DEV Community

Mikhail Shadrin
Mikhail Shadrin

Posted on

We Checked Whether On-Site SEO Predicts AI Citations. The Data Says Mostly No.

#seo #ai #opensource #datascience

Every GEO ("generative engine optimization") tool, including ours until
recently, sells some version of the same pitch: fix your robots.txt, add
Schema.org markup, write FAQ schema, and AI engines will cite you more.

We build one of these tools — Causabi scans sites for AI-crawler readiness
and generates fix files (robots.txt, llms.txt, JSON-LD, FAQ blocks). As part
of validating our own scoring weights, we ran the numbers on whether the
score actually predicts getting cited. Short version: it mostly doesn't,
once brand prominence is in the picture.

What we measured

We scored 44 domains on a 6-category on-site readiness algorithm:

  • robots.txt (AI bots allowed or blocked)
  • Schema.org (Organization/LocalBusiness JSON-LD completeness)
  • FAQ schema (FAQPage markup, 3+ entries)
  • content depth/structure
  • brand/NAP signals
  • freshness (dateModified, recency)

Then we checked how often each domain actually got cited by an AI engine
(Claude, via its web-search tool, one measurement window, a fixed prompt set
per domain).

What we found

We ran the check twice, and I'll give you both runs because the difference
between them is itself informative:

  • Run 1 (July 2): Claude only, n=44 usable domains. Score vs. citation rate: Pearson r = -0.078, Spearman ρ = -0.028. 86% of domains got zero citations regardless of score.
  • Run 2 (July 12): Claude + Gemini, n=41 usable. Spearman ρ = +0.084 (p = 0.60), Pearson r = +0.148 (p = 0.35). Zero-citation share dropped to 58.5% — Gemini names domains and brands much more freely than Claude.

Note the sign flipped between runs (-0.03 → +0.08). At these sample sizes
and p-values that's noise, and that's exactly the point: there is no
statistically significant relationship between on-site readiness score and
citation rate in either direction. If the correlation were real and strong,
two runs ten days apart wouldn't disagree on the sign.

  • The domains that did get cited clustered almost entirely by brand prominence — well-known domains got cited at a noticeably higher rate (~0.16 of prompts in run 1) than everyone else (~0 for the rest of the sample), regardless of how well-optimized their markup was.

Why I'm not overselling this

n=41-44 is small. This is an internal validation exercise for our own
product, not a peer-reviewed study, and I don't want it read as one.
"No significant correlation found" is not the same claim as "we proved
there is no relationship" — at this sample size we can't prove a negative.
Specific caveats:

  • Two engines so far (Claude, Gemini). Citation behavior differs meaningfully across ChatGPT, Grok, and Perplexity — we haven't run the same check across all of them yet.
  • One time window, no longitudinal before/after. We didn't take a domain, improve its score, and watch citations change over months. That's the actually convincing experiment and we haven't run it yet.
  • Prompt-domain matching wasn't blind. Some prompts were picked because a domain plausibly related to that topic, which likely biases toward domains that would get mentioned anyway.
  • "Brand prominence" is a fuzzy variable that probably absorbs some real content-quality signal we're not capturing separately. We can't fully rule out that what looks like "brand wins" is partly "genuinely better/more authoritative content wins," which on-site markup scoring doesn't measure.

What we still think is true, with more confidence

Some things aren't correlational guesses — they're closer to mechanical
facts:

  • robots.txt blocking is binary. If GPTBot, ClaudeBot, or similar are disallowed, that engine cites you zero times, by construction. About 89% of sites we've scanned block at least one AI crawler by default, usually by accident (a blanket Disallow: / that predates AI bots existing).
  • FAQ schema changes extraction, not inclusion. For content that's already in an engine's consideration set, structuring it as self-contained Q&A chunks seems to affect whether it gets pulled into a RAG-style citation — this lines up with published research on chunking behavior. But that's a "how you're cited" lever, not a "whether you're cited" lever.

Where that leaves the product

We're rewriting our own copy to say what the score actually measures:
AI-crawler readiness and machine-readability, not citation probability. No
tool — ours included — can promise the second one. If your on-site work is
mostly aimed at "getting cited more," the more binding constraint for most
sites is probably brand/mentions elsewhere, not another Schema.org type.

The scoring engine and fix generator are open source (MIT) if you want to
see the logic or run it on your own site without touching our SaaS:

pip install causabi-geo
geo-optimizer analyze https://yourdomain.com
geo-optimizer fix https://yourdomain.com --output ./out
Enter fullscreen mode Exit fullscreen mode

Repo: https://github.com/SHADRINMMM/causabi-geo
Site (hosted version + monitoring): https://causabi.com

Top comments (0)