Free GEO Audit

Citation Score: the method, the data, and what it doesn't tell you

Anyone can run this measurement and get the same number. That is the entire reason it is published. Here is how it is calculated, the full 1,205-page data set behind it, and the six things it cannot tell you.

Citation Score is a 0 to 100 measure of whether AI assistants name your business when customers ask your category's buying questions. Ten fixed buyer-intent queries, four assistants, scored monthly. 10 points if you are cited as a source, 5 if mentioned, 0 if absent.

What it is

A number tied to real customer behavior

Citation Score measures the percentage of 10 real buyer-intent queries in your category where AI assistants (ChatGPT, Perplexity, Claude, Gemini) mention or cite your business. On a scale of 0–100.

For example, if you sell e-commerce accounting software: the 10 queries might be "best accounting for Shopify", "Shopify accounting tools reviews", "alternatives to FreshBooks". The exact questions your customers type when shopping.

How it's calculated

Points per query, tracked month to month

Cited as a source (in citations list)10 points
Mentioned by AI (in response text)5 points
Not mentioned or absent0 points
Per 10 queries, max score100 points

Every month we run the same 10 queries and score them the same way. Progress is a number both you and I can see. No ambiguity.

A score from a single engine is not a Citation Score. Their source pools barely overlap, so one engine describes roughly one eleventh of the picture.

Why this metric

Reproducible. Real. Impossible to fake.

Real buyer behavior: These 10 queries are what customers actually type when shopping in your category. Not vanity metrics. Not SEO rankings. Whether AI names you when a customer asks matters.

Reproducible: Anyone can run the same queries and get the same score. Not an algorithm someone else controls. Not a proprietary "visibility index." Just queries and AI responses, the same every month.

Can't be gamed: You can't "optimize for Citation Score" with black-hat tactics. The only way to improve is to actually become quotable: better pages, real citations, trusted positioning.

The guarantee

Why it's the core metric of the sprint

The 30-day SEO + AI visibility sprint comes with a written guarantee: if your Citation Score has not grown within 60 days of the sprint start, Nataliya keeps working free until it does.

That's why we measure monthly against the same 10 queries. It's the only way to make the guarantee credible, and the only reason Citation Score matters.

What the page scan measures

Separately from the brand-level score, the scanner checks each page against six signals. These are page properties, not opinions, and each is machine-checkable.

SignalWhat is checked
ExtractabilityWhether meaningful text comes out of the HTML the server returns, without executing JavaScript
DateWhether a publication or update date is visible in the content
AuthorWhether a named author is present, excluding "Admin" and "The Team"
FAQ schemaWhether FAQPage structured data is present and valid
Answer blockWhether a direct answer sits in the first 30% of the document
llms.txtWhether the file exists at the site root. Measured per site, not per page

The data set: 1,205 live pages

Scanned since January 2026. Share of pages failing each check.

No visible author78.8%
No date on the content77.1%
No FAQ schema62.8%
No llms.txt (per site)51.9%
No direct answer block50.2%
Thin or non-extractable35.7%

Brand level. Median Citation Score across the brands measured: 16 out of 100. Roughly half scored zero, never named as a source by any assistant on any of their own category's buying questions.

The correlation that isn't there. Failure rates did not track with traditional SEO health. Pages on sites with clean architecture, good Core Web Vitals and real backlinks failed extraction at the same rate as everything else. This is the finding we would most like someone to try to replicate or break.

How this lines up with published research

SourceSampleFinding
Kevin Indig18,012 verified ChatGPT citations from 1.2M responses44.2% of citations come from the first 30% of a page. Content buried deep in a long post is roughly 2.5x less likely to be cited
CXL100 Google AI Overview citationsSame position effect, found independently
Qwairy118,000 AI responses, Jan to Mar 2026Only 11% of cited domains appear on more than one platform. Citations per response: Perplexity 21.87, Google AI Mode 8.34, ChatGPT 7.92
Otterly1M+ citations, 202673% of sites carry technical barriers blocking AI crawlers. News and media 20 to 30% of citations, community forums 5.9 to 16.9%

Our 35.7% non-extractable and Otterly's 73% crawler-barrier figure measure different things and should not be compared directly. Theirs counts access barriers at the site level. Ours counts pages that a crawler reaches successfully and still cannot read.

Limitations

Stated plainly, because a method page without them is marketing.

Reproduce it yourself

  1. Write ten buyer-intent queries for your category. No brand names.
  2. Run each in ChatGPT, Perplexity, Claude and Gemini, in clean sessions.
  3. Score 10 for cited with a link, 5 for mentioned, 0 for absent.
  4. Divide by the maximum available and multiply by 100. Ten queries across four assistants gives 400 possible points, so 84 points is a Citation Score of 21.
  5. Repeat monthly with the same queries, in the same order.

If your number differs materially from ours on the same category, we would like to see it.

Related

The practical version

All of this as a 12-point check you can run in twenty minutes, with nothing but a browser: Why AI doesn't recommend your business

See your Citation Score in 30 seconds

The free GEO audit scans your site and calculates your instant Citation Score against your category's 10 key buyer queries, then Nataliya emails you a full human-reviewed report with every fix prioritized by impact.

Scan my site free →