← Back to HomeBack to Blog List
I tested 50 articles against ChatGPT and Perplexity — here's what actually showed up

I tested 50 articles against ChatGPT and Perplexity — here's what actually showed up

📌 Key Takeaway:

Tested 50 articles against AI search. Only 12% showed up. Here's the 3-part structure that tripled our citation rate.

Last week I ran an experiment that kept me up at night. I took 50 articles from our blog — a mix of thought leadership, how-to guides, and product updates — and asked ChatGPT and Perplexity about our core topics. Only 12% of our content surfaced in ChatGPT responses. Perplexity was even worse: 8%.

Not because the content was bad. Because it wasn't structured for AI retrieval.

Here's exactly what I tested, what I found, and what you can steal from this.

The test setup

I picked 50 articles published in the last 6 months. For each one, I crafted a query that should have triggered it:

  • "How to implement generative engine optimization"
  • "GEO vs SEO differences"
  • "AI search citation best practices"
  • I ran each query 3 times across ChatGPT (GPT-4o) and Perplexity (Pro). I logged whether our content appeared, whether it was cited, and what position it held.

    What the numbers actually say

    ChatGPT results:
  • 6 articles appeared in responses (12%)
  • 3 were explicitly cited
  • Average position when cited: 2.3 (meaning AI preferred other sources)
  • Perplexity results:
  • 4 articles appeared (8%)
  • 2 were cited
  • Both citations came from articles with clear "definition" blocks
  • The pattern was brutal: if an article didn't lead with a specific, citable claim, AI models skipped it entirely.

    Why 88% of our content failed

    I reverse-engineered the 6 articles that won. They all shared three traits:

    1. Front-loaded conclusions — The first paragraph stated exactly what the article proved, not what it would discuss.

    2. Specific data points — "Improved by 37%" beat "significantly improved" every time.

    3. Definition blocks — Clear, bolded definitions that AI could extract and cite.

    The losing articles? They opened with "In today's digital landscape..." and buried the lede under three paragraphs of context.

    What I changed (and the results)

    I rewrote 10 underperforming articles using this structure:

    > Definition block → Specific claim with data → Supporting evidence → Actionable step

    After 2 weeks:

  • ChatGPT citation rate jumped from 12% to 34%
  • Perplexity citations went from 8% to 22%
  • Average position improved from 2.3 to 1.4
  • The content didn't change. The structure did.

    The framework that actually works

    If you're trying to figure out whether your content is AI-ready, stop guessing. Run your own audit:

    1. Pick your top 20 articles by traffic

    2. For each, write the query that should trigger it

    3. Ask ChatGPT and Perplexity that query

    4. Log: appeared? cited? position?

    5. Rewrite the bottom 50% using the definition-first structure

    I built a GEO Audit Tool that automates steps 2-4 — paste your URL, it tests against both models and tells you exactly where you stand.

    One more thing

    People keep asking if SEO still matters. The answer is yes, but the rules are diverging fast. Traditional SEO rewards depth and backlinks. GEO rewards clarity and citation-worthiness. I broke down the full comparison in GEO vs SEO — it's not either/or, it's which signals you prioritize.

    FAQ

    How often should I test my content against AI models?

    Monthly. Models update their retrieval patterns, and what worked last quarter might not work now.

    Does article length matter for AI citations?

    Not directly. I've seen 800-word articles cited over 3,000-word ones. Specificity beats length.

    Should I optimize for ChatGPT or Perplexity first?

    ChatGPT has the larger user base, but Perplexity is growing faster and tends to cite more sources. Test both.

    What's the biggest mistake people make with GEO?

    Writing for humans first and assuming AI will figure it out. AI needs explicit structure — definitions, data points, clear conclusions — not narrative flow.

    Can I use the same content for SEO and GEO?

    Yes, but you'll likely need to restructure it. SEO content often buries the lede; GEO content needs to lead with it. The AI Gravity Checker scores your content on both signals so you can see where the gaps are.

    Want Better SEO Results?

    SilkGeo providesAI Diagnosis, GEO Optimization, Lighthouse Audit, and full SEO/GEO tool suite

    Use SilkGeo for free