gptmelo Research · Data Report · July 2026

Every AI Engine Has a Favorite Source

AI Platform Citation Patterns 2026: how ChatGPT, Google AI Overviews, Perplexity, Gemini, and Claude choose who to cite — June 2026 engine snapshots layered on a 680M-citation historical baseline, translated into a platform-by-platform GEO playbook.

Carter WangFounder, gptmelo.com
Data to Jun 2026680M+ historical citations
5 AI enginesChatGPT · AIO · Perplexity · Gemini · Claude
UngatedHTML + PDF, free
The Report in Four Numbers
29.4%
Reddit's share of ChatGPT's tracked citations in 2026 — nearly double Wikipedia's 15.0%. Community has overtaken the encyclopedia (Ahrefs, 9.6M queries)
39.2%
YouTube's share of AI citations by Feb 2026 — overtaking Reddit (20.3%) as the most-cited domain in AI search (Adweek study)
38%
of AI-cited sources rank in Google's top 10. The other 62% get cited without ranking — citations have decoupled from rankings (863K-SERP study)
92.4%
of AI referral traffic now comes from ChatGPT, after 9.9× growth in 19 months (Previsible, July 2026)
01 · The Shift

Citations Are the New Rankings

AI engines don't rank pages — they cite sources. When ChatGPT, Perplexity, or Google AI Overviews answers a buyer's question, it pulls from 3–6 sources, not 10 blue links. That citation set is narrower than a SERP, more concentrated at the top, and selected by different rules. A page can rank #1 on Google and never be cited; a page with no ranking can anchor an AI answer.

The decoupling is now measured: a March 2026 study of 863K SERPs found only 38% of AI-cited sources rank in Google's top 10 (Digital Applied). And the traffic behind the citations is real — AI referral traffic grew 9.9× in 19 months, with 92.4% coming from ChatGPT (Previsible, July 2026), converting at 7.1% — second only to paid search (Similarweb).

But "AI search" is not one channel. This report layers two datasets: Profound's foundational 680M-citation analysis (Aug 2024 – Jun 2025) as the historical baseline, and Ahrefs Brand Radar's monthly engine snapshots plus four other 2026 studies for the current picture (data through June 2026). Comparing the two windows reveals not just who each engine trusts today, but how fast those preferences are shifting — and they shifted dramatically in twelve months. The question this report answers: who does each engine trust now, and how do you become one of them?

ChatGPT → Reddit + Wikipedia·Perplexity → YouTube·AI Overviews → YouTube + social·Gemini → Reddit + YouTube·Claude → Brand sites

02 · The Data

Platform by Platform: Who Gets Cited

Five engines, five sourcing philosophies — all charted on the freshest published data. ChatGPT, AI Overviews, Perplexity, and Gemini use Ahrefs Brand Radar's 2026 monthly snapshots (millions of US queries each); Claude uses OtterlyAI's June 2026 study. Profound's 680M-citation dataset (Aug 2024 – Jun 2025) is retained as the historical baseline inside each card — the deltas between the two windows are where the story is. Each pie is labeled with its own base; bases differ across studies, so compare shapes, not slices, across engines.

ENGINE 01ChatGPT — community first, encyclopedia second

The biggest flip in the dataset. In Profound's Aug 2024 – Jun 2025 window, Wikipedia dominated ChatGPT (7.8% of all citations, 47.9% of its top-10 share). By 2026, Ahrefs Brand Radar's 9.6M-query panel shows Reddit at 29.4% of US citation mentions — nearly double Wikipedia's 15.0%. Community experience has overtaken encyclopedic authority as ChatGPT's default source.

Reddit · 29.4%
Wikipedia · 15.0%
Amazon · 3.4%
Forbes · 1.7%
Business Insider · 1.3%
All other domains · 49.2%
Share of tracked ChatGPT citation mentions · Ahrefs Brand Radar, 9.6M queries, US, 2026 snapshot 📊 Historical baseline: Wikipedia led at 7.8% overall / 47.9% of top-10 share (Profound, 680M citations, Aug 2024 – Jun 2025). And 99.2% of ChatGPT's Reddit citations point to specific discussion threads — not subreddit landing pages (Profound, 180,994 citations).
  • Reddit + Wikipedia together take ~44% of tracked citations — a two-source oligopoly
  • Established publications (Forbes, Business Insider, The Spruce, NY Post, Wired) fill the mid-tier
  • Long tail beyond the top 5 still holds ~49% — specialist pages remain citable
ENGINE 02Google AI Overviews — the social-video surface

Reaching 2B+ monthly users, AIO has shifted hard toward Google-owned and social sources. In Ahrefs' June 2026 snapshot (3M+ US queries), YouTube leads at 20.9% of top-50 mention share with Reddit at 19.6% — and Facebook (11.6%), Instagram (5.2%), and TikTok (3.6%) now outrank most publishers. AIO still tracks the SERP more closely than any other surface: traditional SEO continues to correlate tightly with AIO inclusion (76% of AIO citations pull from the top 10, per Ahrefs).

