AI search platforms do not share sources. Across 50 queries repeated 3 times on 4 platforms, the cross-platform URL overlap was 1.4%. ChatGPT matched Google's Top-3 results only 7.8% of the time. Gemini matched 32.4% of the time. If you treat "AI search" as one channel, you are invisible on most of it.
You have probably heard the advice: "optimize for AI search." The problem is that "AI search" is not one thing. It is four (or five) completely separate retrieval systems that happen to produce natural language answers. And when you look at which URLs those systems actually cite, the overlap is almost zero.
This post presents the citation overlap data from a controlled experiment: 50 queries, each repeated 3 times, across ChatGPT, Claude, Perplexity, and Gemini. We break down exactly how much each platform agrees with the others, how much each agrees with Google, and why the numbers are so low. Every claim traces to published research or reproducible methodology.
For the broader cross-platform citation comparison, see What AI Platforms Actually Cite. For the research that established these overlap metrics, see Query Intent and AI Citation.
🔬 THE 1.4% NUMBER: WHAT IT ACTUALLY MEANS
The headline finding from Lee (2026) is stark: when you send the same query to ChatGPT, Claude, Perplexity, and Gemini, only 1.4% of the cited URLs appear in more than one platform's response. The Jaccard similarity index across all platform pairs was 0.014.
To put that in perspective: if ChatGPT cites 5 URLs for a given query and Perplexity cites 5 URLs for the same query, the expected number of shared URLs between them is less than one. In most cases, it is exactly zero.
This was not a one-off measurement. The study used 50 carefully selected queries spanning informational, discovery, comparison, and validation intent types. Each query was submitted 3 times to each platform to control for response variability. That produced 600 total query-platform observations (50 queries x 3 repetitions x 4 platforms), generating thousands of individual citation URLs for comparison.
| Metric | Value | Source |
|---|---|---|
| Queries tested | 50 (x 3 repetitions each) | Lee (2026) |
| Platforms compared | ChatGPT, Claude, Perplexity, Gemini | Lee (2026) |
| Total observations | 600 (50 x 3 x 4) | Lee (2026) |
| Cross-platform URL overlap (Jaccard) | 0.014 (1.4%) | Lee (2026) |
| Within-platform consistency (ChatGPT) | 0.619 (61.9%) | Lee (2026) |
The within-platform consistency numbers tell an interesting parallel story. When you ask ChatGPT the same question three times, it cites the same URLs about 62% of the time. That is reasonably consistent. But when you compare ChatGPT's citations to Perplexity's citations for the same query, that consistency collapses to near-random levels.
The Bottom Line: Cross-platform citation overlap is not low. It is nearly nonexistent. Each AI platform operates in its own citation universe. A page that Perplexity loves may be completely invisible to ChatGPT, and vice versa.
📊 GOOGLE TOP-3 OVERLAP BY PLATFORM: THE ALIGNMENT MATRIX
If cross-platform overlap is near zero, what about overlap with traditional Google Search? This is where the data gets strategically useful. Lee (2026) measured how often each AI platform's cited URLs matched Google's Top-3 organic results for the same query.
The differences are dramatic:
| Platform | URL Overlap with Google Top-3 | What It Means |
|---|---|---|
| ChatGPT | 7.8% | Almost no alignment with Google's top results |
| Claude | 11.2% | Slightly more Google alignment, still very low |
| Perplexity | 29.7% | Moderate alignment with Google's top picks |
| Gemini | 32.4% | Highest alignment with Google rankings |
This table is one of the most actionable findings in recent AI search research. It tells you exactly how much traditional SEO carries over to each AI platform.
Gemini, which is grounded directly in Google Search infrastructure, shows the highest alignment at 32.4%. That makes architectural sense: Gemini draws from Google's own index and ranking signals. If you rank well on Google, you have a roughly one-in-three chance of being cited by Gemini for the same query.
Perplexity comes in at 29.7%, reflecting its own independent index but with some convergence toward the same high-authority sources that Google favors.
ChatGPT sits at just 7.8%. That number should give pause to anyone whose AI optimization strategy starts and ends with "rank on Google." ChatGPT discovers URLs through Bing's API, then fetches pages in real time. Google's ranking has almost no bearing on what ChatGPT surfaces.
