{
"title": "Why my best-ranking pages got zero AI citations (and what I changed)",
"content": "Last Tuesday I ran a crawl of our top 50 pages by organic traffic. Every single one ranked on page 1 for their target keyword. Then I checked how many showed up as citations in Perplexity and ChatGPT for the same queries.\n\nThree. Out of fifty.\n\nThat gap between traditional rankings and AI visibility is the thing nobody's talking about enough. I've been spending the last six weeks fixing it, and the pattern is becoming clear.\n\n## The problem isn't content quality — it's structure\n\nOur pages were written for humans scanning a SERP. Clear headings, decent intro, some bullet points. That's not enough anymore.\n\nAI engines don't \"read\" pages the way Google's crawler indexes them. They're looking for discrete, self-contained answers they can extract and attribute. When our content buried the actual answer three paragraphs down after a brand story, the AI just skipped us entirely.\n\n> Generative Engine Optimization is the practice of structuring content so AI systems can identify, extract, and cite specific claims — not just index keywords.\n\n## What actually moved the nail\n\nI picked 10 pages and restructured them with three rules:\n\n1. Lead with the answer, not the narrative\n\nOld format: Brand story → Problem agitation → Solution → Answer\nNew format: Direct answer in the first 40 words → Supporting context → Deeper detail\n\nThe first paragraph now reads like a featured snippet on steroids. No throat-clearing.\n\n2. Use explicit attribution for every data point\n\n\"According to our Q3 2025 survey of 1,200 marketers\" hits different than \"studies show.\" I went through and added source attribution to every stat, every claim, every number. Even when the source was our own data.\n\nAI engines weight attributed claims significantly higher than unsupported statements. This makes sense — they're trying to minimize hallucination risk when they pull from your page.\n\n3. Break long answers into discrete Q&A blocks\n\nInstead of one 800-word section on \"how to do X,\" I split it into 4-5 question-headed blocks, each answering one specific sub-question. Each block is 60-120 words.\n\nThis mirrors the format AI engines use when they construct responses. You're essentially pre-packaging your content in the shape they want to serve.\n\n## The results after four weeks\n\nOf the 10 restructured pages:\n- 7 now appear as citations in Perplexity for their primary query\n- 5 show up in ChatGPT responses\n- 4 appear in Google's AI Overviews\n\nThe other 30 unchanged pages? Still sitting at 3 total citations.\n\nTraffic from AI referrals is still small — roughly 4% of what we get from traditional organic. But it's growing 15-20% week over week, and the conversion rate is notably higher. These visitors are arriving with more context and stronger intent.\n\n## The uncomfortable part\n\nThis means the playbook is splitting. You can't just write one version of a page and expect it to win everywhere. The GEO vs SEO comparison I keep coming back to is this: SEO is about matching keywords to queries, GEO is about matching structured knowledge to reasoning chains.\n\nThey overlap, but they're not the same muscle.\n\nI've started running every new piece through a GEO Audit Tool before publishing — not because it gives me a score to chase, but because it surfaces the structural gaps I'd otherwise miss. Things like \"this claim has no source\" or \"this answer is buried below three subheadings.\"\n\nIf you're sitting on great content that ranks well but isn't showing up in AI results, the issue is almost certainly structural, not qualitative. The fix is tedious but straightforward.\n\nI'm tracking all of this in a running doc. Happy to share the full framework if anyone wants it.",
"tags": ["GEO", "AI search", "content strategy", "generative engine optimization", "SEO"],
"summary": "Ranked pages getting zero AI citations? The fix is structural, not qualitative. Here's what changed after restructuring 10 pages for generative engines."
}