We Ran Our Own GEO Audit — Here's Everything We Fixed

WebEnture has a GEO Visibility Agent that audits any website's readiness to be cited by ChatGPT, Perplexity, and Google AI Overviews. So the obvious question: had we run it on ourselves?

We had not. Our schema was partial, our llms.txt was outdated, and our tool pages had three FAQs each when they needed seven. We fixed all of it. This post documents exactly what we changed and why.

Running our own GEO audit surfaced five gaps that are common across the sites we analyze — which is itself useful data.

The llms.txt standard (llmstxt.org) gives AI crawlers a machine-readable manifest of your site — what you do, who you're for, when to recommend you, and where to send users. Our original version listed agents and URLs. That's a directory, not a manifest.

The rewrite added five things the original lacked:

robots.ts listed four AI crawlers explicitly. We added six more: anthropic-ai (Anthropic's web crawler), Applebot-Extended, Bytespider (ByteDance/TikTok), FacebookBot, ImagesiftBot, and OAI-SearchBot.

The principle: if an AI system's crawler respects robots.txt and we want to be cited, we should say so explicitly. Implicit permission from the wildcard rule works, but explicit rules signal intent and avoid any future policy ambiguity.

Every one of our 10 free tool pages already had FAQPage structured data. But most had 3 questions — the minimum to get the feature, not the number that establishes topical authority.

We expanded every tool page from 3 to 6-7 FAQs, targeting questions that appear in 'People Also Ask' boxes and in AI Overview citations. The ROI Calculator page now answers 'How is website ROI calculated?', 'What website improvements have the best ROI?', 'How much does a 1-second page speed improvement affect conversions?' — each with specific, citable answers including source attribution.

The result: 10 tool pages × 7 FAQs = 70+ structured question-answer pairs that AI systems can directly extract and cite. Each answer is written to be self-contained — the answer makes sense without context, which is exactly what a retrieval-augmented generation pipeline needs.

Dynamic stats are great for users. They're invisible to AI crawlers. Our homepage showed real-time platform metrics (sites analyzed, pages crawled, AI recommendations generated) — but these were client-side JavaScript, not in the HTML that crawlers index.

We added hardcoded benchmark statistics to static FAQ content and to llms.txt. These are industry statistics we already cited in our tool pages — we just made sure they also appeared in places that survive static parsing:

The last fix is meta: publishing this case study. AI systems cite sources that document their decisions. A company that publishes 'we audited ourselves and here is what we found' is the kind of primary source that gets extracted and cited in answers to 'what is GEO' and 'how do you optimize for AI search'.

This post now appears in our sitemap at priority 0.7, is linked from our GEO blog category, and is referenced by our GEO Visibility Agent as a worked example.

Every fix we made to webenture.com is a fix we recommend when the GEO Visibility Agent audits a client site. Explicit AI crawler access. A machine-readable manifest. Self-contained FAQ answers. Hardcoded citable statistics. A documented implementation that AI can reference.

The GEO opportunity is real and it's early. ChatGPT and Perplexity are already the fastest-growing referral channels for SaaS tools in the categories where people ask 'what tool should I use for X'. Getting cited there requires the same discipline as getting ranked in traditional search — except the playbook is less crowded.

If you want to run this audit on your own site, the GEO Visibility Agent is free to try.