Perplexity Cites 2.8x More Sources Than ChatGPT, Only OpenAI Depends on Wikipedia

Analysis of 118K+ answers reveals dramatic differences in AI citation behavior: Perplexity averages 21.87 citations per question while ChatGPT uses 7.92, and OpenAI is the only model citing Wikipedia significantly at 4.8%.

Research

We analyzed 118,101 AI-generated answers with 669,065 citations across 8 major providers during Q3 2025. The data reveals dramatic differences in how each platform selects, cites, and positions sources.

What works for ChatGPT will fail for Perplexity. Wikipedia strategy succeeds on OpenAI, fails everywhere else. Platform-specific optimization isn't optional - it's essential. Our findings align with Princeton's GEO research showing platform-specific optimization is essential, and Search Engine Land's analysis confirming Wikipedia dominates ChatGPT citations.

This report quantifies exactly how different these providers are, and what that means for your GEO strategy.

The Citation Volume Spectrum: Not All AI Providers Cite Equally

Here's how much each provider relies on external sources:

The spread is massive: Perplexity cites 2.76x more sources per question than ChatGPT (21.87 vs 7.92).

What this means: A brand that appears in 10% of Perplexity responses might appear in only 3.6% of ChatGPT responses - not because their content is worse, but because the platforms have fundamentally different citation strategies.

Understanding the Metrics

Citations per question: How many sources the AI model references per answer. Higher numbers indicate the model relies more heavily on external sources.

Unique Sources: Total distinct domains cited during the quarter. Indicates breadth of source discovery.

Diversity Score: Shannon entropy normalized to 0-100. Higher scores mean citations are distributed across many sources rather than concentrated on a few.

Average Position: Mean placement of citations in responses. Lower numbers (early positioning) typically correlate with higher authority.

The Wikipedia Dependency: Only One Provider Relies on It

Here's where conventional GEO wisdom breaks down completely.

Wikipedia Citation Rates by Provider:

ChatGPT is 4.8% dependent on Wikipedia. Every other major provider: effectively 0%.

Why this is critical: The standard GEO advice - "get a Wikipedia page for AI visibility" - only works for one platform. If your target audience uses Perplexity, Gemini, or Google AI, Wikipedia presence provides minimal direct benefit.

What OpenAI Does Differently

OpenAI's Top Cited Sources:

1. Wikipedia (4.8%)
2. Google.com (various properties)
3. Le Monde (major news)

Everyone else's Top Cited Sources:

1. Specialized business/legal platforms
2. Industry-specific content sites
3. Regional authorities

The pattern: OpenAI uses Wikipedia as a knowledge foundation, then cites specialized sources. Other providers skip the encyclopedia layer entirely and go straight to specialized content.

Strategic Implication: If you're optimizing for ChatGPT specifically, Wikipedia presence is valuable. For multi-provider GEO, specialized vertical content is non-negotiable.

Provider-by-Provider Deep Dive

Perplexity: The Citation Champion (21.87 cit/question)

Profile:

• Total Citations: 276,139 across 12,623 questions
• Unique Sources: 37,399 domains
• Diversity Score: 79.8 (diverse, but less than Google)
• Average Position: 5.66 (mid-response)
• Top Sources: Specialized legal platforms, government portals

What Perplexity prioritizes:

• Specialized business and legal content
• French-language vertical-specific sites
• Practical, actionable information
• Recent, updated content

Why Perplexity cites so much: The platform's value proposition is comprehensive research. More citations = more thorough answers = differentiation from ChatGPT's concise responses.

Optimization Strategy for Perplexity:

1. Create deep vertical content: Not 500-word blog posts, but comprehensive 3,000+ word guides
2. Structure for scanning: Perplexity pulls specific sections, so make each section independently valuable
3. Update frequently: Perplexity appears to favor recently updated content
4. Don't rely on Wikipedia: 0% Wikipedia share means this tactic is wasted effort

ChatGPT (OpenAI): The Wikipedia Dependent (7.92 cit/question)

Profile:

• Total Citations: 190,894 across 24,099 questions
• Unique Sources: 42,592 domains (highest breadth)
• Diversity Score: 83.4
• Average Position: 2.82 (very early)
• Top Sources: Wikipedia, Google.com, Le Monde

What ChatGPT prioritizes:

• Encyclopedic sources (Wikipedia: 4.8%)
• Mainstream authoritative sites
• News outlets with editorial standards
• .edu domains

Why ChatGPT cites less: The model is trained to provide concise answers drawing primarily from training data, using citations mainly for verification and recent information.

Optimization Strategy for ChatGPT:

1. Wikipedia presence is valuable - 4.8% share means thousands of potential citations
2. Aim for early positioning - Average position 2.82 means top-3 sources dominate
3. Mainstream credibility matters - News outlet coverage helps
4. Don't compete with specialized sites - ChatGPT prefers broad authority over niche expertise

Google AI Overview: The Diverse Giant (17.93 cit/question)

Profile:

• Total Citations: 79,565 across 4,439 questions
• Unique Sources: 25,785 domains
• Diversity Score: 89.4 (highest)
• Average Position: 10.08 (latest)
• Top Sources: Google.com, YouTube, LinkedIn

What Google AI Overview prioritizes:

• Google properties (own platforms)
• Video content (YouTube)
• Social platforms (LinkedIn)
• Diverse, multi-perspective sources

Optimization Strategy for Google AI Overview:

1. YouTube content is essential - Video platform is a top-3 source
2. Expect late positioning - Average 10.08 means citations appear throughout response
3. Don't expect dominance - High diversity means no single source gets significant share
4. Google properties matter - Citations from Google ecosystem get preferential treatment

Gemini: The Regional Specialist (17.11 cit/question)

Profile:

• Total Citations: 122,251 across 7,145 questions
• Unique Sources: 20,549 domains
• Diversity Score: 77.0
• Average Position: 4.67 (early)
• Top Sources: French business blogs, specialized legal platforms

What Gemini prioritizes:

• Regional, French-language content
• Business and entrepreneurship resources
• Local service providers
• Practical implementation guides

Microsoft Copilot: The Selective Curator (2.47 cit/question)

Profile:

• Total Citations: 148 across 60 questions
• Unique Sources: 111 domains
• Diversity Score: 97.0 (highest possible)
• Average Position: 1.90 (earliest)
• Top Sources: PCMag, Forbes, Rolling Stone

Multi-Provider Strategy: The Overlap

Most brands need visibility across multiple providers. Here's what works everywhere: Universal Tactics:

1. Comprehensive, structured content - Works for all providers
2. Niche authority - Specialized expertise beats generic content everywhere
3. Regular updates - All providers favor fresh information
4. Clear, scannable formatting - Essential for AI parsing

Provider-Specific Additions:

• ChatGPT: Add Wikipedia
• Perplexity: Add depth (3,000+ words)
• Google AI Overview: Add YouTube
• Gemini: Add localization
• Microsoft Copilot: Add press coverage

Key Takeaway: One Size Does NOT Fit All

The data is unambiguous:

• Wikipedia strategy works for ChatGPT only (4.8% vs 0% elsewhere)
• Citation volume strategies work for Perplexity (21.87 cit/question) but not Copilot (2.47)
• YouTube optimization matters for Google but not ChatGPT
• Regional content wins on Gemini but underperforms on ChatGPT

The winning approach: Platform-specific optimization based on where your audience actually uses AI, measured with provider-specific metrics.