Only 11% of AI-Cited Sources Overlap Between Platforms, Here's What Each One Actually Trusts

The short answer

AI search engines choose what to cite by retrieving content from their own indexes, and each platform trusts fundamentally different sources. When Profound measured which domains get cited by both ChatGPT and Perplexity for the same topics, the overlap was just 11%. Nine out of ten sources cited by one platform were completely ignored by the other.

The reason is structural. Each AI search platform draws from different source pools. ChatGPT relies on Bing and heavily cites Wikipedia, which accounts for 7.8% of all its citations. Perplexity runs its own index of over 200 billion URLs and favors Reddit, which represents 6.6% of its total citations, roughly three-and-a-half times ChatGPT's Reddit reliance. Google AI Overviews pull approximately 38% of their cited sources from existing top-10 organic results. Gemini provides no clickable citation at all in 92% of its answers.

Every major AI search platform uses Retrieval-Augmented Generation, retrieve content from the web, synthesize an answer, cite specific sources, but the retrieval step is where they diverge. Different indexes, different trust signals, different source preferences. A single optimization strategy cannot cover them all.

This article breaks down what each platform actually trusts, where the citation process fails, and what that means for businesses trying to get cited across the AI search landscape.

The shared process: Retrieval-Augmented Generation

Every AI search platform runs a process called Retrieval-Augmented Generation, or RAG. RAG is the three-step process AI search platforms use to answer queries: retrieve relevant content from the web or a proprietary index, synthesize a response by combining information from multiple sources, and cite specific sources by attributing parts of the answer to the pages it drew from.

Those citations are the new visibility. They're the equivalent of appearing on page one of Google, except there are typically only a handful of citation slots per answer, often far fewer than ten. If your content isn't selected during retrieval, it can't be cited in the answer.

Every platform runs this process. The differences are in how each step works, and those differences determine who gets cited and who gets ignored.

ChatGPT

ChatGPT's citation system relies on Bing's search index as its retrieval backbone. When ChatGPT needs current information, it queries Bing and applies its own logic to decide what's worth extracting and citing. It strongly favors encyclopedic, well-attributed content from established authority domains, making entity-level signals the primary lever for ChatGPT visibility.

What sources does ChatGPT cite most?

The source preferences are distinctive. Profound's analysis of 680 million citations found that Wikipedia accounts for 7.8% of all ChatGPT citations, the highest single-domain share of any AI platform. Among ChatGPT's top 10 most-cited sources, Wikipedia's dominance is even more pronounced, representing 47.9% of that group's share. Reddit accounts for about 1.8% of total citations. The rest skews toward established authority: major publications, government domains, educational institutions, and business platforms like G2 and Forbes.

This tells you what ChatGPT values: encyclopedic, factual, well-established information. Content that reads like a reference source, with definitions, statistics, and well-attributed claims, performs better than marketing copy or opinion pieces.

How does entity recognition affect ChatGPT visibility?

The practical implication is that entity work matters for ChatGPT visibility. Your Wikipedia presence (if your company meets notability criteria), your Wikidata entry, and your presence on authoritative platforms like G2 and industry publications are the kinds of signals ChatGPT's retrieval system recognizes.

The exact mechanism by which Wikidata entities and sameAs links feed into ChatGPT's citation process isn't documented by OpenAI, but the pattern is consistent with how knowledge graphs support entity resolution: a company with a well-connected web presence across authoritative platforms has a structural advantage over one whose primary online presence is its own website alone.

Aivarize's cross-platform audits have revealed a specific blind spot here: companies that score well on traditional authority metrics but have thin entity footprints outside their own domain tend to underperform on ChatGPT relative to what their SEO strength would predict.

What are ChatGPT's citation blind spots?

ChatGPT's citation process is far from transparent. Research analyzing 14,000 real-world conversations found that GPT-4o generates 24% of its responses without explicitly fetching any online content at all, raising questions about how much of what it presents actually comes from retrieved sources versus its training data.

Perplexity

Perplexity's citation system is built on its own proprietary search index of over 200 billion URLs, independent of both Google and Bing. It heavily favors Reddit (6.6% of total citations), YouTube (2.0%), and recently updated content, making freshness and community presence the primary levers for Perplexity visibility.

