The Invisible Layer Your Website Is Missing
Every day, AI crawlers visit your website. GPTBot, ClaudeBot, PerplexityBot - they request your pages, parse the response, and move on. What they take away from that visit isn't what you think.
Two visitors, two very different experiences
When a human visits your homepage, the browser renders HTML - it parses styles, executes JavaScript, builds the visual layout, and displays the result. The visitor sees your headline, your pricing, your value proposition.
When an AI crawler visits that same URL, it sees none of that. It reads the raw response: a tangle of HTML tags, inline styles, script bundles, and metadata built for browsers, not for machines trying to extract meaning. Your actual content - the product name, the company description, the key claims - is buried in that noise. The AI has to guess at what matters.
This isn't a hypothetical problem. It's the structural reason why a site can rank well in traditional search and remain effectively invisible to AI-generated answers.
What structured data was built for - and what it's become
Structured data, specifically Schema.org markup delivered as JSON-LD, was designed to solve exactly this problem. Instead of making an AI infer that your page is about a software product, you declare it explicitly: this is a SoftwareApplication, here is its name, here is its price, here is the publisher. The meaning is stated, not implied.
For years, SEO professionals used structured data primarily to earn rich results in Google Search - the star ratings, product images, and enhanced listings that improve click-through rates. That use case still exists for many schema types: Product, Article, LocalBusiness, Event, and others continue to produce rich results when implemented correctly.
But two once-popular schema types no longer deliver what many sites still expect. HowTo rich results were fully deprecated by Google in September 2023. FAQPage rich results are now restricted to authoritative government and health websites - for most sites, no visible enhancement in search. This is a rich results problem, not a schema problem. What's no longer true is that either type will earn you an enhanced listing in Google. What remains true is that semantically clear markup communicates more reliably to machines than unstructured HTML does.
What's shifted is the second job structured data now has, one that matters regardless of rich results. Schema markup makes content legible to AI systems in ways that raw HTML doesn't. When Google's AI Overviews surface information about a product or service, structured data gives the system explicit signals about what the content represents. The relationship between schema and citation in systems like ChatGPT or Perplexity is less direct - those systems draw on training data and crawled content more broadly - but the underlying principle holds: explicit structure communicates more reliably than implicit meaning.
The parallel representation your site isn't serving
The deeper issue is that structured data embedded in HTML - even perfect JSON-LD - is still delivered alongside all the noise AI crawlers have to filter out. The markup is present, but it isn't surfaced cleanly.
A growing approach to this problem is to give AI systems a separate, dedicated layer of content: files that exist in parallel with your website and are built specifically for machine consumption. Not HTML pages cleaned up for readability, but formats designed from the ground up to communicate meaning - schema.json files that declare your entities and relationships, Markdown versions of your content stripped of presentation, and llms.txt files that give AI crawlers a structured index of what's on your site. It's worth noting that llms.txt is an emerging convention, not yet a formally ratified standard, but it's being adopted by a growing number of sites as a practical signal for AI crawlers.
This is where the idea of a second voice for your website becomes practical. The same content, expressed twice: once for humans in HTML, and once for machines in formats they're built to process.
If you want to understand how schema markup maps to this kind of AI-readable output, the schema markup generator on Geordy walks through the full breakdown - supported schema types, use cases, validation tools, and how schema.json fits into the broader stack of AI-facing formats.
Why implementation details matter more than intent
Most sites that have taken steps toward structured data have done so incompletely. Common problems include schema that references entities not present on the page, markup where the structured data contradicts the visible HTML - a price in schema that doesn't match the on-page price - and outdated implementations built around deprecated schema types.
These aren't edge cases. For AI systems that use structured signals to gauge how much to trust a source, inconsistency is worse than absence. A page with no schema is ambiguous. A page with schema that conflicts with its own content sends a signal that actively undermines reliability.
The practical implication is that structured data isn't a one-time task. It needs to stay synchronized with your actual content as products change, pages update, and new content is added. That's a maintenance problem most sites never adequately address. Geordy.ai addresses this by scanning your site and automatically generating the full set of AI-facing formats - schema, Markdown, llms.txt, and others - and keeping them current as your content changes, without touching how your site looks or functions for human visitors.
The layer that makes everything else legible
The shift here isn't about any single format or schema type. It's about recognizing that your website has always had two audiences - humans and machines - but for most of the web's history, the machine-facing layer was an afterthought. That's no longer a workable position. AI systems now represent a real and growing channel for discovery, and they process content by fundamentally different rules than search crawlers do. The sites that understand this early - and build that second layer deliberately - are the ones with a structural advantage as that channel matures. Getting cited isn't about luck. It's about legibility.
In practice, this shift is less about replacing your existing site and more about extending it. The human experience remains unchanged, while the machine-facing layer quietly improves how your content is interpreted, indexed, and surfaced across AI systems.
As AI-driven discovery grows, the sites that invest in clarity—not just visibility—will be the ones that consistently appear in answers, not just search results.
