Figure 1 illustrates the global distribution of publications related to AIGC following the release of ChatGPT. The data indicate that, since then, an increasing number of nations have engaged in AIGC research. Specifically, over 80% of countries/regions in Asia and Europe have contributed to this body of work. African countries with relevant outputs in 2023 grew by 750%, with 24 African countries having published AIGC-related papers as of the data collection date. This trend demonstrates that AIGC research and applications are no longer concentrated in a few technology-developed regions but are spreading worldwide.

The data suggest a regional imbalance in AIGC-related research. Asia and Europe have the highest number of publications, while the Americas have the highest percentage of AIGC-related papers, which is close to 1%. Notably, AIGC-related papers in Asia are mainly concentrated in China (46%), while in the Americas they are mainly concentrated in the United States (65%), reflecting the fact that AI technology has become the focal point of the global technology competition landscape. Although Oceania has fewer publications overall, its contribution from Australia and New Zealand is notable. In contrast, smaller nations in Oceania have yet to significantly engage in AIGC research, likely due to limited resources. Africa, with 266 papers from 24 countries, faces constraints in research resources and infrastructure, potentially slowing AIGC development. Nonetheless, many African nations are making efforts to explore AIGC’s societal applications and build their technological capacities, even if only through an article.

Figure 2 illustrates the keywords co-occurrence network and the evolving trends of subject terms in global AIGC literature. Over time, researchers’ focus has shifted from early terms like ‘conversational agent’ to ‘Chatbot’, and more recently to ‘ChatGPT’, eventually broadening to ‘generative AI’. This progression reflects the advancement of human-computer interaction from basic conversational robots to sophisticated generative AI platforms, highlighting ongoing improvements in user experience, functional complexity, and application scenarios. Technologically, attention has transitioned from early concepts such as ‘deep learning’, ‘GAN (generative adversarial networks)’, and ‘CNN (convolutional neural network)’ to ‘NLP (natural language Ppocessing)’ and ‘transformer’ architectures, and now to ‘LLM (large language models).’ This shift demonstrates the evolution of AI from traditional deep learning methods to generative AI technologies centered around NLP, particularly the development of Transformer models and LLM.

High-frequency keywords indicate that since the release of ChatGPT, ‘artificial intelligence’ has emerged as the dominant keyword in AIGC research. As a prominent application, ‘ChatGPT’ naturally draws significant attention and discussion within the field. Additionally, the increasing frequency of terms such as ‘LLM’ and ‘generative AI’ in recent literature suggests a growing focus on the fundamental technologies and core architectures of generative AI. This shift highlights a transition in research from application-oriented tools to foundational technologies, signalling an expansion of AIGC studies from single model applications to a broader technical foundation and theoretical framework.

Red nodes around the high-frequency keywords highlight recent trends in AIGC research. On one hand, terms such as ‘education’, ‘higher education’, ‘patient education’, and ‘health’ indicate the expanding application of AIGC technology in education and medicine. This suggests that AIGC is extending beyond science and academic research into various societal domains, showing promise in addressing practical problems. On the other hand, the frequent co-occurrence of keywords like ‘Prompt engineering’, ‘task analysis’, ‘performance’, ‘accuracy’, ‘quality’, and ‘readability’ reflects a growing focus on the performance and precision of AIGC models. Researchers are exploring ways to enhance model output quality, improve content readability, and optimize task execution. This shift underscores the importance of performance optimization as AIGC technology advances from development to practical application. Additionally, the appearance of terms such as ‘anxiety’, ‘challenges’, and ‘future’ indicates that while technological advancements are a focus, researchers are also considering potential challenges and social risks associated with AIGC. The rapid growth of generative AI raises issues related to ethics, security, privacy, and social anxiety, prompting scholars to contemplate future developments and address these concerns.

The size of the circle indicates the frequency of the word; the colour distribution is the time evolution after standardization; the connecting line indicates the strength of the co-occurrence relationship.

Figure 3 illustrates the distribution of AIGC-related research performance across continents and macro topics. The number of published papers indicates that AIGC technology has expanded beyond electrical engineering, electronics and computer science into various disciplines. However, its application in foundational fields such as mathematics and physics remains limited. This suggests that while AIGC technology has driven advancements in various disciplines, its potential in fundamental sciences could become a significant area of interest in the future, particularly for addressing complex mathematical modelling, theoretical validation, and physical simulation problems.

Examining the distribution of AIGC papers across disciplines by continent reveals that Asia leads in publication volume, followed by Europe and the Americas, with Oceania and Africa contributing fewer results. Asia’s leading position may be attributed to its advancing scientific infrastructure, demographic advantages, and strong national policies. Europe and the Americas are notable for their technological innovations and focus on the application of AIGC in social sciences and clinical and life sciences, which reflects their emphasis on the social and ethical implications of AIGC. This focus is supported by robust science and technology policies, social security, and academic freedom in these regions. In Oceania, AIGC research has not yet made significant inroads into Mathematics, potentially due to a concentration of resources in specific application areas and relatively weaker basic theoretical research. Africa’s AIGC research is emerging, with fewer publications across disciplines, and has yet to produce research in engineering and materials science, suggesting constraints in scientific resources and technological development in the region.