YouTube · 20.9%
Reddit · 19.6%
Facebook · 11.6%
Google.com · 6.0%
Instagram · 5.2%
Other top-50 · 36.7%
Mention share among AI Overviews' top-50 cited domains · Ahrefs Brand Radar, US, June 2026 📊 Historical baseline: in Profound's Aug 2024 – Jun 2025 dataset AIO was the most balanced surface — Reddit 2.2%, YouTube 1.9%, Quora 1.5%, LinkedIn 1.3% of all citations. The 2026 shift is toward video and closed social platforms.
  • Social platforms (Reddit, Facebook, Instagram, TikTok, Quora) take an outsized share vs every other engine
  • If you rank top-5 for a query, you are likely cited in AIO for related prompts
  • Wikipedia is just 4.8% here vs 15% on ChatGPT — encyclopedic content matters least on AIO
ENGINE 03Perplexity — from Reddit-first to video-first

The second big 2026 flip. In Profound's 2024–25 window, Reddit owned Perplexity (6.6% of all citations, 46.7% of top-10 share). Ahrefs' June 2026 snapshot (3.1M US queries) now puts YouTube on top at 32.4% — the most concentrated top source of any assistant — with Reddit second at 16.6% and Wikipedia at 8.2%. Perplexity also cites ~3× more sources per answer than ChatGPT (often 6+), rewarding pages with clear methodology and data tables.

YouTube · 32.4%
Reddit · 16.6%
Wikipedia · 8.2%
Alibaba · 5.1%
Oreate AI · 2.9%
Other top-50 · 34.8%
Mention share among Perplexity's top-50 cited domains · Ahrefs Brand Radar, US, June 2026 📊 Historical baseline: Reddit 46.7% of top-10 share (Profound, Aug 2024 – Jun 2025). Commerce is Perplexity's new wildcard — Alibaba, Etsy, Target, and Walmart all crack the June 2026 top 10 as its shopping features mature.
  • Most citation-transparent surface: linked references up front, often 6+ per answer
  • First-hand experience content ("I tried…", "after 6 months…") still disproportionately cited via Reddit
  • Comparison tables with consistent columns get pulled aggressively for commercial queries
ENGINE 04Gemini — the Google-property loyalist

Ahrefs Brand Radar's June 2026 snapshot (3M+ US queries, 14.5M prompts analyzed) puts Reddit at 27.5% of Gemini's top-50 mention share, with YouTube (13.7%) and Wikipedia (12.7%) close behind — the community + video + encyclopedia trio covers over half of Gemini's source attention. Uniquely, retail and marketplace domains (Walmart 2.8%, eBay 1.9%, Target 1.3%) crack the top sources — no other engine cites commerce pages this heavily.

Reddit · 27.5%
YouTube · 13.7%
Wikipedia · 12.7%
Forbes · 2.9%
Walmart · 2.8%
Other top-50 · 40.4%
Mention share among Gemini's top-50 cited domains · Ahrefs Brand Radar, US, June 2026 📊 2026 update: Google properties anchor Gemini heavily — yet Gemini shares only 38.5% of its top-200 cited domains with AI Overviews. Even inside Google, the two surfaces are separate channels. Profound also found Gemini over-indexes on sub-1K-view long-tail YouTube videos (ChatGPT cites high-view videos 1.5× more often).
  • Pulls YouTube, Play Store, Maps, and Knowledge Graph entities more than any other engine
  • Only engine where retail domains (Walmart, eBay, Target, Best Buy) earn top-50 citation share
  • Video transcripts are first-class citable content — a spoken answer is a source
ENGINE 05Claude — the brand-site reader

OtterlyAI's June 2026 study (379,321 citations, 16,406 domains, SaaS/tech queries) shows Claude is the outlier: brand and product domains take 64.0% of all citations — 4× the next category. Social media is just 0.9%, and Reddit was cited zero times. Claude's sourcing is also the most distributed: its top-10 domains hold only 9.5% of citations, and 32.9% of cited domains appear exactly once.

