--- title: "Do Gemini, GPT and Claude prefer the same kind of content?" description: "Mostly yes on content quality: Gemini, GPT and Claude share most preferences. But they cite different sources, so one good page will not show up everywhere." canonical: "https://underneath.agency/resources/do-ai-engines-prefer-same-content" published: 2026-10-11 updated: 2026-10-11 publisher: "Underneath (https://underneath.agency/agent)" entity: "https://underneath.agency/.well-known/entity.json" --- Guide · AI search # Do Gemini, GPT and Claude prefer the same kind of content? Mostly, yes: in controlled tests, Gemini, GPT and Claude rewarded largely the same qualities in a page. Where they differ is in which sources they go looking for, how many they cite and how sensitive they are to changes. One well-built page can serve all of them, but showing up in each engine still depends on where that engine searches. ## The short version 1. In a lab test, 78.95% to 84.21% of the content preferences found for Gemini, GPT and Claude were shared between each pair ([Wu and colleagues](https://arxiv.org/abs/2510.11438)). 2. Topic moved preferences more than engine did: under Gemini, rules for shopping questions overlapped with rules for research questions by only 34.78% to 40.00%. 3. Engines differ in sources: 93.5% of ChatGPT’s sources for well-known brands were independent publishers, while 25.1% of Gemini’s were brand sites ([Chen and colleagues](https://arxiv.org/abs/2509.08919)). 4. In [our study of 80 buyer questions](https://underneath.agency/research/ai-citations-google-rankings-study), 8.3% of ChatGPT’s citations ranked in Google’s top 10 for the question, against 25.7% of Claude’s. 5. In [our brand agreement study](https://underneath.agency/research/ai-assistants-brand-agreement-study), 66.3% of the options recommended for a question came from one assistant only. ## How much do the engines’ content preferences overlap? A lot. Researchers at Carnegie Mellon built test AI search engines on Gemini, GPT and Claude ([Wu and colleagues](https://arxiv.org/abs/2510.11438)). They then had AI models explain each engine’s choices and boil them down into rules about what it preferred in a page. On the same set of research questions, Gemini and GPT shared 78.95% of their rules. Gemini and Claude shared 84.21%, as did GPT and Claude. Each engine kept a few preferences of its own, and rules tuned to one engine worked best on that engine. Topic mattered more than engine. Under Gemini, rules for two sets of open research questions overlapped by 88.24%, but rules for shopping questions overlapped with them by only 34.78% to 40.00%. Shopping rules leaned toward practical guidance over in-depth explanation. Read these figures with care. The engines were smaller, cheaper versions (Gemini 2.5 Flash-Lite, GPT-4o mini and Claude 3 Haiku), each choosing among five documents handed to it. The overlap was measured on keywords the researchers labeled by hand. This is a simulation, not a test of live ChatGPT or Gemini. ## What do all of them seem to reward? Substance, stated clearly. Rules shared by all three engines included comprehensive coverage, credible sources, accurate and current facts, and a neutral tone ([Wu and colleagues](https://arxiv.org/abs/2510.11438)). They also shared clear structure, specific evidence and stating the conclusion at the start. We describe [what AI-preferred content looks like](https://underneath.agency/resources/what-content-do-ai-engines-prefer) in more detail. A separate test by researchers at the software company Sprinklr points the same way. They ran 252,000 trials on six AI models, changing one thing at a time in two competing sources ([Vishwakarma and colleagues](https://arxiv.org/abs/2605.25517)). Eleven of 18 content factors, 61%, mattered in at least four of the six models. Four mattered in every model: matching the topic, stating a price, carrying a recent date and appearing earlier in the list of sources. Layout changes alone had no consistent effect. Those results are unpacked in [why an AI cites a rival’s page first](https://underneath.agency/resources/why-ai-cites-competitor-page-first). ## Where do the engines differ? They differ in where they look, how many sources they show and how easily they are moved. University of Toronto researchers classified the sources four AI engines cited for brand questions ([Chen and colleagues](https://arxiv.org/abs/2509.08919)). | Engine | Independent publishers | Brand sites | Social and community | |---|---|---|---| | ChatGPT | 93.5% | 6.5% | 0% | | Claude | 87.3% | 6.8% | 5.9% | | Gemini | 63.4% | 25.1% | 11.5% | | Perplexity | 67.4% | 8.8% | 23.8% | These figures are for well-known brands. The engines also differ in how many sources they cite. In one study of 602 test questions, ChatGPT cited 6.88 sources per question on average and Perplexity 16.35 ([Zhang Kai and colleagues](https://arxiv.org/abs/2604.25707)). ChatGPT, though, drew more heavily on each source it cited. They differ in what signals they respond to. In that study, ChatGPT responded most to relevance to the question. Google responded to closeness in meaning to the question and answer, plus definitions; Perplexity to relevance, headings and length. In the Sprinklr test, 50% of factors mattered for Claude 3.5 and only 33% for Gemini 2.5. ## If preferences overlap, why do the engines cite different pages? Partly because they search differently, and partly because they choose differently from the same evidence. In [our study of 80 buyer questions](https://underneath.agency/research/ai-citations-google-rankings-study), 8.3% of ChatGPT’s citations ranked in Google’s top 10 for the question, against 16.6% for Gemini and 25.7% for Claude. The result is little shared ground. In [our brand agreement study](https://underneath.agency/research/ai-assistants-brand-agreement-study), 66.3% of options recommended for a question came from one assistant only. All four assistants agreed on the first pick for 10.0% of questions. Different sources explained little of