The Regional Spillover Effects on Innovation From Generative Pre-Trained Transformer (GPT) and Large Language Models (LLMs)

Research by: Daniel Broby & Mario Cerilles Jr.

Executive Summary

The release of ChatGPT by OpenAI constitutes a major technological event in the evolution of artificial intelligence. As a generative pre-trained transformer built on large language model architecture, ChatGPT substantially widened public, commercial, and developer awareness of AI systems. This article examines whether that technological event generated measurable innovation spillovers, as reflected in subsequent GPT and LLM-related patent filings in the United States and Europe.

The study is motivated by a central question in innovation economics: whether the disclosure and diffusion of a major frontier technology induce follow-on inventive activity. Also, whether institutional systems differ in their capacity to capture that activity through formal intellectual property channels. To address this question, the authors employ a Difference-in-Differences event-study design. They compare GPT and LLM-related patent filings with a control group of unrelated computing patents over a two-year window surrounding the launch of ChatGPT. The empirical design therefore separates the specific effect associated with the ChatGPT release from general trends in computing-related patenting.

The results show a marked post-release increase in GPT and LLM-related patent filings in the United States. This finding is consistent with the interpretation that ChatGPT acted as an innovation shock, stimulating follow-on applications, technical adaptations, and commercially oriented inventive activity. By contrast, the corresponding effect in Europe is weaker. The article interprets this divergence as evidence that institutional conditions shape the translation of technological opportunity into patentable innovation. Differences in patent eligibility criteria, procedural complexity, examination practice, and the broader regulatory environment appear to affect the speed and extent with which AI-related inventions are formalised.

The contribution of the article lies in connecting generative AI diffusion with measurable intellectual property outcomes. Rather than treating AI innovation as a uniform global process, the paper shows that regional legal and institutional architectures condition the observable response to a common technological event. In doing so, it advances the literature on innovation spillovers, artificial intelligence, and comparative technology policy.

The legal contribution of the article is to position patent law not merely as a passive registration mechanism, but as an institutional determinant of innovation capture. By comparing the United States and Europe, the study shows that doctrinal standards governing software, algorithms, and AI-related inventions may influence the observed geography of patenting after a technological shock. The analysis is therefore relevant to debates on patent eligibility, examination consistency, and the treatment of computer-implemented inventions. It suggests that legal uncertainty can attenuate the translation of technical experimentation into formal intellectual property claims, particularly in fields where the boundary between abstract computation and patentable technical contribution remains contested.

The policy implication is clear. Patent systems that reduce unnecessary procedural frictions, provide greater clarity on software and AI-related eligibility. They support timely examination may be better positioned to formalise innovation after major technological shocks. If jurisdictions seek to capture the economic value generated by frontier AI, they must consider the institutional mechanisms through which inventions are protected, financed, and commercialised.

For Europe, the findings suggest that regulatory caution must be balanced against the need to maintain competitiveness in AI development. For the United States, the results reinforce the importance of an institutional environment that allows rapid technological advances to be translated into recognised intellectual property.

Overall, the article provides timely empirical evidence on how generative AI reshapes innovation incentives and how jurisdictional differences influence the capture of AI-driven economic value.

To cite this article:

Broby, D. & Cerilles, M., Jr. (2026). The regional spillover effects on innovation from generative pre-trained transformers (GPT) and large language models (LLMs). Economics of Innovation and New Technology. https://doi.org/10.1080/10438599.2026.2676054

To access the article:

https://doi.org/10.1080/10438599.2026.2676054

About the journal