AI diffusion (science)
AI adoption in European patents and scientific publications
AI
Science
Patents
Tracker measuring the diffusion of artificial intelligence in EPO patents and scientific publications. Data: REGPAT (OECD), OpenAlex, patCit.
AI patents per year
Number of EP patents citing publications classified as AI (observed + predicted by ML).
Share of AI patents (% of total)
Share of EP patents containing at least one NPL citation identified as AI.
Top 10 countries — AI patents (inventors)
REGPAT fractional weights, inventor geography.
Top 10 technology fields (IPC4)
Number of AI patents per IPC4 group.
Trend in AI patents — observed vs. predicted
Observed: NPL citations linked to an OpenAlex publication (patCit, DOI written in the citation, or title found in arXiv, Crossref or Semantic Scholar) whose primary topic is in the AI list. Predicted: NPL citations that could not be linked to any publication, classified as AI by an ML model applied to their title.
Corpus: EP patents with at least one NPL citation, dated by their first publication. Source: Google Patents, patCit, OpenAlex, REGPAT 202602.
2024 provisional: some of the patents published in 2024 do not yet have a search report in the data; their number will be revised upward, while the share of AI patents is less affected.
2024 provisional: some of the patents published in 2024 do not yet have a search report in the data; their number will be revised upward, while the share of AI patents is less affected.
AI share of patents (%, by year)
Percentage of EP patents classified as AI among all patents with NPL citations: all AI patents, and those with an observed AI citation.
Total NPL corpus vs. AI patents
Absolute size of the corpus of patents with non-patent literature (NPL) citations. The log scale makes it possible to compare growth rates.
AI patents by country — top 25 (inventor geography)
REGPAT fractional weights. Each patent contributes to several countries in proportion to its inventors. “AI share” ranks countries by the share of their patents that are AI.
Source: REGPAT 202602 · EPO_Inv_reg · Weighting: reg_share × inv_share.
Total volume and share of AI patents by country
EP patents of the country (horizontal axis, log scale) and share of AI patents (vertical axis). Bubble size follows the number of AI patents. Top left: countries relatively specialized in AI.
Europe map — AI patents by NUTS1 region
REGPAT fractional weights by major region (NUTS1). Share of Europe: the region’s share of AI patents across European regions. Local AI share: share of the region’s patents that are AI.
World map — AI patents by country
Worldwide distribution of patents, 1990–2024, REGPAT weighting by inventors or applicants.
AI patents by IPC4 field — volume and intensity
Volume: number of AI patents in the field. Intensity: share of the field’s patents that are AI.
IPC4 = first 4 characters of the International Patent Classification.
Trend over time — top 8 IPC4 (AI patents per year)
Diffusion of AI across technology fields, 1990–2024.
Composition of AI patents by IPC4 field (% per year)
Share of each field among the AI patents of the top 8 IPC4 fields (total = 100%), 1990–2024. A patent can belong to several fields.
AI scientific publications — annual trend
AI publications (OpenAlex primary topic) cited by the AI patents in the corpus and linked to an OpenAlex publication (patCit, DOI or title found), by the article’s year of publication. Each publication is counted once.
Only citations linked to an OpenAlex publication are counted; citations predicted by the model are not. A recent publication appears only once it has been cited by an already published patent: the series does not measure recent scientific output. OpenAlex dates part of the IEEE archives to late 2002; for these publications, the year given in the patent citation is used.
Top 20 AI research topics
OpenAlex primary topics of the AI publications cited by the patents in the corpus (observed citations). Number of distinct publications.
Top 25 applicants — European AI patents
Applicants with the largest number of European AI patents (1990–2024). Subsidiaries of the same group are combined (e.g., Microsoft Corporation and Microsoft Technology Licensing). The fractional equivalent accounts for joint filings.
Source: REGPAT 202602, patCit, OpenAlex · Observed and predicted AI patents, each patent counted once.
Top 25 inventors — European AI patents
Most prolific inventors in EPO patents citing AI scientific publications. The fractional equivalent weights by the number of co-inventors.
Source: REGPAT 202602, patCit, OpenAlex · Fractional weighting per patent.
Most cited scientific publications
Publications most cited by European AI patents — ranked by number of distinct citing patents.
Source: patCit (EPO), Google Patents, OpenAlex · Only citations linked to a publication are counted.
Trend in AI patents — top 15 applicants (1990–2024)
Number of AI patents per year for each main applicant (fractional equivalent). Shows the rise of new players and the decline of older ones. Choose the applicants to compare.
Source: REGPAT 202602, patCit, OpenAlex · Top 15 applicants by cumulative volume 1990–2024, subsidiaries combined.
About this tracker
This dashboard analyzes European patents (EPO) between 1990 and 2024 that cite scientific publications identified as artificial intelligence, as well as the main players (firms, inventors and publications) at the heart of this dynamic.
Author
This tracker is developed by Gabriela Terra — HEC Paris.
What this tracker measures
| Tab | Content |
|---|---|
| Overview | Overall KPIs, AI patents per year, top countries and IPC fields |
| AI patents | Observed (direct citation) vs. predicted (ML model) |
| Geography | Distribution by inventor country (REGPAT) and NUTS1 maps |
| IPC fields | AI intensity by technology class (IPC4) |
| AI publications | AI scientific publications cited by patents |
| Key players | Top applicants, inventors and most cited publications |
Identifying AI patents
A patent is classified as an AI patent using two complementary methods:
- Observed : the patent cites at least one publication whose primary topic in OpenAlex is one of 34 AI topics. The link is made through patCit (EPO), through the DOI written in the citation, or by finding the cited title in arXiv, Crossref or Semantic Scholar.
- Predicted : for NPL citations that could not be linked to any publication, an ML model classifies the title of the cited publication as AI or not (threshold chosen to reproduce the number of AI citations observed on recent patents). The patent is predicted AI if at least one of its citations is classified as AI. The model first serves to flag the citations to check: those it deems possibly AI are searched for in bibliographic databases, and only those that cannot be found keep its prediction.
Counting
- One patent = one EP application. The EPO often publishes the same patent several times (A1, A2, A3, B1…); these publications are grouped and counted only once, in the year of the first publication.
- Geography (countries, regions) uses REGPAT fractional weighting; each patent weighs 1 in total.
- Scientific publications (AI publications and Key players tabs) can only be identified for observed citations.
Detailed method: methodological note Artificial intelligence in the service of innovation (Bergeaud, Delbouve, Terra, 2026).
Sources
| Source | Role |
|---|---|
| Google Patents | NPL citations of EP patents (raw text), publications up to September 2026 |
| patCit | Resolution of non-patent literature citations (up to about 2018) |
| arXiv, Crossref, Semantic Scholar | Search for cited publications without a DOI |
| OpenAlex | Classification of publications (AI topic) |
| REGPAT (OECD 202602) | Geolocation of inventors (fractional) |
Update
Annual, after a new extraction of citations (Google Patents) and, where applicable, a new REGPAT edition. Data as of October 1, 2026.