this: when one assistant recommended an option and another did not, the second assistant’s own cited pages named it 42.1% of the time. Other audits agree. A 2026 survey reports one audit in which only 26% of cited domains were cited by both Bing Chat and Perplexity ([Martinez](https://arxiv.org/abs/2607.14035)). A shared taste in content does not mean a shared shortlist. ## Does a page good enough for one engine work for the others? On one study’s scoring, pages cited by several engines were of higher quality. A study of business software questions audited 1,100 pages cited by Brave, Google’s AI Overviews and Perplexity ([Kumar and Palkhouski](https://arxiv.org/abs/2509.10762)). The engines cited pages of very different quality on the authors’ own scoring: an average of 0.727 for Brave against 0.300 for Perplexity. Pages cited by more than one engine scored 71% higher on quality than pages cited by just one. The authors have a commercial research affiliation, and the score is their own. ## Does page structure need tailoring for each engine? Probably not much. A team from Japanese universities grouped six engines, including older products such as Bing Chat, by how they search ([Yu and colleagues](https://arxiv.org/abs/2603.29979)). They then restructured 200 articles for them. The same rewrites raised citation rates by 17.3% on average across all groups. The authors suggest each kind of engine leans on different features, such as a clear summary up front or sections that stand alone. Those weights come from their own predictive model and their own grouping of engines, and a recent survey cautions that the work relies on automated judges. Treat them as hypotheses, not settings to tune. ## What should you do about it? Build one strong page, then work on each engine’s route to it. 1. Write for the shared preferences: a direct answer first, full coverage, specific evidence, credible sources, accurate dates and prices, and a neutral tone. On wording itself, see [whether readable writing helps AI visibility](https://underneath.agency/resources/does-readable-writing-help-ai-visibility). 2. Do not fork content per engine. Most preferences are shared; tailored rules did best in lab tests, but shared rules still helped. 3. Map where each engine looks. ChatGPT leans on independent publishers, Gemini cites more brand sites, Perplexity more community sources. 4. Track each engine separately. A win in one does not carry over, so measure citations engine by engine. 5. Ask your category’s buying questions in each assistant every few weeks and note which sources each one cites. If you want help building pages that work across engines, see our [generative engine optimization service](https://underneath.agency/services/generative-engine-optimization). ## What does the research not tell us yet? No study has compared live versions of Gemini, ChatGPT and Claude on the same pages over time. - The overlap figures come from small, older model versions choosing among five documents, not from live products. - Content preference tests feed engines a fixed set of pages, so they say little about how each engine searches. - Source mixes come from specific question sets, countries and dates, and engines change after updates. - Several studies are vendor-authored or use the authors’ own scoring tools. - Nobody has measured whether the same page earns the same clicks or trust from each engine’s users. ## Frequently asked questions ### Do I need different content for ChatGPT, Gemini and Claude? Mostly no. In a lab test, 78.95% to 84.21% of content preferences were shared between each pair of engines. ### Why does ChatGPT cite different sources than Gemini? They search differently. For well-known brands, 93.5% of ChatGPT’s sources were independent publishers, while Gemini cited brand sites 25.1% of the time. ### Does ranking in Google help with every AI assistant equally? No. In our study, 25.7% of Claude’s citations ranked in Google’s top 10 for the question, against 8.3% of ChatGPT’s. ### Will optimizing for one AI engine hurt my visibility in another? Not on current evidence. Rules learned from one engine still improved results on others in lab tests, though tailored rules worked best on their own engine. ## Sources - Wu, Zhong, Kim and Xiong (2025), [What Generative Search Engines Like and How to Optimize Web Content Cooperatively](https://arxiv.org/abs/2510.11438), arXiv:2510.11438. - Vishwakarma, Kumar and Jamidar (2026), [What Gets Cited: Competitive GEO in AI Answer Engines](https://arxiv.org/abs/2605.25517), arXiv:2605.25517. - Chen, Wang, Chen and Koudas (2025), [Generative Engine Optimization: How to Dominate AI Search](https://arxiv.org/abs/2509.08919), arXiv:2509.08919. - Zhang Kai, He Xinyue and Yao Jingang (2026), [From Citation Selection to Citation Absorption: A Measurement Framework for Generative Engine Optimization Across AI Search Platforms](https://arxiv.org/abs/2604.25707), arXiv:2604.25707. - Kumar and Palkhouski (2025), [AI Answer Engine Citation Behavior An Empirical Analysis of the GEO16 Framework](https://arxiv.org/abs/2509.10762), arXiv:2509.10762. - Yu, Yang, Ding and Sato (2026), [Structural Feature Engineering for Generative Engine Optimization: How Content Structure Shapes Citation Behavior](https://arxiv.org/abs/2603.29979), arXiv:2603.29979. - Martinez (2026), [Optimizing Visibility in Generative Engines: A Critical Survey of Generative Engine Optimization (2023-2026)](https://arxiv.org/abs/2607.14035), arXiv:2607.14035. - Underneath (2026), [Do ChatGPT, Gemini, Perplexity and Claude cite pages that rank?](https://underneath.agency/research/ai-citations-google-rankings-study) - Underneath (2026), [Do ChatGPT, Gemini, Perplexity and Claude agree on brands?](https://underneath.agency/research/ai-assistants-brand-agreement-study) --- This is the Markdown twin of https://underneath.agency/resources/do-ai-engines-prefer-same-content. The HTML page is canonical. Publisher: Underneath, https://underneath.agency/agent. Site index: https://underneath.agency/llms.txt.