Africa and Oceania achieve the highest scores for AIGC research and the combination of electrical engineering, electronics and computer science in terms of IRW and CNCI. This outcome is contrary to initial expectations. Analysis of other indicators reveals that although Africa and Oceania publish fewer articles overall, the high-impact nature of their research results in a relatively high influence around the world. A detailed review of the papers from these regions indicates that many are produced in collaboration with institutions in Asia, Europe, and the Americas. This suggests that, despite limited resources, enhancing international cooperation can significantly boost scientific research output and global impact. Meanwhile, the Americas excel across all indicators, demonstrating strong overall influence. Asia and Europe show similar scores in CNCI, but Asia has a higher number of highly cited papers and a greater proportion of papers in the Documents in the top 1% compared to Europe, highlighting Asia’s superior performance in high-impact research.

In the fields of clinical and life sciences, the findings differ from the latter. Among Asia, Europe, and the Americas, where the volume of publications allows for comparison, the Americas lead in most impact indicators and normalized indicators, followed by Europe and Asia. Despite Oceania’s relatively small number of publications, its proportion of high-impact papers remains notable, while Africa’s performance across all indicators is comparatively lower.

In social sciences, Oceania’s influence stands out distinctly. Despite having only 189 papers, Oceania has a proportion of Documents in the top 1% and top 10% of approximately 15.64% and 35.45%, respectively, significantly surpassing the other continents. Additionally, Oceania’s CNCI is 4.28, the highest among the continents. Africa, with 71 publications in social sciences, has about 9.86% of its papers in the Documents in the top 1%, ranking just behind Oceania. Although Asia, Europe, and the Americas show close performance in social sciences, Europe exhibits a higher citation impact than Asia and the Americas, resulting in relatively higher IRW and CNCI for Europe.

Notably, among the 24 articles published in the field of mathematics in the Americas, the citation impact exceeds 10, and the proportions of Documents in the top 1% and top 10% are 25% and 37.5%, respectively, indicating a strong influence of the Americas in AIGC research within mathematics. The reference value of the measurements for other disciplines is limited due to the small number of papers.

Based on the number of published papers, this article examines the top 10% of countries in each continent and identifies their top 5 micro-topics. The analysis reveals that research from countries in Asia, Europe, and the Americas predominantly centers on five micro-topics: deep learning, artificial intelligence in healthcare and medicine, human-robot interaction, natural language processing (NLP), and open data.

China leads in the number of publications within the field of deep learning. Additionally, NLP is a significant research topic in China, aligning with the country’s rapid advancements in artificial intelligence applications and language models. However, in terms of citation impact, articles from China show slightly lower or similar levels compared to global and disciplinary averages, with a notable disparity when compared to the United States. This suggests that China still faces challenges in accumulating high-quality, high-impact research outcomes. This gap may be attributed to variability in paper quality or the fact that many highly cited papers originate from leading research institutions in other countries.

The United States exhibits exceptional citation influence, particularly in artificial intelligence in healthcare and medicine, as well as deep learning. Notably, in the deep learning field, the U.S. not only publishes a significant number of papers but also achieves substantial global impact. This underscores the United States’ leading position in AI fundamental research and technology application, particularly in practical fields such as medicine and healthcare.

Unlike other countries in Asia, Europe, and the Americas, the UK does not feature open data among its top five research topics. Instead, it focuses on plagiarism, with notable citation performance in this area. This indicates that the UK’s research on AI applications in academic ethics, anti-cheating measures, and educational technology has garnered significant attention within the academic community.

Although African countries have fewer AIGC publications, they have increased their scientific research influence through international collaborations, particularly with Asia, Europe, and the Americas. Egypt’s research aligns closely with that of European and American countries, while South Africa’s focus areas differ. In addition to human-robot interaction and the technology acceptance model, South Africa’s research also addresses specific issues such as health literacy, Sibutramine, and plagiarism.

Australia demonstrates strong citation influence in globally prominent topics such as deep learning, artificial intelligence in healthcare and medicine, and human-robot interaction. Additionally, Australia’s research on plagiarism and Sibutramine is notable, reflecting its emphasis on academic integrity and health technology applications.

Although the number of papers published by different countries across various topics is relatively similar, citation influence varies significantly. For instance, India excels in human-computer interaction and open data. Italy and Spain have notable citation influence in human-computer interaction, with Spain’s NLP papers particularly influential, though its deep learning papers are less impactful. These variations highlight differences in AIGC-related resource allocation, technological strengths, and research needs among countries, providing valuable insights into the global development landscape.