AI Citability Scoring: Make Your Pages Quotable
Score any page 0-100 on how likely ChatGPT, Perplexity, and Gemini are to quote it, with block-level rewrites for the passages that score worst
AI engines do not cite pages. They cite passages. A page can rank first on Google and still get skipped because the answer is buried in paragraph four and every sentence starts with 'it' instead of naming the subject.
Who it's for: SEO consultants, content marketers, technical writers, documentation owners, agency teams selling GEO audits, publishers chasing AI referral traffic
Example
"Score https://acme.com/guide for AI citability" → A GEO-CITABILITY-SCORE.md with an overall score out of 100, a weighted breakdown across five categories, the three strongest and three weakest content blocks named by heading, and rewritten opening sentences for every block scoring below 60
CLAUDE.md Template
New here? 3-minute setup guide → | Already set up? Copy the template below.
npx ccplaybooks add geo-citability# AI Citability Scoring
Score a web page on how likely AI systems (ChatGPT, Claude, Perplexity, Gemini, Bing Copilot) are to quote it as a source, then produce specific rewrites that raise the score.
## Core Insight
AI language models cite passages that meet specific structural criteria. Research from Princeton, Georgia Tech, and IIT Delhi (2024) found GEO-optimized content achieves 30-115% higher visibility in AI-generated responses. The key finding: AI systems preferentially extract passages that are **134-167 words long**, **self-contained** (understandable without surrounding context), **fact-rich** (specific statistics, dates, named entities), and **answer a question in the first 1-2 sentences**.
This differs from traditional SEO copywriting, which optimizes for keyword density and engagement metrics. Citability optimizes for **extractability**: how easily an AI system can pull a passage out of your page and present it as a direct answer.
---
Get new playbooks like this one
One email a week with new Claude Code workflows. Free, like everything here.
No spam. Unsubscribe anytime.
What This Does
Scores a page on extractability: how easily an AI system can pull a passage out and present it as a direct answer. Claude segments the page at each H2 and H3, scores every block on five weighted categories, averages them into a page score, and writes specific rewrites for the blocks that drag the average down.
This is the structural half of GEO. It does not judge whether your content is good, it judges whether it is quotable. geo-content covers the other half, scoring the same page on E-E-A-T to see whether an engine has reason to trust the source at all. geo-platform-optimizer goes wider, tuning a whole site per engine.
From Zubair Trabzada's geo-seo-claude toolkit, packaged as a single CLAUDE.md.
Quick Start
Step 1: Create a Project Folder
mkdir -p ~/Documents/GEOCitability
Step 2: Download the Template
Click Download above, then:
mv ~/Downloads/CLAUDE.md ~/Documents/GEOCitability/
Step 3: Start Working
cd ~/Documents/GEOCitability
claude
Give Claude a URL: "Score https://example.com/guide for AI citability." It fetches the page, segments it, scores each block, and writes GEO-CITABILITY-SCORE.md.
The Five Categories
| Category | Weight | What it measures |
|---|---|---|
| Answer Block Quality | 30% | Does each section open with a 1-2 sentence direct answer? |
| Passage Self-Containment | 25% | Can a block be understood in isolation, with its subject named? |
| Structural Readability | 20% | Clean heading hierarchy, short paragraphs, tables, lists |
| Statistical Density | 15% | Specific numbers with named sources, not "many" and "most" |
| Uniqueness & Original Data | 10% | Information not available on a higher-authority page |
Block score = (Answer × 0.30) + (SelfContain × 0.25) + (Structure × 0.20) + (Stats × 0.15) + (Unique × 0.10). The page score is the average across blocks, plus a "citability coverage" figure: the percentage of blocks scoring above 70.
What the Research Says
The scoring is built on published GEO findings, which the template carries as a reference table:
| Finding | Source |
|---|---|
| Optimal cited passage length: 134-167 words | Bortolato 2025, AI Overview passage analysis |
| Definition patterns increase citation rate 2.1x | Georgia Tech 2024 |
| Adding statistics increases citation by 40% | Princeton GEO study 2024 |
| Authority quotations increase citation up to 115% in some categories | IIT Delhi 2024 |
| Content with source citations is cited 20-25% more often | Perplexity and ChatGPT search |
Self-Containment Checklist
Applied to every passage. Five questions, and most pages fail on the first:
- Does it explicitly name the subject, rather than "it," "this," or "they"?
- Can someone understand the main point reading only this passage?
- Does it contain at least one specific fact, statistic, or named entity?
- Is it between 50 and 200 words?
- Does it avoid opening with "But," "However," or "And," which imply prior context?
Examples
The template carries a matched pair so Claude has a target, not an abstraction.
High citability, 58 words:
"Content delivery networks (CDNs) are distributed server systems that cache and serve web content from locations geographically close to end users. A CDN reduces latency by 50-70% on average by serving assets from edge servers rather than a single origin server."
Names its subject, opens with a definition, carries three specific data points.
Low citability, 52 words:
"If you've ever wondered why some websites load faster than others, the answer might surprise you. There's this amazing technology that has been around for a while now."
Similar length, but the topic is never named, there are zero facts, and no sentence works pulled out of context.
Per-Engine Preferences
The template includes what each system favors, which shapes the rewrite advice:
| Engine | Preference |
|---|---|
| ChatGPT Search | Explicit definitions, named sources, recent dates. Cites 2-4 sources. |
| Perplexity | Fact-dense passages with statistics. Cites 4-8 sources. Values recency. |
| Claude | Well-structured, comprehensive passages. Nuance over brevity. |
| Gemini / AI Overviews | Concise 40-60 word answer blocks. Favors existing top-10 organic results. |
| Copilot | High-authority domains with clear factual claims. |
For a full per-engine audit rather than a passage-level one, see geo-platform-optimizer.
Tips
- Run it on your best-performing page first. A page already getting search traffic with a low citability score is the highest-leverage rewrite you have.
- Fix Answer Block Quality before anything else. It carries 30% of the score and usually means moving one sentence to the top of a section.
- Question-based H2s ("What is X?", "How does X work?") map directly to AI queries and lift both the answer and structure scores.
- Convert any comparison of three or more things into a table. AI systems extract table data with high accuracy, and it is the fastest structural win available.
- Keep both score files after a rewrite. The per-section score table makes it obvious which edits worked.
Limitations
- Scoring is judgment-based. Expect a few points of variance between runs on the same page, and treat the block ranking as more reliable than the absolute number.
- Citability is one input among several. A perfectly structured page from a domain with no authority signals still may not get cited. Pair it with geo-content.
- The rubric is tuned for informational and comparison content. Narrative essays, landing pages, and creative writing score low by design and should not be optimized against it.
- "Expected citability lift" figures in the output are estimates, not measurements.
- The full geo-seo-claude repo runs this scoring through a Python engine with page-fetch tooling. This template reproduces the rubric and procedure for Claude to apply directly.