The JS rendering point is critical. Even though bots like GPTBot technically have headless capabilities, they often fall back to text-only extraction for non-priority pages to save compute. We see a lot of "invisible" content in e-com especially because of this.
One other signal to check: internal linking structure. AI crawlers seem to respect semantic clusters more than traditional pagerank flow. If your "about" page isn't semantically linked to your "product" page in a way the LLM understands as a relationship, it often hallucinates the connection.
Thanks for the detailed feedback. Those are the next items on my list now. Will add headless browser research capabilities to go around java script issues. Will also add semantic clustering check.
Seems like you are quite well versed with the space. Would you be open to sharing some interesting resources or getting on call with me to share if you have struggled with this problem and what your workflow looks like?
One signal I think matters a lot for GEO that traditional SEO ignores: specificity of claims. AI models seem to prefer content that makes concrete, verifiable statements over vague authority pages. Would be cool to see Potatometer check for that kind of content quality signal.
Both of these are making it into the roadmap. The semantic internal linking cluster point is particularly interesting, most tools treat internal links as a PageRank signal but you’re right that LLMs parse them differently. And the specificity of claims check is something I hadn’t explicitly scored for but makes intuitive sense given how RAG systems work. Adding both.