How does Perplexity's index differ from Google's and Bing's?

Perplexity doesn't use Bing or Google. It maintains its own proprietary search index of over 200 billion URLs, crawled and indexed independently. Pages that rank well on Google don't automatically appear in Perplexity's results.

The source preferences are striking. Among Perplexity's top 10 most-cited domains, Reddit accounts for 46.7% of that group's share, and 6.6% of total citations, roughly three-and-a-half times ChatGPT's Reddit reliance. YouTube accounts for 2.0% of total citations (13.9% among the top 10). Community-generated content, forum discussions, and user experiences carry far more weight here than on any other platform.

Why does content freshness matter more on Perplexity?

Freshness is the other critical factor. Practitioner research (from Digitaloft, cited by Onely) found that 76.4% of ChatGPT's most-cited pages were updated within the last 30 days. Perplexity shows an even stronger recency preference, practitioner data from ConvertMate's analysis of over 10,000 domains suggests content updated within 30 days performs substantially better on Perplexity specifically. An authoritative article from six months ago may be passed over in favor of a Reddit discussion from last week.

Aivarize has seen this directly in client audits. Companies with strong SEO rankings are often surprised to find they're nearly invisible on Perplexity because their content hasn't been updated in months. It's the platform where "publish and forget" is punished most visibly.

How many sources does Perplexity cite per response?

Perplexity cites more sources per response than other platforms, averaging around 21 citations per answer compared to roughly 3–8 typical elsewhere. More citation slots means more opportunities to appear, but also means Perplexity is cross-referencing sources and looking for consistency.

There's a catch, though. Perplexity visits approximately 10 relevant pages per query but only cites 3 to 4 of them. That means roughly 6–7 relevant sources go uncredited per response. Writing citable, extractable content improves your chances of being among the 3–4 that make the cut.

For businesses, Perplexity rewards active participation and regular publishing. A company blog updated weekly with well-sourced, discussion-worthy content will outperform a comprehensive but static website. Building genuine presence in relevant Reddit communities and producing YouTube content both directly feed Perplexity's source preferences.

Google AI Overviews

Google AI Overviews draw approximately 38% of their cited sources from pages already ranking in Google's top-10 organic results, the highest overlap between traditional search ranking and AI citation of any platform. This makes strong SEO the foundation for AI Overview visibility, with content structure and fan-out query coverage as the GEO layer on top.

How much do organic rankings influence AI Overview citations?

Google AI Overviews have a fundamentally different relationship with existing search results than any other AI platform.

In Ahrefs' mid-2025 study of 1.9 million citations, 76% of the sources cited in AI Overviews came from pages already ranking in Google's top 10 organic results for the same query. However, an updated analysis from early 2026, using 4 million AI Overview URLs, found this has dropped to approximately 38%, likely driven by Google's switch to Gemini 3 and more aggressive fan-out query behavior.

Even at 38%, this is still the highest overlap between traditional ranking and AI citation of any platform. If you rank well on Google, you have an advantage in AI Overviews, but it's no longer the near-guarantee it appeared to be a year ago.

What content structures do AI Overviews favor?

Ranking alone isn't enough. AI Overviews favor specific content structures: question-based headings that match how users phrase queries, FAQ sections, comparison tables, and definition boxes. Content optimized for featured snippets tends to perform well in AI Overviews too, as the patterns overlap significantly.

AI Overviews appear selectively, triggering on roughly 20% of all Google searches according to Ahrefs' analysis of 146 million SERPs. The rate is higher for informational and how-to queries, lower for navigational or transactional searches. Understanding which queries in your industry trigger AI Overviews is part of the optimization.

YouTube content plays a notable role. Google's AI can pull information from YouTube transcripts, descriptions, and metadata, especially for how-to and tutorial queries. This creates a unique advantage for businesses with video content in Google's ecosystem.

What are fan-out queries and why do they matter?

As the overlap between organic rankings and AI citations continues to shift, a newer factor is becoming increasingly important: fan-out queries. Google AI Overviews don't just answer the query you typed. They generate sub-queries, sometimes called fan-out queries, to expand the original search before synthesizing a response. Content that answers related questions within the same page is more likely to be pulled into the expanded synthesis. Optimizing for these secondary queries is becoming as important as ranking for the primary term.

