I finally got ChatGPT to cite my SaaS. Here is the exact markup (and why your React app is invisible to AI).

A few weeks ago, I shared a painful experiment on this blog: I opened Perplexity, asked it to rank my product, and watched it place me dead last.

The problem wasn't my codebase. My site had server-side rendering, a perfect sitemap, and explicit allow-rules in robots.txt. The model could read everything, but it had no semantic reason to trust or cite any of it.

I spent the last few weeks rebuilding the architecture of my directory, CitableHub, to focus entirely on Generative Engine Optimization (GEO).

The results this week?

ChatGPT read the directory feed, recommended 3 listed tools by name, and cited the source 8 times.

Brave Search ranked a listed product #1.

Google AI Overview generated a "People also ask" specifically about a listed product.

Here is the technical reality of AI discoverability that most developers are ignoring right now, and how to fix it.

1. HTML is for humans. llms.txt is for agents. You spend hours obsessing over your CSS and React components. AI crawlers don't care. They want high-signal, markdown-formatted text.

We implemented a machine-readable feed (llms.txt standard) that strips away the UI and serves pure context. If your SaaS doesn't have an endpoint providing structured context for language models, you are forcing the AI to guess what your tool does by scraping your marketing headers. It will guess wrong.

1. The Identity Layer (JSON-LD)
   Inconsistent metadata kills entity resolution. If your dev.to bio, your GitHub repo, and your website's tags describe your app differently, the LLM treats them as unverified claims. We enforced strict structured schema across the board so the AI sees one canonical truth.

2. The "Ask AI" Verification
   To prove this works, I built an "Ask AI" button into every listing on CitableHub.

Instead of trusting my word, users can click one button that opens any of 6 major AI assistants (ChatGPT, Claude, Gemini, etc.) with a pre-filled prompt that reads their product page live and forces the AI to summarize it and cite the source.

The Takeaway for Builders
Traditional SEO was about keywords and backlinks. Generative SEO is about structure, parseability, and verified entities.

If you just shipped a tool, stop worrying about keyword density. Start thinking about how an LLM parses your DOM.

I built CitableHub as a free registry to automate this process for developers. We structure your SaaS data so AI assistants actually show it when users ask for recommendations.

There are over 629 tools already listed. Drop your project in, use the "Ask AI" button on your profile, and see if the models actually understand what you built.

Has anyone else implemented an llms.txt file yet? Curious to hear how it impacted your crawler logs.