Claude lands at 11.2%, slightly higher than ChatGPT but still in the single-digit-adjacent range. Claude uses external search for URL discovery but applies its own content evaluation criteria that diverge substantially from Google's.
The Bottom Line: Google rank is a meaningful predictor of AI citation only for Gemini and (to a lesser degree) Perplexity. For ChatGPT and Claude, Google rank is essentially irrelevant to citation probability. If your entire AI strategy is "rank higher on Google," you are reaching at most 32% of the AI citation landscape.
🏗️ WHY THE OVERLAP IS SO LOW: FOUR ARCHITECTURAL REASONS
The 1.4% overlap is not random noise. It is the predictable result of four fundamental differences in how these platforms find, evaluate, and select sources.
1. Different Retrieval Architectures
The most consequential difference is the retrieval pipeline itself:
| Platform | Architecture | URL Discovery Method |
|---|---|---|
| ChatGPT | Live fetching | Bing API lookup, then real-time page fetch via ChatGPT-User |
| Claude | Live fetching | External search API, then real-time page fetch via Claude-User |
| Perplexity | Pre-built index | PerplexityBot crawls proactively; retrieves from own index at query time |
| Gemini | Google-grounded | Retrieves directly from Google Search infrastructure |
ChatGPT and Claude never see a page unless an external search engine surfaces it first. Perplexity only sees pages its own crawler has already visited. Gemini only sees pages that Googlebot has indexed. These are four different versions of the web, and the overlap between them is naturally limited.
2. Different Indexes
Even when two platforms use search APIs for discovery, they use different search engines. ChatGPT routes through Bing. Claude uses its own search integration. Perplexity maintains a proprietary index. Gemini uses Google.
Bing and Google do not index the same pages. Research on traditional search engine overlap has consistently shown that different search engines maintain substantially different indexes, even for identical queries (Hoechstoetter & Lewandowski, 2015). When AI platforms inherit these different indexes, they inherit the divergence.
3. Different Ranking and Selection Signals
Even when two platforms find the same candidate URL, they may make different citation decisions. Each platform applies its own criteria for deciding whether a source is worth citing:
- ChatGPT favors authoritative, well-structured pages with clear factual claims. It tends to cite fewer sources but with higher apparent authority.
- Claude is more conservative with citations overall, emphasizing synthesis over source attribution.
- Perplexity cites more sources per response and shows preference for fresh content. Its index skews 3.3x fresher than Google's for medium-velocity topics (Lee, 2026).
- Gemini inherits Google's authority signals, favoring pages with strong E-E-A-T indicators and existing search visibility.
4. Different Content Format Preferences
Platforms differ in which content types they can even process:
| Content Type | ChatGPT | Claude | Perplexity | Gemini |
|---|---|---|---|---|
| Server-rendered HTML | Yes | Yes | Yes | Yes |
| Client-side JS content | No | No | Partial | Yes (via Googlebot) |
| YouTube video | No | No | Yes | Yes |
| Reddit threads (API) | No | No | No | No |
| Reddit threads (Web UI) | Yes (17%) | No | Yes (20%) | Yes |
| PDF documents | Partial | Partial | Yes | Yes |
A YouTube video cannot be cited by ChatGPT regardless of its quality. A page that requires JavaScript rendering is invisible to ChatGPT and Claude. These format-level incompatibilities alone explain a significant portion of the overlap gap.
The Bottom Line: The 1.4% overlap is not a bug. It is a structural feature of how AI search works. Four different retrieval methods, four different indexes, four different ranking signals, and four different format capabilities produce four largely non-overlapping citation sets.
For a deeper dive into the architectural differences, see our Generative Engine Optimization Guide.
🔄 THE CROSS-PLATFORM CITATION MATRIX
To make the pairwise relationships concrete, here is the full cross-platform citation matrix showing URL-level overlap between every platform pair. All values represent Jaccard similarity coefficients from the 50-query consistency dataset (Lee, 2026).
| ChatGPT | Claude | Perplexity | Gemini | |
|---|---|---|---|---|
| ChatGPT | 0.619* | 0.021 | 0.018 | 0.012 |
| Claude | 0.021 | 0.583* | 0.015 | 0.019 |
| Perplexity | 0.018 | 0.015 | 0.647* | 0.024 |
| Gemini | 0.012 | 0.019 | 0.024 | 0.601* |
Diagonal values represent within-platform consistency (same query asked 3 times).
