AI for Science 2026: The State of AI Use among Researchers

Source: Fudan University / Nature Research Intelligence (Springer Nature) · 2026-07

1. AI assistance concentrates on information- and text-intensive tasks, but its role diminishes where judgment is required.

Information gathering and improving or editing papers are the two activities AI supports most. Nearly half of respondents use AI ‘every time’ or ‘most times’ for each (10.5% and 33.3% for information gathering, and 12.2% and 32.4% for improving or editing papers). By contrast, fewer than a quarter of respondents use AI consistently for reviewing papers (5.8% every time, 17.7% most times). Overall, AI appears to be reinforcing the exploratory and generative stages of research while leaving tasks that rest on accountability and professional judgment largely untouched.

2. Frequent AI use declines with researcher seniority, driven more by senior researchers opting out than by lighter use among active users.

Defining frequent users as those who report using AI ‘every time’ or ‘most times’, for improving or editing papers, the share falls from about 46.8% among researchers with fewer than three years’ experience to 36.1% among those with 20 or more years and from roughly 47.8% to 33.4% for information gathering.

3. General-purpose models dominate, with tool preferences broadly similar worldwide.

When survey participants were asked to name the three AI tools they used most frequently, general-purpose models accounted for 75.9% of all mentions, far ahead of writing and editing tools (7.2%), literature and evidence discovery tools (5.6%), and search tools (5.3%). Among the general-purpose models, ChatGPT leads at 36.8%, followed by Gemini at 19.4% and Claude at 6.5%. This pattern is broadly consistent across countries or regions, with ChatGPT ranking first everywhere. DeepSeek, however, is used much more often in China (48.8%) than in other countries or economies (usually below 10%).

4. Intensive AI use for content discovery is notably more common in East Asia.

Across the pooled sample, about one in seventeen researchers (5.7%) relies primarily or exclusively on AI tools to identify relevant research, and rates in East Asian countries are roughly twice this. This advantage holds across experience levels. Among researchers with fewer than three years’ experience, heavy use reaches 21.2% in Japan and 15.2% in China, compared with 5.7% in the US and 8.9% in Europe. The gap narrows with seniority, but does not disappear: among researchers with more than ten years’ experience, Japan (8.6%) and China (6.0%) continue to outpace the US (5.2%) and Europe (4.3%).

5. Most researchers appear to access AI tools independently and often without incurring costs, while those who use AI more intensively for content discovery are more likely to pay for such tools themselves.

Overall, 45.4% of respondents use their primary AI tool free of charge, 41.3% fund it personally, and only 11.1% report that their institution pays for the license. Free access is most pronounced for DeepSeek (93.9% of its users), while Microsoft Copilot is the sole listed tool accessed predominantly through institutional funding (48.5%). A follow-up regression analysis to understand the associations between heavy use and other factors (Table 3 in Appendix B) indicates that researchers who rely most heavily on AI for content discovery remain more likely to pay themselves, even after accounting for differences in research experience, sector, and discipline.

6. AI tools perform strongly in content discovery, with satisfaction driven primarily by their ability to identify relevant research and provide supporting references.

Nearly two-thirds of respondents (65.2%) claimed they were satisfied or very satisfied with how their primary AI tool supports the discovery of new research content, while 29.5% were neutral, and only 5.3% were dissatisfied. Among the more than 1,000 responses to open-ended questions in the survey, the most frequently cited reason for satisfaction was the capacity to point researchers to credible sources and uphold findings with references (235 mentions), followed by accuracy of outputs (148 mentions), as well as productivity gains (90 mentions).

7. Researchers’ responses suggest that AI is primarily viewed as a tool for augmenting human capabilities, rather than as a substitute for human judgment.

Meanwhile, concerns over accuracy and integrity limit their trust. Across more than 7,500 responses to open-ended questions, AI was described as making research faster, more accessible, and more equitable. The most frequently cited hopes were gains on productivity (3,442 mentions) and literature discovery (2,019 mentions). Concerns, however, were even more pronounced: accuracy and hallucination issues dominated (4,935 mentions), raised more often than any single hope, followed by worries about loss of critical thinking (1,167 mentions) and research integrity (702 mentions).