesn’t impact how AI systems see or cite your content today.”
Original Search Engine Land article: https://searchengineland.com/llm-only-pages-ai-search-467690
.png)