Several patterns emerge from this matrix:
Perplexity and Gemini show the highest cross-platform overlap (0.024). This makes sense because both maintain their own indexes that tend to converge on high-authority sources. They are the two platforms most likely to independently discover the same page.
ChatGPT and Gemini show the lowest overlap (0.012). This is the most architecturally distant pair. ChatGPT discovers through Bing and fetches live. Gemini discovers through Google and serves from index. They operate in almost completely separate URL spaces.
Within-platform consistency ranges from 0.583 to 0.647. Perplexity is the most self-consistent (0.647), likely because its pre-built index provides more stable retrieval than live-fetching approaches. Claude is the least self-consistent (0.583), reflecting more variability in its source selection.
The Generative Engine Optimization (GEO) framework from Aggarwal et al. (2024) anticipated this fragmentation. Their research demonstrated that optimization strategies that boost visibility by up to 40% in one generative engine context may have minimal impact in another. The domain-specific nature of effective GEO strategies maps directly onto the platform-specific citation patterns we observe here.
The Bottom Line: No two platforms share more than 2.4% of their citations. The practical ceiling for "optimize once, appear everywhere" is roughly 2%. That is not a strategy. That is a rounding error.
🎯 WHAT THE CONSISTENCY DATA REVEALS
The 50 queries x 3 repetitions design was specifically chosen to separate within-platform variance from cross-platform divergence. Here is what that separation tells us.
Within-Platform: Reasonably Stable
When you ask ChatGPT the same question three times, you get the same cited URLs about 62% of the time. That is stable enough to be strategically meaningful. If your page gets cited once, it will likely get cited again for similar queries.
| Platform | Within-Platform Jaccard | Interpretation |
|---|---|---|
| Perplexity | 0.647 | Most consistent (index-based retrieval) |
| ChatGPT | 0.619 | Moderately consistent (Bing + live fetch) |
| Gemini | 0.601 | Moderately consistent (Google-grounded) |
| Claude | 0.583 | Least consistent (more synthesis variability) |
Cross-Platform: Nearly Random
When you compare any two platforms' citations for the same query, overlap drops to 1.2% to 2.4%. This is barely above what you would expect from random URL sampling. The platforms are not disagreeing on rankings. They are drawing from different pools entirely.
This has a direct implication for monitoring. If you track your AI visibility on only one platform, you are measuring at most 25% of the AI search landscape. A page that appears in 0% of ChatGPT citations could appear in 30% of Perplexity citations. Single-platform monitoring creates dangerous blind spots.
For a head-to-head breakdown of the two most different platforms, see ChatGPT vs Perplexity vs Gemini.
📋 IMPLICATIONS: THE MULTI-PLATFORM OPTIMIZATION FRAMEWORK
The overlap data leads to one unavoidable conclusion: you must optimize for each platform separately. Here is the framework the data supports.
Platform-Specific Optimization Priorities
| Platform | Google Top-3 Overlap | Primary Optimization Lever | Secondary Lever |
|---|---|---|---|
| ChatGPT (7.8%) | Very low | Bing indexation + server-side rendering | Clean HTML structure, factual density |
| Claude (11.2%) | Low | Content quality + structured claims | Internal linking, high content-to-HTML ratio |
| Perplexity (29.7%) | Moderate | Fresh sitemaps + crawl accessibility | Schema markup (dateModified), recency signals |
| Gemini (32.4%) | Moderate-high | Traditional Google SEO | E-E-A-T signals, YouTube companion content |
The "Coverage vs. Depth" Tradeoff
Given that optimizing for all four platforms requires different tactics, most teams face a resource allocation decision:
Broad coverage approach: Ensure your pages meet baseline requirements for all platforms (server-side rendering, schema markup, sitemap freshness, Bing submission). This gets you into the candidate pool everywhere but does not guarantee citations anywhere.
Deep platform approach: Pick the 1 to 2 platforms where your audience is most active and optimize aggressively for those. Accept lower visibility on others.
