2
1 Comment

What We Learned Building an AI Search Visibility Tool

Over the last few months, we've been building TriRank, a tool that helps businesses understand how they appear in AI search.

One thing surprised us: AI visibility is very different from traditional SEO.

A page that ranks well on Google isn't always the one AI assistants cite. In many cases, AI systems seem to prefer pages with clear structure, strong topical authority, and consistent information across the web.

That changed how we think about content. Instead of chasing keywords alone, we've started paying much more attention to creating useful resources, improving site structure, and making information easier for AI systems to understand.

We're still learning, and I'd love to hear from others working in this space.

Have you noticed any differences between ranking well in Google and being mentioned by ChatGPT or other AI search tools?

posted toAvatar for product TriRank
TriRank
  1. 1

    Yeah, consistent with what we've found running audits across a couple hundred sites. Two patterns beyond the structure/topical-authority point:

    1. Crawler access is the silent killer. We checked 100 crisis/safety nonprofits and 50 news outlets with the same rubric — the nonprofits split into "actively fine with AI crawlers" vs "just never decided," and the undecided group (55 of 100) had zero deliberate signal either way: no llms.txt, no crawler-specific robots.txt rules, nothing. A page can have perfect structure and still be invisible if nobody ever pointed the crawler at it or explicitly allowed it. Worth checking before optimizing content.

    2. Independent corroboration outweighs on-page quality more than people expect. A site describing itself as the authority on X matters less than other sites (that the model already trusts) describing it the same way, unprompted. Self-published claims and third-party corroboration aren't treated the same, even when the text is nearly identical.

    Curious what TriRank's data shows on the crawler-access side specifically — is that something you're scoring, or mostly content/structure signals right now?