> **来源：[研报客](https://pc.yanbaoke.cn)** # AI for Science: Summary of the State of AI Use Among Researchers ## Core Content This report, published in 2026, presents an in-depth analysis of AI adoption and usage among researchers globally, based on a survey conducted by Springer Nature, Fudan University, and the Shanghai Academy of AI for Science. It explores how AI is integrated into the research workflow, the role of different AI tools, and the funding and trust dynamics surrounding their use. ## Main Findings ### 1. AI Adoption and Penetration - AI is most frequently used for **information- and text-intensive tasks**, such as **literature search and discovery** and **improving or editing papers**, and least used for **judgment-based, collaborative, or administrative tasks**. - **Younger researchers** tend to use AI more frequently than their senior colleagues, especially for content discovery and paper editing. - The use of AI is **uneven across the world**, with **East Asian countries** (particularly **China, Japan, and South Korea**) showing significantly higher rates of AI use compared to other regions. - **Institutional funding** for AI tools is **rare**, with most researchers using **free or self-funded tools**. ### 2. AI Tools in the Research Workflow - **General-purpose large language models (LLMs)** dominate AI usage, accounting for **75.9% of all mentions**. - **ChatGPT** is the most commonly used tool, followed by **Gemini**, **Claude**, and **DeepSeek**. - **DeepSeek** is used **much more frequently in China** (48.8%) than in other countries (usually below 10%). - **Microsoft Copilot** is the only tool with **predominant institutional funding** (48.5%), likely due to its integration with enterprise software. - **Free access** is most common for **DeepSeek** (93.9%), while **personal funding** is widespread, especially in **hospitals, self-employed, and academia**. ### 3. Funding for AI Tools - **Institutional funding** accounts for only **11.1%** of AI use, while **free tools** and **personal funding** make up **45.4%** and **41.3%**, respectively. - **Personal funding** is the most common, suggesting that researchers perceive **high value** in AI tools, even if they are not institutionally supported. - **Unclear** funding sources account for **2.2%** of responses, indicating a lack of transparency in how access is financed. ### 4. Heavy AI Users for Content Discovery - Approximately **5.7% of researchers** rely **primarily or exclusively** on AI for content discovery, a group referred to as **'heavy users'**. - **East Asian countries** have **twice the rate** of heavy users compared to the global average. - **Younger researchers** are more likely to be heavy users than senior researchers. - **Personal payment** is more strongly associated with heavy use than institutional funding. ### 5. AI Satisfaction and Concerns - **65.2% of researchers** are **satisfied or very satisfied** with AI tools for content discovery. - **Satisfaction** is driven by AI’s ability to **identify relevant research** and **provide supporting references**. - **Major concerns** include **accuracy and hallucination**, **loss of critical thinking**, and **research integrity**. - **Productivity gains** and **literature discovery** are the most frequently cited hopes for AI in research. ## Key Insights - AI is seen as a **supportive tool**, not a **replacement for human judgment**. - **Trust remains limited** due to concerns about **accuracy and data security**. - **Regional differences** are significant, with **East Asia** leading in AI adoption. - **Institutional support** is still **underdeveloped**, and **access to advanced tools** depends largely on **personal funding**. - **General-purpose models** are the **primary entry point** for AI use, while **specialized tools** are still relevant in specific contexts. ## Conclusion The report highlights a **growing reliance on AI** in research, particularly in **literature search and paper editing**, with **East Asian researchers** being the most active. While AI is widely used and perceived as beneficial, **trust issues** persist, especially regarding **accuracy and integrity**. The **funding landscape** is dominated by **free and self-funded tools**, with **institutional support** still playing a **limited role**. The findings suggest a **gradual shift** toward the use of **general-purpose AI models**, but **specialized tools** will continue to have a **complementary role** in research workflows.