The data from Lee (2026) suggests the broad coverage approach has a higher expected value for most sites because the baseline requirements overlap more than the citation-level signals do. Server-side rendering helps with ChatGPT, Claude, and Perplexity. Schema markup helps with all four. Sitemap freshness helps with Perplexity and (indirectly) Google-based platforms.
The Bottom Line: The minimum viable AI optimization strategy includes: (1) server-side rendering, (2) schema markup with dateModified, (3) XML sitemap submission to both Google and Bing, and (4) monitoring across at least two platforms. Anything less leaves you blind to the fragmentation the data reveals.
For a complete walkthrough of these tactics, see our Generative Engine Optimization Guide. To check your current visibility across platforms, try the AI Visibility Quick Check.
🧪 METHODOLOGY NOTES
Transparency on how these numbers were produced:
- Query selection: 50 queries drawn from Google Autocomplete across 8 industry verticals (healthcare, finance, e-commerce, SaaS, legal, real estate, education, local services).
- Repetition design: Each query submitted 3 times per platform on different days to measure within-platform consistency.
- Platforms tested: ChatGPT (GPT-4o with web search), Claude (with search enabled), Perplexity (default mode), Gemini (with Google grounding).
- Citation extraction: All cited URLs extracted programmatically from platform responses. Only explicit inline citations counted (not knowledge-based claims without attribution).
- Overlap metric: Jaccard similarity index (intersection over union of URL sets) computed for all pairwise platform combinations and for within-platform repetition pairs.
- Google comparison: Google Top-3 organic results collected via SERP API for each query. URL-level match computed against each platform's citation set.
The larger dataset from Lee (2026) covering 19,556 queries across the same platforms produced consistent overlap metrics, confirming that the 50-query consistency subset is representative.
For a discussion of how AI consensus patterns form across platforms, see How AI Platforms Build Consensus.
❓ FREQUENTLY ASKED QUESTIONS
If only 1.4% of citations overlap, does that mean I need completely different content for each platform?
No. The same content can get cited by multiple platforms, but you need to make it discoverable and accessible to each platform's retrieval system. The content itself can be identical. The distribution and technical accessibility need to be platform-aware. Submit your sitemap to both Google and Bing. Ensure server-side rendering. Use schema markup. These baseline steps increase your chances across all platforms without requiring separate content.
Which platform pair has the most citation overlap?
Perplexity and Gemini share the highest pairwise overlap at 2.4% (Jaccard = 0.024). Both maintain their own indexes that tend to converge on high-authority, well-known sources. The lowest overlap is between ChatGPT and Gemini at 1.2%, reflecting their architecturally distant retrieval methods (Bing live-fetch vs. Google index).
Does ranking on Google help me get cited by ChatGPT?
Barely. ChatGPT's URL overlap with Google's Top-3 results is only 7.8% (Lee, 2026). ChatGPT discovers URLs through Bing, not Google. Your Google rank has almost no direct effect on ChatGPT citation probability. If you want ChatGPT citations, focus on Bing indexation and clean, server-rendered HTML.
How often does the same platform cite the same sources for repeated queries?
Within-platform consistency ranges from 58.3% (Claude) to 64.7% (Perplexity). This means if your page gets cited once, there is roughly a 60% chance it will be cited again for the same query on the same platform. Perplexity is the most consistent because its pre-built index provides more stable retrieval than live-fetching approaches.
What is the minimum I should do to cover all four platforms?
Four baseline steps cover the broadest ground: (1) server-side rendering so all crawlers can read your content, (2) schema markup with dateModified timestamps, (3) XML sitemap submission to both Google Search Console and Bing Webmaster Tools, and (4) visibility monitoring on at least two platforms (we recommend ChatGPT and Perplexity as the most architecturally different pair). For a free check of your current status, use our AI Visibility Quick Check.
📚 REFERENCES
- Aggarwal, P., Murahari, V., Rajpurohit, T., Kalyan, A., Narasimhan, K., & Deshpande, A. (2024). "GEO: Generative Engine Optimization." KDD 2024. DOI
- Hoechstoetter, N., & Lewandowski, D. (2015). "What Users See: Structures in Search Engine Results Pages." arXiv preprint. arXiv
- Lee, A. (2026). "Query Intent, Not Google Rank: What Best Predicts AI Citation Behavior." Preprint v5. DOI