--- title: "Do ChatGPT, Copilot, Google and Perplexity cite the same sources?" description: "No. Studies find ChatGPT, Copilot, Google and Perplexity cite largely different pages for the same question, so success on one engine rarely carries over." canonical: "https://underneath.agency/resources/do-ai-search-engines-cite-the-same-sources" published: 2026-10-10 updated: 2026-10-10 publisher: "Underneath (https://underneath.agency/agent)" entity: "https://underneath.agency/.well-known/entity.json" --- Guide · AI search # Do ChatGPT, Copilot, Google and Perplexity cite the same sources? No. For the same question, the major AI search engines cite mostly different pages, and often entirely different websites. Being cited by one engine says little about whether the others will cite you. ## The short version 1. In a one-day test of 15 commercial prompts, 96.4% of the cited web addresses appeared in only one of four engines, [a vendor-run study found](https://arxiv.org/abs/2609.22655). 2. For the same product question, ChatGPT and Gemini shared on average only 5.4% of the sites they displayed, in [a 2026 audit of product recommendations](https://arxiv.org/abs/2609.18729). 3. Even Google’s own products disagree: [our comparison](https://underneath.agency/research/ai-mode-vs-ai-overviews-study) found AI Mode and AI Overviews shared 13.9% of cited pages on the same searches. 4. The engines differ more from each other than from themselves on another day: the vendor study found same-engine results about 42 times more similar than cross-engine results. ## Do AI search engines cite the same pages for the same question? Rarely: most cited pages appear on only one engine. [Tannenbaum](https://arxiv.org/abs/2609.22655) ran 15 commercial prompts through ChatGPT, Microsoft Copilot, Google and Perplexity on 6 June 2026. Among 528 distinct cited web addresses, 509 appeared in only one engine. That is 96.4% of all observed addresses. Pairs of engines rarely overlapped at all. Across 73 same-prompt comparisons, 84.9% of engine pairs shared no cited address. The author founded a company that sells AI visibility software, and his data covers one niche on one day, so treat the exact figures with care. Independent studies point the same way. [Uberti-Bona Marin and colleagues](https://arxiv.org/abs/2609.18729) asked 117 product questions in the Netherlands in September 2026. For the same question, ChatGPT and Gemini shared only 5.4% of their displayed sites on average, and 76.7% of comparisons shared no site at all. ## Do the engines at least agree on their top sources? No, the disagreement holds at the top of the list too. In the Tannenbaum data, no pair of engines shared a single address among their first five citations, across all 60 comparisons on the ten prompts every engine answered. No single engine gives a full picture either. ChatGPT captured 42.6% of the addresses cited across all four engines, Google 24.3%, Perplexity 23.6% and Copilot 11.4%. Watching one engine means missing most of the sources the others use. Engines also differ in [how many sources they cite per answer](https://underneath.agency/resources/how-many-sources-ai-search-engines-cite). Our own data adds a check on Google. When AI Mode and AI Overviews both answered the same search, [they led with the same source in 20.9% of searches](https://underneath.agency/research/ai-mode-vs-ai-overviews-study). ## Do Google’s own AI products agree with each other? Only partly, even though they share Google’s index. In our study of 400 searches, AI Mode and AI Overviews shared a mean 13.9% of cited pages where both cited sources. In 30.5% of those searches they had no cited page in common. Gemini differs again. [Grossman and colleagues](https://arxiv.org/abs/2604.27790) compared Google Search, AI Overviews and Gemini on 11,500 queries. On average, only 18% of the sources returned by either AI Overviews or regular search were retrieved by both, and AI Overviews and Gemini were the least similar pair. A larger study by [Zhang and colleagues](https://arxiv.org/abs/2512.09483) covered 55,936 queries across six AI engines and Google and Bing in mid-2025. Only 38% of cited domains appeared in both AI and traditional results. We look further at [how AI citations compare with Google’s results](https://underneath.agency/resources/do-ai-search-engines-cite-the-same-sites-as-google). ## Are the differences just random noise? No: engines differ far more from each other than from their own results on another day. Tannenbaum found that the same engine’s cited addresses changed by 67.0% between 5 and 6 June. Even so, one engine’s list was about 42 times more similar to its own list a day earlier than to another engine’s list on the same day. Engine family matters. [Yang](https://arxiv.org/abs/2507.05301) studied news citations in an AI search comparison site in 2025. Models from the same company scored 0.82 to 0.99 on a similarity scale where 1 means identical news sources. Across companies, scores fell to between 0.11 and 0.58. Our [four-assistant study](https://underneath.agency/research/ai-assistants-brand-agreement-study) of 80 buyer questions found the same split. In 43.8% of question pairs, two assistants shared no cited domain at all. ## Does it matter which version of an engine you test? Yes. The app a customer uses and the developer version a tracking tool queries can cite different sites. In the product audit, ChatGPT’s app and its developer version shared an average of 12.0% of their displayed sites for the same question, asked moments apart. For 