Back

Negligible participation in the scientific literature of AI unicorn startups

Loisel, Q. E. A.; Sandoval Lentisco, A.; Ioannidis, J. E. A.

2026-07-16 scientific communication and education
10.64898/2026.07.15.738744 bioRxiv
Show abstract

Artificial intelligence (AI) development is concentrated within private firms, yet their participation in scientific publishing remains poorly understood. We conducted a bibliometric analysis of all 317 AI unicorn startups (1998-2025). Only 1,389 eligible peer-reviewed publications and 688 preprints involving some leading startup contributions were identified. More than half of startups (52.4%) produced no qualifying scientific output, and only 24 firms (7.6%) produced any highly cited papers ([≥]200 citations). Scientific influence was highly concentrated: the top 10% of firms accounted for 96.8% of citations, while three startups accounted for 92 of 134 firm-attributed highly cited papers. Firm valuation was not associated with publication productivity or highly cited output, whereas funding raised showed weak associations. Overall, participation in formal scientific communication among AI unicorn startups is negligible, comprising only 0.1% of the overall AI literature in 2025. Most leading developers of AI technology do not engage with the scientific literature, raising concerns for the transparency, reproducibility, and accountability of this rapidly moving innovation frontier.

Matching journals

The top 5 journals account for 50% of the predicted probability mass.

1
PLOS ONE
● 5266 papers in training set
Top 7%
22.0%
Similar papers in this journal
2
eLife
● 5828 papers in training set
Top 3%
18.6%
Similar papers in this journal
3
Nature Human Behaviour
○ 95 papers in training set
Top 0.3%
4.9%
Similar papers in this journal
4
FACETS
● 14 papers in training set
Top 0.1%
4.3%
Similar papers in this journal
5
Scientific Reports
○ 3612 papers in training set
Top 25%
4.1%
Similar papers in this journal
50% of probability mass above
6
Proceedings of the Royal Society B: Biological Sciences
● 393 papers in training set
Top 2%
3.5%
7
Journal of Cognitive Neuroscience
● 135 papers in training set
Top 0.7%
3.3%
8
Conservation Biology
○ 17 papers in training set
Top 0.1%
3.2%
9
PeerJ
◐ 308 papers in training set
Top 3%
2.7%
10
PLOS Biology
● 486 papers in training set
Top 2%
2.5%
11
mBio
● 833 papers in training set
Top 7%
1.7%
12
Nature Neuroscience
○ 252 papers in training set
Top 3%
1.7%
13
FASEB BioAdvances
○ 18 papers in training set
Top 0.1%
1.7%
14
Heliyon
○ 152 papers in training set
Top 4%
1.5%
15
eneuro
● 439 papers in training set
Top 5%
1.5%
16
Proceedings of the National Academy of Sciences
● 2444 papers in training set
Top 31%
1.4%
17
Wellcome Open Research
○ 67 papers in training set
Top 0.6%
1.3%
18
PLOS Digital Health
● 106 papers in training set
Top 4%
1.1%
19
F1000Research
○ 88 papers in training set
Top 3%
1.0%
20
iScience
○ 1154 papers in training set
Top 30%
1.0%
21
Frontiers in Integrative Neuroscience
○ 15 papers in training set
Top 0.1%
0.8%
22
BMJ Open
● 601 papers in training set
Top 12%
0.8%
23
npj Digital Medicine
○ 118 papers in training set
Top 4%
0.6%
24
PLOS Computational Biology
● 1863 papers in training set
Top 21%
0.6%
25
Philosophical Transactions of the Royal Society B
● 51 papers in training set
Top 1%
0.6%
26
BMC Public Health
○ 158 papers in training set
Top 6%
0.6%

"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.