So there's this thing happening that not everyone is talking about publicly, but a lot of people are doing privately.
Some websites are creating content that's designed to be read by AI systems, not humans. Hidden pages, machine-readable text layers, structured data specifically for LLM consumption.
It's called "LLM-only pages" or sometimes "AI-optimized hidden content." And it's raising questions about where the line is between smart optimization and something shadier.
I've been looking into this because I keep seeing it mentioned in private conversations, even as it stays off the main SEO conference stages. Let me share what I've learned.
What Are LLM-Only Pages?
The basic idea is simple: create web content that AI crawlers can see and learn from, but that human visitors never see.
These pages are stuffed with structured data, entity relationships, and optimized text that LLMs can easily parse. They're designed to influence how AI systems understand and reference your brand.
Think of it as writing for the machines, not for people.
There are different implementations. Some use hidden HTML layers that render for bots but not in browsers. Others create separate pages that aren't linked from the main navigation. Some deliver content conditionally based on user agent. And others use JSON-LD and structured data that goes beyond normal schema markup.
Some people are using React components that specifically check for AI crawler user agents (GPTBot, ClaudeBot, PerplexityBot, etc.) and only render content for those visitors.
Why People Are Doing This
The motivation isn't hard to understand.
Traditional web content is designed for humans: visuals, interactivity, brand storytelling. But AI crawlers don't care about your beautiful design or your compelling narrative arc. They want clear, structured, extractable information.
There's often a mismatch between what makes good human content and what makes good AI-consumable content.
LLM-only pages are an attempt to serve both audiences optimally—humans get the content designed for them, AI systems get the content designed for them.
The people doing this argue that AI systems need different formats than humans, that providing AI-readable content helps AI give better answers, that it's just like having mobile and desktop versions of a site, and that the information is the same, just presented differently.
What the Research Shows
Is it actually working? The early data suggests yes, at least for some implementations.
Companies using comprehensive LLM optimization strategies report higher citation rates in AI responses. Some studies show pages with robust structured data and AI-friendly formats appear up to 10x faster in AI search results.
Adobe released an LLM Optimizer tool specifically because many websites have content that's invisible to AI agents due to JavaScript rendering. Their tool identifies what AI can and can't see—and a lot of sites discover their most important content is invisible to AI crawlers.
Gartner predicts a 25% decline in traditional search as AI handles more queries directly. If that's true, optimizing for AI visibility becomes business-critical, not optional.
The Ethical Questions
Here's where it gets complicated.
The critics argue that if content is hidden from humans but shown to AI, that's deception. It's essentially cloaking, which Google has always penalized. It creates a two-tier internet: one for people, one for machines. And if everyone does this, the web becomes optimized for bots, not users.
The defenders counter that it's not deception if the information is the same, that you're providing the same content in different formats for different consumers, that it's no different than creating machine-readable feeds alongside human-readable pages, and that AI systems serve human users, so optimizing for AI ultimately serves humans.
Honestly? Both sides have points.
The Cloaking Comparison
Traditional cloaking—showing one thing to search engines and something different to users—has been against Google's guidelines forever. It's considered manipulative.
Is LLM-only content the same as cloaking?
The technical answer: kind of. You're showing different content to different visitors based on user agent.
The nuanced answer: it depends on what you're doing with it.
If you're creating LLM-only pages that contain the same information as your human-facing pages, just formatted differently, that's arguably not deceptive. You're adapting presentation, not misleading anyone.
If you're creating LLM-only pages that make claims you wouldn't make on your public site, that's more clearly problematic.
The proponents of LLM-only content emphasize that the information should be consistent: "The human version and the bot version say the same thing. One is interactive, the other is plain text."
How It's Actually Implemented
For those curious about the technical side:
The simplest approach uses user agent detection. AI crawlers identify themselves (GPTBot, ClaudeBot, PerplexityBot, etc.), so you can serve different content to those user agents.
There are React components specifically designed for this. They check the user agent against a list of 30+ known AI bot signatures and conditionally render content.
More sophisticated approaches use structured data layers—extensive JSON-LD that provides machine-readable versions of your content without creating fully separate pages.
Some implementations use server-side rendering to ensure AI crawlers see fully rendered content (since many AI crawlers don't execute JavaScript), while human visitors get the standard client-rendered experience.
The Sustainability Question
Here's something that concerns me about this approach: what happens when it scales?
If LLM-only pages become widespread, AI companies will adapt. They'll discount content that's only visible to their crawlers. They'll look for cloaking signals. They'll prioritize content that humans actually engage with.
This pattern has happened in SEO repeatedly. A tactic works until it's widespread, then the algorithms adapt to neutralize or penalize it.
The people using LLM-only pages today might get an advantage. The people using them in two years might get penalized. That's how these things often play out.
A More Sustainable Alternative
Instead of creating hidden content for AI, there's an argument for making your human-facing content more AI-friendly.
This means using clear, extractable structures like headings, logical organization, and well-organized information. It means including clean summaries that could be quoted. It means implementing comprehensive schema markup. It means ensuring content renders without JavaScript for crawlers. And it means being explicit about entities and relationships.
This approach serves both audiences with the same content. No hidden layers. No user agent detection. Just well-structured content that works for humans and machines.
It's less aggressive than LLM-only pages, but it's also less risky.
Where I Land on This
I've been thinking about this a lot, and here's my current position:
I understand why people do it. The gap between human-optimized and AI-optimized content is real. Serving both audiences well is a legitimate goal.
The "same information, different format" version seems defensible. If you're genuinely providing the same content in a more machine-readable way, that's not fundamentally dishonest.
The slippery slope concerns are real. Once you start creating bot-only content, the temptation to optimize it specifically for AI ranking factors—even at the expense of accuracy—is hard to resist.
It's probably not sustainable long-term. AI companies will eventually discount content that humans don't see and engage with.
There are safer ways to improve AI visibility. Better structure, comprehensive schema, ensuring crawlability—these approaches improve AI visibility without the ethical complications.
Practical Recommendations
If you're considering this approach, here's my advice:
Start with the basics first. Make sure your existing content is AI-crawlable and well-structured before creating new bot-only content. Many sites have AI visibility problems that can be fixed without hidden content.
If you do create LLM-only content, keep it consistent. The information should match your public content exactly. Don't make claims to bots that you wouldn't make to customers.
Document what you're doing. If you're implementing user-agent-specific content, have a clear rationale and be prepared to explain it.
Watch the signals. If AI systems start discounting your bot-only content, be ready to pivot.
Consider the reputation risk. If it came out that you were showing different content to AI systems than to humans, how would your customers react? How would journalists describe it?
The Bigger Picture
LLM-only pages are one manifestation of a larger question: how should web content adapt to AI systems?
The old model was simple: create content for humans, let search engines index it. The content was the same for everyone.
Now we have AI systems that consume and synthesize content differently than humans do. Creating optimal experiences for both audiences is a real challenge.
LLM-only pages are one solution. Better-structured human content is another. Comprehensive schema markup is another. Each has tradeoffs.
The right answer probably depends on your specific situation, your risk tolerance, and your ethical comfort level. I don't think there's a universal rule here yet.
What I do think is that being thoughtful about it matters. Making deliberate choices rather than just copying what competitors are doing. Understanding the tradeoffs rather than assuming any tactic that works is fair game.
The AI search landscape is new enough that the norms haven't been established. The companies experimenting now are partially defining what becomes acceptable practice. That's a responsibility worth taking seriously.