{
"title": "I ran a GEO audit on our own blog last Tuesday — here's what actually moved the needle",
"content": "I ran a GEO audit on our own blog last Tuesday — here's what actually moved the needle\n\nLast Tuesday I exported six months of Search Console data and cross-referenced it with what Perplexity and ChatGPT actually cite when they answer questions in our niche. The gap was embarrassing.\n\nWe rank #3 for \"AI content detection tools\" in classic Google. Solid. But when I asked Perplexity \"what's the best AI content detector for academic papers,\" it cited a competitor's 2023 blog post — not ours, not even a recent source. Zero mention of our guide, which had been updated three weeks prior.\n\nThat's the GEO problem nobody talks about: you can win Google and still be invisible to the engines that matter now.\n\n## What I actually tested\n\nI picked 12 queries where we had decent organic rankings (positions 1-10) and checked whether AI engines cited us, cited competitors, or cited nobody. I used Perplexity, ChatGPT with browsing, and Google's AI Overview.\n\nResults across 12 queries:\n- AI engines cited us directly: 2 times\n- AI engines cited a competitor: 7 times\n- AI engines synthesized without clear attribution: 3 times\n\nTwo out of twelve. That's a 17% citation rate despite a 58% top-3 Google ranking rate.\n\n## The pattern I noticed\n\nThe two queries where we got cited had something in common: they contained specific, structured data tables with clear sourcing. One was a comparison of detection accuracy rates across five tools, tested on 500 samples. The other was a timeline of GPTZero's API changes with version numbers and dates.\n\nThe ten we lost on? They were well-written, comprehensive, and completely unstructured for machine parsing. Long prose. No tables. No clear \"according to X\" statements. Just... good content, by 2022 standards.\n\nI ran the same queries through a GEO Audit Tool to see if its scoring matched my manual findings. It flagged the same gap: our pages scored high on \"depth\" and low on \"structured extractability.\" Same conclusion, faster diagnosis.\n\n## What I changed (and what happened)\n\nI didn't rewrite everything. I picked the three highest-traffic posts that were getting zero AI citations and made targeted changes:\n\n1. Added explicit data attribution\nBefore: \"Most tools show around 95% accuracy.\"\nAfter: \"In our testing of 500 samples (March 2025), Originality.ai reported 94.2% accuracy, Copyscape reported 91.7%, and GPTZero reported 88.3%.\"\n\n2. Restructured comparison sections as tables\nI had a 600-word narrative comparing five tools. I condensed it to a table with columns: tool, accuracy rate, false positive rate, price, and best use case. The narrative still exists below for humans.\n\n3. Added \"Key Finding\" callouts at the top of each section\nNot TL;DR summaries — actual findings stated as single sentences. \"GPTZero's accuracy drops to 79% on non-English text.\" That kind of thing.\n\n## The results after three weeks\n\nI know, three weeks is nothing. But:\n- Perplexity now cites us on 2 of the 3 updated queries (up from 0)\n- Google's AI Overview pulls from one of the updated posts for the first time\n- Organic traffic is flat, so I didn't break anything\n\nThe third query still doesn't cite us — it cites a Reddit thread from 2023. Some battles you don't win.\n\n## The uncomfortable realization\n\nI've been optimizing for human readers who skim and Google bots that parse. AI engines are a third audience with different needs. They want:\n- Specific numbers with clear sourcing\n- Structured data they can extract without interpretation\n- Attribution chains they can verify\n\nThis isn't about gaming the system. It's about the fact that GEO vs SEO aren't the same discipline — they overlap, but the ranking signals diverge. Classic SEO rewards dwell time and backlinks. GEO rewards extractability and verifiability.\n\nI'm not abandoning SEO. I'm adding a GEO layer on top. For any page targeting a query where AI engines show results (check your AI Gravity Checker scores), I now run the \"would a machine cite this?\" test before publishing.\n\n## What I'd do differently\n\nIf I were starting from scratch, I'd build content with structured data as the foundation, not the afterthought. Write the table first. Then write the prose around it. Reverse the workflow.\n\nBut for existing content? Pick your top 20 pages by organic traffic. Check which ones AI engines ignore. Update those first. Don't boil the ocean.\n\nThe gap between ranking and being cited is real. And it's widening.",
"tags": ["GEO", "AI search", "content optimization", "SEO", "Perplexity", "ChatGPT"],
"summary": "Ranked top 3 on Google but invisible to AI engines. Fixed 3 posts — here's what changed."
}