60.9% of pairs, they shared no site. This matters for measurement. [A dashboard built on developer access](https://underneath.agency/resources/api-ai-visibility-monitoring) may not show what buyers actually see in the app. Asking the same engine again also changes the list. In the same audit, two runs of one question shared on average 26.0% of displayed sites in ChatGPT, 29.8% in Gemini and 45.9% in AI Overviews. For ChatGPT, the number of distinct sites seen per question rose from 2.65 after one request to 5.56 after three. A single answer is one sample, not the engine’s settled view. We explain [why citations change between checks](https://underneath.agency/resources/why-ai-search-citations-change) separately. ## Is anything shared across engines? A small core of well-known publishers, and little else. [Chen and colleagues](https://arxiv.org/abs/2509.08919) compared Claude, ChatGPT and Perplexity on car questions in August 2025. Each engine’s sources were mostly its own: 50.3% of Claude’s domains, 60.8% of ChatGPT’s and 56.5% of Perplexity’s were exclusive to that engine. The domains all three shared were mainly large review outlets such as Car and Driver, Edmunds and Consumer Reports. In consumer electronics, the shared core was TechRadar, Tom’s Guide and RTINGS. That study used developer versions of the engines and thanked a GEO vendor for support. Rankings differ too. In [our rankings study](https://underneath.agency/research/ai-citations-google-rankings-study), 8.3% of the pages ChatGPT cited ranked in Google’s top 10 for the question, against 25.7% for Claude. ## What should you do about it? Treat each AI engine as its own channel, with its own sources to win. 1. Decide which engines your buyers use, and measure each one separately. If you need one number, see [how to combine engines into one score](https://underneath.agency/resources/combine-ai-engines-visibility-score). 2. Do not read one engine’s results as a proxy for the others. 3. Test the consumer apps, not only developer versions, because their sources differ. 4. Repeat measurements over several days, since one engine’s sources change daily. 5. Prioritize the few publishers that every engine cites in your category. They are the closest thing to a shared foundation. If you want help measuring across engines, see our [generative engine optimization service](https://underneath.agency/services/generative-engine-optimization). ## What does the research not tell us yet? The research shows engines diverge, but not why, nor how that changes over months. - The sharpest figures come from small samples, such as 15 prompts on one day from a company with a commercial interest. - Studies use different units: exact pages, domains, displayed sources or news outlets only. - Few studies cover business-to-business questions, so divergence in those categories is unmeasured. - No study we reviewed tracks whether divergence is growing or shrinking over time. - Citation is not the same as influence: the sources an engine displays are a small part of what it reads. ## Frequently asked questions ### If ChatGPT cites my site, will Perplexity cite it too? Not necessarily. In one four-engine test, 96.4% of cited addresses appeared on only one engine, though that test covered just 15 prompts. ### Do Gemini and Google AI Overviews use the same sources? No. A study of 11,500 queries found AI Overviews and Gemini were the least similar pair among Google Search, AI Overviews and Gemini. ### Why do AI engines cite different websites? Each engine runs its own searches and applies its own selection. News citations from models by the same company scored 0.82 to 0.99 on similarity, while cross-company scores fell as low as 0.11. ### Should I track every AI engine separately? Yes, if your buyers use more than one. The broadest single engine captured only 42.6% of the addresses cited across four engines in one test. ## Sources - Tannenbaum (2026), [Scoring With the Engine: Retrieval Exposure, Cross-Engine Divergence, and the Limits of Engine-Agnostic GEO Scores](https://arxiv.org/abs/2609.22655), arXiv:2609.22655. - Uberti-Bona Marin et al. (2026), ["If I Had to Buy Just ONE: Galaxy S26 Ultra": Auditing AI-Generated Product Recommendations](https://arxiv.org/abs/2609.18729), arXiv:2609.18729. - Grossman et al. (2026), [How Generative AI Disrupts Search: An Empirical Study of Google Search, Gemini, and AI Overviews](https://arxiv.org/abs/2604.27790), arXiv:2604.27790. - Zhang et al. (2025), [Source Coverage and Citation Bias in LLM-based vs. Traditional Search Engines](https://arxiv.org/abs/2512.09483), arXiv:2512.09483. - Yang (2025), [News Source Citing Patterns in AI Search Systems](https://arxiv.org/abs/2507.05301), arXiv:2507.05301. - Chen et al. (2025), [Generative Engine Optimization: How to Dominate AI Search](https://arxiv.org/abs/2509.08919), arXiv:2509.08919. - Underneath (2026), [AI Mode vs AI Overviews: how different are the sources?](https://underneath.agency/research/ai-mode-vs-ai-overviews-study) - Underneath (2026), [Do ChatGPT, Gemini, Perplexity and Claude agree on brands?](https://underneath.agency/research/ai-assistants-brand-agreement-study) - Underneath (2026), [Do ChatGPT, Gemini, Perplexity and Claude cite pages that rank?](https://underneath.agency/research/ai-citations-google-rankings-study) --- This is the Markdown twin of https://underneath.agency/resources/do-ai-search-engines-cite-the-same-sources. The HTML page is canonical. Publisher: Underneath, https://underneath.agency/agent. Site index: https://underneath.agency/llms.txt.