For businesses, Google AI Overviews are the most SEO-adjacent AI platform. Strong traditional SEO is the foundation. The GEO layer on top focuses on content structure: making sure your already-ranking content is formatted so AI can extract clean, quotable answers from it.

Find out where your business stands. We score brands across all five GEO dimensions and show you exactly where you're visible, where you're not, and what to do about it. Get your GEO visibility score →

Gemini

Gemini is Google's standalone AI assistant, and it draws from Google's full ecosystem, Search, YouTube, Scholar, Maps, and Business Profiles, to select sources. YouTube carries particularly strong weight, but Gemini's severe attribution gap (no clickable citation in 92% of answers) makes brand-level entity recognition even more critical than direct link visibility.

Why does YouTube matter so much for Gemini?

YouTube carries particular weight. Ahrefs' study of 75,000 brands found YouTube brand mentions are the single strongest correlating factor with AI citation rates across the platforms studied (ChatGPT, AI Mode, and AI Overviews, with a correlation of ~0.737). Gemini, given its deep integration with Google's ecosystem and YouTube specifically, is likely to share or amplify this pattern. Businesses with active YouTube channels see noticeably stronger representation across Google's AI surfaces.

Google Business Profile data also matters, especially for local and company-related queries. A complete, verified profile with accurate information helps Gemini represent your business correctly when users ask about you or your category.

What is Gemini's attribution problem?

The attribution problem is severe. Research found that Gemini provides no clickable citation in 92% of its answers, and 34% of responses are generated without explicitly searching the web at all. This means Gemini uses your content to inform its answers but rarely tells the user where the information came from.

For businesses, this makes brand presence and entity recognition even more important: even if Gemini doesn't link to you, being part of its knowledge base influences how it talks about your industry.

Copilot

Bing Copilot shares ChatGPT's retrieval infrastructure (both use Bing) but applies its own weighting, with a notable lean toward business publications like Forbes and Gartner. For B2B businesses, Copilot optimization overlaps heavily with professional content strategy: thought leadership, industry analysis, and an active LinkedIn presence.

What sources does Copilot favor?

Based on Profound's citation data and practitioner observation, business publications carry particular weight: Forbes and Gartner appear frequently in Copilot's business-related responses, with Forbes showing a notably higher citation count on Copilot than on other platforms.

LinkedIn presence appears to matter more for Copilot than for other platforms, which makes sense given that Microsoft owns both Bing and LinkedIn. This is Aivarize's inference based on the platform's architecture and observed citation patterns, not a published study.

What technical signals help with Copilot visibility?

Two technical signals are worth noting for Copilot specifically.

Bing Webmaster Tools verification confirms your site's legitimacy in Bing's index, a baseline requirement for Copilot visibility.

IndexNow implementation goes further. It notifies Bing of content changes in real time, improving how quickly Copilot discovers new or updated pages. For businesses publishing frequently, this is the fastest path to Copilot freshness.

Grok

Grok, built by xAI, draws heavily from X/Twitter for its real-time information. Where ChatGPT relies on Wikipedia and Perplexity relies on Reddit, Grok's distinctive signal is X/Twitter activity, making it the platform most responsive to social discussion, trending conversations, and brand mentions in posts.

How does X/Twitter activity influence Grok?

Brands that are actively discussed, mentioned, or shared on X see stronger representation in Grok's responses. This makes Grok particularly relevant for industries where social conversation drives purchasing decisions, such as technology, consumer products, and media.

Large-scale citation analysis for Grok is still limited compared to the other platforms covered here, so these observations are based on Grok's stated design and early practitioner reports rather than the kind of multi-million-citation datasets available for ChatGPT or Perplexity. For businesses with an active X/Twitter presence, Grok is worth monitoring separately.

How do the platforms compare side by side?