Brand / product sites · 64.0%
News / media · 14.9%
Blogs / personal · 5.4%
Education · 2.7%
Government / NGO · 2.4%
Other categories · 10.6%
Share of Claude citations by source category · OtterlyAI, 379K citations, SaaS/tech queries, June 2026 📊 Only 13% of Claude's cited domains overlap with ChatGPT's (4.2% at URL level) — Claude is a separate channel with separate winners. What little social it cites is professional: LinkedIn alone is 43.1% of its social citations.
  • First-party product, docs, and comparison pages are the primary citation route
  • Long tail rewards breadth: modest-authority sites get cited if a page answers one question completely
  • Do not port a Reddit-first playbook here — build your site like documentation, not advertising
The core asymmetry (June 2026): Reddit is 29.4% of ChatGPT's citations, 27.5% of Gemini's, 19.6% on AIO, 16.6% on Perplexity — and zero on Claude, where brand sites take 64%. YouTube leads Perplexity and AIO but barely registers on Claude (0.3%). A single content strategy cannot serve engines this different — platform-specific GEO is not optional.

03 · What Changed in 2026

Three Shifts Since the Original Dataset

Shift 1 — YouTube overtook Reddit; the old favorites flipped

The biggest single change since Profound's Aug 2024 – Jun 2025 window: by February 2026, YouTube accounted for 39.2% of AI citations, overtaking Reddit at 20.3% (Adweek study, via Digital Applied). Lantern's February 2026 analysis of 200M+ citations independently found YouTube the most-cited domain by more than 2× the second-ranked domain. By June 2026 the per-engine leaders had flipped outright: ChatGPT's #1 went from Wikipedia to Reddit (29.4% vs 15.0%), and Perplexity's from Reddit to YouTube (32.4% vs 16.6%), per Ahrefs Brand Radar. AI engines increasingly parse video transcripts as source material — a video with a clear spoken answer is now a citable asset.

Shift 2 — Concentration intensified

The 2026 5W AI Platform Citation Source Index found the top 15 domains absorb ~68% of the AI answer pipeline. Yet the paradox cuts the other way too: Evertune's 200M-prompt analysis found even the most-cited domain rarely exceeds 5% of total citations — meaning the long tail is where brands win. Profound's own 27M-prompt study found 97.4% of AI citations come from non-Tier-1 earned media: niche blogs, trade publications, and review sites — not the New York Times.

Shift 3 — Citations decoupled from rankings

Only 38% of AI-cited sources rank in Google's top 10 (863K SERPs, March 2026). Q2 2026 data confirms the selection logic: pages that lead with a quotable number, a clear definition, or a named framework are cited 2–3× more often than higher-authority pages that bury the same fact in prose. Extractability now beats authority.

PatternAug 2024 – Jun 2025 (Profound)2026 status
Top domain overallWikipedia (ChatGPT) / Reddit (AIO, Perplexity)Flipped: Reddit #1 on ChatGPT (29.4%) and Gemini (27.5%); YouTube #1 on Perplexity (32.4%) and AIO (20.9%) — Ahrefs, June 2026
ConcentrationTop-10 sources dominate each platformTop 15 domains absorb ~68% of citation volume (5W Index)
Ranking linkAssumed correlatedOnly 38% of cited sources rank top-10 — 62% cited without ranking
TLD split.com 80.4% · .org 11.3% · .io/.ai ~2.8%.com dominance stable; .ai TLD presence growing with tech queries
Traffic impactEarly-stageAI referrals up 9.9× in 19 months; ChatGPT converts at 7.1%
Sources: Profound (680M citations); Adweek/Digital Applied (Feb–Apr 2026); 5W Index 2026; Previsible (Jul 2026); Similarweb (Apr–May 2026); Peec AI.

04 · The Playbook

What To Do About It, Per Platform

Mapped to the gptmelo GEO Framework — each move below names the layer it belongs to.

Target engineHighest-leverage moveFramework layer
ChatGPTGenuine thread-level participation in the 5–10 subreddits your buyers read (Reddit is now 29.4% and 99.2% of its Reddit citations are threads); keep a clean Wikipedia-adjacent entity presence for the 15% encyclopedia layerLayer 3 · Authority
PerplexityPublish video with clearly spoken answers (YouTube is now 32.4%); back it with primary data, named methodology, and data tables for the 6+-source answersLayer 2 + 4 · Content & Citation
AI OverviewsKeep classic SEO strong (76% of AIO citations pull from the top 10); add YouTube and social distribution — video + social now anchor over 60% of AIO's top-source shareLayer 1 + 2 · Technical & Content
GeminiPublish video with clearly spoken answers (transcripts are citable — even sub-1K-view niche videos get cited); complete Google Business Profile, Knowledge Graph, and marketplace listingsLayer 2 · Content
ClaudeBuild your own site like documentation: reference-grade product, comparison, and "how it works" pages (64% of Claude citations are brand sites — and Reddit is worth zero here)Layer 2 + 4 · Content & Citation

