{
"title": "I tested 50 product pages against ChatGPT and the results broke my mental model of SEO",
"content": "Last week, I ran a test that made me rethink everything I thought I knew about GEO.\n\nI took 50 product pages from a B2B SaaS client — all had decent traditional SEO: proper H1s, meta descriptions, backlinks, the works. Then I crafted 10 buyer-intent queries like \"best project management software for remote teams\" and \"how to choose a CRM for small business.\"\n\nI ran each query through ChatGPT, Perplexity, and Google's AI Overviews. I tracked which pages got cited, which got paraphrased without citation, and which got completely ignored.\n\nThe Setup\n\nThe pages spanned three categories:\n- 20 feature pages (describing product capabilities)\n- 15 comparison pages (vs competitors)\n- 15 use case pages (specific industry applications)\n\nAll had been optimized for traditional search. Average word count: 1,200. Average DA: 38. Average backlinks: 47.\n\nThe Pattern That Emerged\n\nOf the 50 pages, only 14 got cited at least once across all three engines. Here's what separated them:\n\n1. Specific statistics beat generic claims. Pages with concrete numbers (\"reduces onboarding time by 3.2 days on average\") were cited 4x more often than pages with vague claims (\"streamlines onboarding\"). This was the single biggest differentiator.\n\n2. Structured data mattered more than I expected. Pages with schema markup for FAQ, HowTo, or Product got picked up 60% more often. The AI engines are clearly using structured data as a signal for parseable content.\n\n3. Traditional DA was nearly irrelevant. The cited pages had an average DA of 34. The uncited pages averaged DA 41. I double-checked this because it seemed wrong, but the pattern held.\n\nEngine Differences\n\nPerplexity cited the most pages (11 of 14), but also paraphrased without attribution more often. ChatGPT was more conservative — only cited 8 pages, but always with clear attribution. Google's AI Overviews were the pickiest, citing just 6 pages, all of which had specific statistics and FAQ schema.\n\nWhat I Changed\n\nI took 5 of the uncited pages and made three specific changes:\n- Added concrete statistics from customer data (even if rough estimates)\n- Implemented FAQ schema with direct, concise answers\n- Restructured the intro to lead with the conclusion, not the problem\n\nAfter two weeks, 4 of the 5 pages started appearing in AI citations for at least one query. The fifth page was a feature page with no customer data to leverage — it's still not cited.\n\nThe Bigger Picture\n\nMost GEO advice I see online treats it like SEO 2.0 — same tactics, different search box. That's wrong.\n\nAI engines don't rank pages the way Google does. They're looking for:\n- Specific, verifiable information they can quote\n- Clear structure they can parse without hallucinating\n- Direct answers to implicit questions in the query\n\nIf you're optimizing for AI search, stop thinking about keywords and start thinking about what an AI would need to construct a confident answer. What data points would it need? What structure would make extraction easy?\n\nThe Tools\n\nI used GEO Audit Tool to baseline the pages before making changes. It specifically checks for AI-readiness factors like structured data completeness and content specificity — things traditional SEO tools completely miss.\n\nFor tracking visibility across AI engines, AI Gravity Checker has become my go-to. It shows you exactly which queries trigger citations for your pages, so you can see if your changes actually moved the needle.\n\nIf you're still treating GEO and SEO as the same thing, this breakdown of GEO vs SEO is worth reading. The overlap is smaller than you think, and the tactics that work for one can actually hurt the other.\n\nWhat I'm Testing Next\n\nI'm currently running a follow-up test with 100 pages across three industries to see if the pattern holds for ecommerce and local service businesses. Early results suggest the \"specific statistics\" effect is even stronger in YMYL niches — which makes sense, given the higher accuracy bar.\n\nI'll share the full dataset when it's complete. For now, if you're doing GEO work, start by auditing your pages for specificity. That alone might move the needle more than any other single change.",
"tags": ["GEO", "AI search", "SEO", "content optimization", "structured data"],
"summary": "Specific stats beat DA in AI search. 50-page test shows what actually gets cited."
}