Who Do AI Engines Cite?

the top citation sources for ChatGPT, Google, Perplexity, Gemini & Claude
~40%
of all AI citations go to Reddit
(5W Index, 680M citations)
11%
domain overlap between ChatGPT & Perplexity citations — each engine is its own game
3–6
sources per AI answer vs ~10 in a SERP — a much smaller winners' circle
Engine by Engine

Google AI Overviews

mirrors the SERP… less every month
Top sources: YouTube (~30% of AIOs!), Wikipedia, Reddit, Quora
  • Heavy bias to UGC + video — YouTube has a ~200x edge over any other video source
  • Top-10 organic → AIO citation link is weakening: ~76% in 2025, under 40% by 2026
  • Strong brand signals (branded search, mentions) matter more here than elsewhere
play: video + ranked page + entity-clean schema

ChatGPT

reference + community brain
Top sources: Wikipedia, Reddit, Forbes, G2, Medium
  • Leans encyclopedic + discussion content; loves "best of" listicles & review hubs
  • Cites brand-owned domains rarely (~0.6% of answers) — 3rd-party mentions do the work
  • Most volatile engine: citation sources reshuffle in weeks, not years
play: get talked about on Reddit, G2, press & listicles

Perplexity

the academic + news nerd
Top sources: Reddit, LinkedIn, YouTube, NIH, Wikipedia
  • Most generous brand citer (~13% of answers cite brands — 20x+ ChatGPT)
  • Favors fresh news, studies, .gov/.edu & company pages over creator content
  • LinkedIn company pages punch way above their weight here
play: publish original data, keep company pages sharp

Gemini / AI Mode

keeps it in the Google family
Top sources: YouTube, Google properties, Reddit, LinkedIn, Facebook
  • Clear preference for Google-owned surfaces — Maps, YouTube, Business Profiles
  • UGC domains gained most after the Sept 2025 ranking shakeups
  • AI Mode & AI Overviews cite the same URLs only ~14% of the time — treat them separately
play: YouTube presence + fully built-out GBP

Claude

the careful librarian
Top sources: Wikipedia, Reddit, docs & reference sites, news wires
  • Patterns track ChatGPT's reference-heavy profile in cross-engine studies
  • Rewards semantically complete, well-structured documentation (think Zapier-style pages)
  • Smaller search volume today, but included in every major 2026 citation index
play: deep, self-contained reference pages win

⭐ The cross-engine 5

cited (almost) everywhere
Reddit · Wikipedia · YouTube · LinkedIn · Forbes/news
  • Top 15 domains absorb ~68% of the AI answer pipeline
  • Journalism = ~27% of all citations, ~49% on time-sensitive queries
  • But no single domain exceeds ~5% on any one platform — it's a long tail with fat outliers
presence on 4+ trusted platforms ≈ 2.8x more ChatGPT appearances

Vertical winners differ

SaaS / softwareG2, Reddit, Capterra, vendor docs
HealthNIH, .gov, major hospital systems
FinanceBloomberg, SEC, Investopedia, NerdWallet
Local / servicesGBP, Yelp, Reddit threads, local news
Media / techNYT, Reuters, The Verge, TechRadar
find YOUR vertical's citation set before writing anything

What actually gets a page cited 📌

  • Quotable data points, definitions & extractable tables beat raw domain authority
  • ~44% of LLM citations come from the first 30% of the text — front-load the answer
  • Depth, readability & freshness outrank backlinks as citation signals
  • Structured content (headings, lists, FAQ) is the most effective format
  • Optimized review profiles (Trustpilot, G2) drive co-mentions when buyers ask about competitors

The borrowed-authority playbook

  1. Map your vertical's citation set — run your money questions through each engine, log every cited domain
  2. Win your own pages — answer-first, structured, front-loaded, fresh (see the AEO process ↓)
  3. Get onto the cited 3rd parties — Reddit presence, listicle inclusions, G2/review profiles, PR & news wires, LinkedIn company page, YouTube
  4. Track citation share monthly — same question set, every engine, trended; volatility is measured in weeks
THE BIG IDEA

You can't rank your way into AI answers from your domain alone.
Be present where the engines already look — and make every page of yours the easiest thing in the index to quote.

Synthesized from 2025–2026 citation studies: 5W AI Platform Citation Source Index (680M citations), Profound, Semrush (325K prompts), Peec AI (30M sources), Lantern (200M citations), Ahrefs & Seer Interactive. Shares vary by study & month — treat as direction, not gospel.