Platform	Primary Index	Top Source Bias	Avg Citations per Response	Freshness Sensitivity	Key Optimization Lever
ChatGPT	Bing	Wikipedia (7.8% of all citations)	3–8	Moderate	Entity recognition & authority domains
Perplexity	Proprietary (200B+ URLs)	Reddit (6.6% of all citations)	~21 (cites 3–4 per query)	Very high (30-day window)	Fresh content & community presence
Google AI Overviews	Google Search	Own organic top-10 (~38% overlap)	Varies by query	Moderate-high	SEO foundation + structured content
Gemini	Google ecosystem (Search, YouTube, Scholar)	YouTube (~0.737 brand correlation)	Rarely provides clickable links (92% uncited)	Moderate	YouTube & Google Business Profile
Copilot	Bing	Forbes, Gartner, business publications	3–8	Moderate (IndexNow helps)	Business content & LinkedIn presence
Grok	X/Twitter + web	X/Twitter activity	Limited data	Very high (real-time)	Active X/Twitter presence

Why the 11% overlap changes everything

Optimizing for multiple AI platforms requires a multi-channel strategy because each platform retrieves from a different index, applies different trust signals, and favors structurally different source types, producing citation lists that overlap by as little as 11%.

When Profound measured which domains get cited by both ChatGPT and Perplexity for the same topics, the overlap was just 11%. Nine out of ten domains cited by one platform were ignored by the other. (Separately, academic research measuring domain-level overlap between AI platforms and Google search found similarly low overlap across the board, confirming that generative engines surface fundamentally different source ecosystems.)

The reasons are structural. ChatGPT trusts Wikipedia and established authority sites. Perplexity trusts Reddit and fresh content. Google AI Overviews trust their own organic rankings. Each platform has built its own retrieval system with its own biases, and those biases produce fundamentally different citation lists.

One caveat worth repeating from Aivarize's GEO guide: citation behavior is volatile. Research from the same Profound dataset shows that 40–60% of the domains cited in AI responses change within a single month, even for identical queries. The platform-specific patterns described in this article are real and measurable, but they're probabilistic, not permanent.

This is why multi-platform measurement matters, and why the Aivarize GEO Scoring Index tests across ten platforms rather than one. A single optimization strategy cannot cover all platforms. A business that focuses exclusively on Wikipedia entity work (optimizing for ChatGPT) will see minimal benefit on Perplexity, which values community discussion and freshness. Strong Google rankings help with AI Overviews but have no direct effect on ChatGPT or Perplexity.

What all platforms share

Despite their differences, every AI platform values four content characteristics:

Structured, extractable content. All platforms need to pull coherent passages from your pages. Self-contained paragraphs that answer a specific question without requiring surrounding context are universally preferred. Practitioner analysis suggests AI-cited passages tend to cluster in the range of roughly 100 to 200 words, with a practical sweet spot around 120 to 180 words, though this varies by topic and platform.
Specific, well-attributed data. Content containing specific numbers, percentages, and named data sources is cited more frequently across all platforms. The foundational GEO study (Aggarwal et al., KDD 2024) found that adding source citations, relevant quotations, and embedded statistics could collectively boost visibility by up to 40% on the Position-Adjusted Word Count metric, the strongest effect of any optimization strategies tested.
Clear sourcing and verifiability. Content that cites its own sources, linking to studies, naming experts, and referencing specific data, signals reliability to AI systems trained to prefer verifiable information.
Entity clarity. All platforms perform better when they can clearly identify what your content is about and who created it. Knowledge Graph connections and consistent brand information across platforms strengthen entity recognition. Schema markup contributes to this, though no major AI system currently parses JSON-LD semantically — the benefit comes from schema making content clearer as text, and only when the markup is attribute-rich rather than generic boilerplate.

These shared signals are the foundation. Platform-specific optimization builds on top of them. Aivarize covers the content side in detail in: How to Write Content That AI Will Actually Cite →

What should you do with this?

Start by understanding which AI platforms matter most for your business. If your customers are researchers or professionals, Perplexity may be more relevant than ChatGPT. If your traffic comes primarily from Google, AI Overviews are your first priority. If you serve a B2B audience, Copilot's business content preferences matter.

Then measure where you currently stand on each platform. Search for your product category, your competitors' names, and the problems you solve on ChatGPT, Perplexity, and Google, and check whether your brand appears in any cited source. Test the same query across platforms; the gaps are often surprising. Where you're missing, use the platform-specific levers above: entity work for ChatGPT, freshness and community presence for Perplexity, structured content for AI Overviews.