Three cross-platform moves that compound

  • Lead with the quotable fact (Layer 4). One extractable number or definition per H2 section. Pages structured this way are cited 2–3× more than higher-authority pages without it.
  • Ship primary data quarterly (Layer 3). Original research with stated methodology earns citations across all five surfaces for 6–12 months post-publish — it makes you the cited source instead of the citing one.
  • Monitor all five surfaces, not just Google (Layer 5). The same query produces visibly different citation sets per engine. Track your AI Mention Rate monthly per platform; the gaps tell you which playbook row to run next.
The 97.4% opportunity: AI engines overwhelmingly cite non-Tier-1 sources. You don't need to outrank the New York Times — you need one extractable, authoritative, unique page per question your buyers ask. That is a content problem, not a budget problem.
See your own citation pattern

Which engines cite you — and which don't?

gptmelo monitors your brand's mentions across ChatGPT, Perplexity, Gemini and more, archives every answer, and flags the questions where competitors are cited and you're not. Run a free site audit →


FAQ

Frequently Asked Questions

Which source does ChatGPT cite most?

Reddit — 29.4% of tracked US citation mentions in Ahrefs Brand Radar's 2026 data (9.6M queries), nearly double Wikipedia's 15.0%. This is a flip: in Profound's Aug 2024 – Jun 2025 dataset, Wikipedia led with 7.8% of all citations (47.9% of top-10 share) and Reddit was a distant second.

Which source does Perplexity cite most?

YouTube — 32.4% of top-50 mention share in June 2026 (Ahrefs), the most concentrated top source of any assistant, ahead of Reddit at 16.6%. Another flip: Reddit led Perplexity with 46.7% of top-10 share in 2024–25. Perplexity still cites ~3× more sources per answer than ChatGPT, often 6 or more.

Does ranking #1 on Google guarantee AI citations?

No. Only 38% of AI-cited sources rank in Google's top 10 (863K-SERP study, March 2026). AI Overviews still track the SERP closely; ChatGPT, Perplexity, and Claude select on extractability, authority signals, and community validation instead.

Is one GEO strategy enough for all AI engines?

No. Wikipedia is 13× more important on ChatGPT than on AI Overviews; Reddit is 3× more important on Perplexity than on AIO — and cited zero times by Claude, where 64% of citations go to brand sites instead. The engines share a preference for extractable, authoritative, unique content (Layers 2–4 of the GEO Framework) — but the distribution channels differ per platform.

Which source does Claude cite most?

First-party brand and product websites — 64.0% of all Claude citations in OtterlyAI's June 2026 study (379K citations, SaaS/tech queries), 4× the next category. Claude cited Reddit zero times and spreads citations across a long tail: its top-10 domains hold just 9.5% of citations, so modest-authority sites can earn citations with well-structured, self-contained pages.


05 · Methodology & Sources

Data Sources & Citation

This report synthesizes published third-party citation datasets; percentages are reported as published by each source. Foundational platform data: Profound, "AI Platform Citation Patterns" (680M citations, Aug 2024 – Jun 2025, updated Aug 2025). 2026 updates: Digital Applied Q2 2026 citation analysis (5,000+ queries, 5 surfaces); Adweek YouTube citation study (Feb 2026); Lantern 200M-citation analysis (Feb 2026); 5W AI Platform Citation Source Index 2026; Previsible AI Traffic Report (6.77M sessions, Jul 2026); Similarweb clickstream (Apr–May 2026); Profound YouTube/Reddit deep-dives (Dec 2025 – Jan 2026). Current engine snapshots (Ahrefs Brand Radar, US, 2026): ChatGPT (9.6M queries; percentages via BuzzStream's share calculation), Perplexity (3.1M queries, June 2026), Google AI Overviews (3M+ queries, June 2026), and Gemini (3M+ queries, June 2026) — note Ahrefs "mention share" is share of the top-50 sources' summed citations, except ChatGPT (share of all tracked mentions). Claude: OtterlyAI Claude Citation Study (379,321 citations, 16,406 domains, SaaS/tech queries, June 2026 — note the vertical skew when generalizing). Third-party figures were not independently re-measured by gptmelo; where studies conflict (e.g., YouTube-vs-Reddit share), both figures are reported with attribution.

How to cite this report

Wang, C. (2026). AI Platform Citation Patterns 2026: How ChatGPT, Google AI Overviews & Perplexity Choose Their Sources. gptmelo Research. https://www.gptmelo.com/resources/ai-platform-citation-patterns-2026