AI Citation Statistics 2026

How many sources do AI engines cite?

• ChatGPT (GPT-5.5 with browsing): 3.4 average sources per response (OpenAI, 2025).
• Perplexity: 4.1 average sources per response (Perplexity public, 2025).
• Google AI Overviews: 4.2 average citations per Overview, down from 7.8 in early 2024 (SE Ranking, 2025).
• Claude (with web search): 3.8 average sources per response (Anthropic, 2025).
• The trend across all engines: citation slots are decreasing as confidence in fewer sources increases.

Which domains get cited most?

• Reddit is the #1 cited domain in ChatGPT for product-research queries (Profound, 2025 sample of 50K+ queries).
• Reddit citations grew 4.7x in 2025 after Google's Reddit content partnership and OpenAI's Reddit data deal.
• Wikipedia, YouTube, Reddit, and G2 collectively account for 38% of buyer-intent ChatGPT citations (BrandRank.ai, 2025).
• Owned brand sites account for 24% of citations on buyer-intent queries.
• The remaining 38% comes from review sites (Capterra, Trustpilot), industry blogs, and forums.

What types of pages get cited?

• Pages with FAQ schema are 3.4x more likely to be cited in AI Overviews than equivalent pages without (roast.page, 2026).
• Pages with HowTo schema are 2.8x more likely to be cited on "how do I X" queries (roast.page, 2026).
• Pages with valid llms.txt are cited 2.1x more often on average than equivalent pages without (early data, small sample).
• Pages under 800 words on commercial keywords are 4x less likely to be cited (Google's March 2026 spam update specifically demoted thin pages).
• Pages with direct, specific answers in the first 100 words are cited 5x more often than pages that bury the answer (roast.page analysis, 2026).

How do brand mentions affect citation rate?

• Brand mentions on third-party sites correlate r=0.62 with AI citation rate — substantially higher than backlink correlation (r=0.31) (Profound, 2025).
• Mentions on review sites (G2, Capterra, ProductHunt) have the highest correlation: r=0.71 with citation rate.
• Reddit/Hacker News mentions correlate r=0.58 with citation rate, particularly strong for technical categories.
• Each new G2 review increases citation rate ~0.4% on average (Profound, 2025).
• Brands appearing in 5+ "best of" lists are cited 3.2x more often than brands appearing in 0-1 lists (BrandRank, 2025).

Citation accuracy — how often do AI engines get it right?

• 71% citation accuracy across major AI engines (Stanford HAI, 2025) — up from 51% in 2023.
• 29% of cited claims are partially or fully wrong (Stanford HAI, 2025).
• ChatGPT accuracy: 73%, Claude: 76%, Perplexity: 68%, Google AI Mode: 79% (Stanford HAI, 2025).
• Pricing-related claims are the most frequently wrong category (44% inaccurate when stated specifically — often citing outdated pricing).
• Compliance and security claims are most accurate (87%) because they're more often pulled from structured data.

Citation share by category

• Technical/developer tools: 60% of buyer-intent queries are now AI-search-driven (Pew, 2025).
• SaaS/B2B: 15-30% of buyer-research queries on AI engines first (Gartner, 2025).
• Ecommerce/consumer: 5-10% AI-driven currently, growing fast.
• Local services: Under 5% (still dominated by Google Maps and classical SEO).
• The 70/20 gap between awareness and implementation (Edelman, 2025) is the single largest unsold opportunity in marketing tech.