A GEO agency with a site that AI couldn't read.
We audit client sites for AI visibility problems regularly. When we turned the same process on our own site, the findings were more significant than expected.
The sitemap contained 345 URLs. Of those, 334 returned 404 errors — dead pages that had been built, deleted, reorganized, or renamed at some point in the site's history but never removed from the sitemap. Every time an AI crawler walked that sitemap — Googlebot, Perplexity's crawler, GPTBot — it followed hundreds of broken paths. A model trying to build a confident picture of what TRIVOX is and does hit a wall of dead ends.
The llms.txt file was the more immediate problem. llms.txt is a file designed specifically for AI crawlers — it tells language models what pages to read to understand a site. TRIVOX had one. It listed dozens of pages that no longer existed. Instead of pointing AI crawlers at the site's best content and entity information, llms.txt was directing them at a graveyard of 404s. A model that read it would either ignore the site or form a confused, incomplete picture of who we are and what we do. Neither outcome is acceptable for a company whose core service is helping AI answer that exact question about its clients.
What the cleanup actually involved.
The fix wasn't cosmetic. It was systematic — touching the sitemap, the llms.txt, the content architecture, and the schema layer.
Sitemap cut to live pages only.
We audited every URL in the sitemap, confirmed which pages returned a 200 status, and rebuilt the sitemap to include only pages that exist. 334 dead entries removed. The remaining pages are the ones that actually describe what TRIVOX does, who it serves, and where it operates. A clean sitemap is the first thing a crawler reads — getting this right is not optional.
llms.txt rebuilt from scratch.
We wrote a new llms.txt file that maps AI crawlers directly to the pages that matter: service pages, case studies, the GEO library, location pages, and the core entity definition of TRIVOX AI. Each entry describes what the page covers in plain language, so a model reading llms.txt gets a useful index of the site's knowledge, not a list of URLs to guess at. This is the file that tells AI systems what to read to understand us.
Answer-shaped content added for every service.
The site had service descriptions but lacked the direct, question-answering passages that AI assistants pull quotes from. We added clear, answer-shaped content to every core service page: what the service is, who it's for, what it produces, and how it works — stated plainly in one or two sentences an AI can lift verbatim. Not marketing copy. Machine-readable answers.
Entity data and schema tightened.
Organization schema with correct name, URL, service area, and description. Service schema for each core offering. FAQPage schema on the library pages. Consistent NAP across the site and the Google Business Profile. Every signal pointing at the same clear picture: TRIVOX AI, Toronto, AI automation and GEO, these services, these areas. Confidence for a model comes from agreement across sources — we made the sources agree.
Why the self-case matters more than a polished client story.
Client case studies are valuable, but they're also easy to cherry-pick. What we did to our own site is harder to spin.
When we found 334 dead pages and a broken llms.txt on the site of a company selling AI visibility services, we had a choice: quietly fix it without mentioning it, or publish the honest account. We published it because the honest account is more useful — it shows exactly what a GEO audit finds and what the remediation actually looks like, on a real site with a real mess to clean up.
The result is a site that a model can read. Every page the sitemap lists exists. Every page llms.txt recommends delivers useful content. Every service has answer-shaped content a model can quote. Every schema signal confirms what we do and where we do it. That is the foundation. Without it, the rest of the GEO work — content, citations, entity clarity — sits on broken ground. This is what GEO foundation work looks like in practice. For clients who want the same audit applied to their site, that's what our GEO visibility engagement starts with.
The concrete state of the site after the work.
The outcomes here are technical — because the foundation is technical. The downstream benefit is a site AI can actually cite.
Sitemap: 345 URLs → clean
334 dead pages removed from the sitemap. Every URL that remains is a live page with real content. Crawlers following the sitemap now find content, not errors.
llms.txt: rebuilt and accurate
The file AI crawlers read to understand TRIVOX now points at pages that exist and describe what we do. The index AI models build from it is accurate and complete.
A citable, machine-readable site
Entity data, schema, answer-shaped content, and a clean sitemap now work together. A model reading this site can form a confident, accurate picture of what TRIVOX is — which is the precondition for recommending it.
Related, and what to do next.
Questions, answered.
Why does a broken sitemap hurt AI visibility?
AI crawlers — including GPTBot, Google's AI crawler, and Perplexity's indexer — walk your sitemap to discover content. If the sitemap lists hundreds of pages that return 404 errors, the crawler wastes crawl budget on dead ends and builds an incomplete picture of your site. Clean sitemaps mean more useful content gets indexed and the model forms a more accurate view of what you do.
What is llms.txt and why does it matter?
llms.txt is a file, placed at your domain root, that tells AI language models and crawlers which pages to read to understand your site and business. It's specifically designed for AI crawlers, not traditional search bots. If it lists dead pages or is missing entirely, AI models either skip it or form a confused picture of who you are. A well-built llms.txt is one of the most direct GEO levers available.
Is this kind of technical audit something TRIVOX does for clients?
Yes. Every GEO engagement starts with an audit that covers sitemap health, llms.txt state, schema markup, entity consistency, and NAP agreement across the web. What we found and fixed on our own site is the same diagnostic we run for clients — the same problems appear frequently, often without the business knowing.
Does fixing the technical foundation guarantee AI recommendations?
No, and we won't pretend otherwise. Technical GEO foundation — clean sitemap, llms.txt, schema, entity data — is necessary but not sufficient. It makes your site legible to AI. Whether a model names you also depends on content quality, review volume, citation breadth, and how your authority compares to competitors in the same category. Foundation work removes the blockers; the rest builds the signal.
How long did the rebuild take?
The audit was a day's work. The sitemap cut and llms.txt rebuild were done within the week. Schema and content additions across service pages took a few more weeks. The full foundation rebuild was roughly three to four weeks of focused work — not a months-long project.