More than half of the 'unicorn companies' in the AI field have never published a peer-reviewed paper or preprint.

AI is a rapidly growing field in both science and business, and has produced numerous unicorn companies valued at over $1 billion (approximately 16 billion yen). However, it has been pointed out that more than half of the unicorn companies in the AI field do not publish their research results in the form of peer-reviewed papers or preprints.

Negligible participation in the scientific literature of AI unicorn startups | bioRxiv
https://www.biorxiv.org/content/10.64898/2026.07.15.738744v1

AI's top startups are barely publishing their research | Science | AAAS
https://www.science.org/content/article/ai-s-top-startups-are-barely-publishing-their-research

Among the AI startups emerging, primarily in developed countries, some boast that their technology will bring about a major revolution in software development, drug discovery, and scientific research. Therefore, a research team led by Professor John Ioannidis of Stanford University School of Medicine investigated how these AI startups publish their research findings in scientific literature and advance science.

Ioannidis has long been a researcher of the integrity and ethics of scientific research, and in 2005, he attracted attention with his paper ' Why Most Published Research Findings Are False, ' which questioned the reproducibility of medical research. In 2015, he also pointed out the scarcity of peer-reviewed papers published by Theranos , a blood testing startup that was later revealed to be a massive fraud.

In 2019, we reported on the practice of 'citation farming,' where authors cite their own papers in order to increase the number of citations, which is an indicator of paper evaluation.

What is the reality of 'inflating citation counts' that distorts the value of academic papers? - GIGAZINE

This study identified 317 unicorn AI companies that existed between 1998 and 2025, and searched for related academic papers and preprints. It then selected papers where company researchers played a major role as first or last author, examining the extent to which startups contribute to scientific research. The final dataset reportedly includes 1,389 peer-reviewed papers and 688 preprints.

The analysis revealed that more than half of AI startups have not published a single paper that meets the requirements. Looking at the number of citations, the top 5% of companies account for more than 90% of all citations, with OpenAI accounting for approximately 40% of these, followed by the Chinese computer vision company Megvii and the AI platform Hugging Face. OpenAI has approximately 4,500 employees, but only 8 researchers have written five or more papers that meet the requirements.

Regarding these results, Ioannidis questioned, 'It seems like a very strange contradiction that in a field that is said to be revolutionizing the way science is conducted and is so advanced in terms of scientific potential, there is absolutely no scientific backing. How can we determine that their claims are true, verified, and reproducible?'

According to Nur Ahmed, an AI researcher at the University of Arkansas, this situation is largely due to the structure of the AI industry. For example, in the pharmaceutical industry, researchers can profit by protecting the content they publish in scientific papers with patents, but in the AI industry, there is little profit to be gained from publicly disclosing technological advancements. For instance, in 2017, Google published a paper on Transformer, which forms the basis of today's large-scale language models, and obtained patents for parts of it, but apparently no one has paid them any compensation.

Furthermore, while the peer review process for scientific papers can take months to several years, the AI startup field operates at a much faster pace. Therefore, many AI startups adopt methods of quickly publishing their research results not in the form of academic papers, but in the form of blogs or technical reports. It's important to note that this study does not track these non-academic publications.