Why it’s here
Filed beside LIT-482, and the pairing is the point. That paper argues the effects worth measuring live at the platform and collective scale rather than the individual one. This one finds that industry-tied research is sparse in exactly that cluster and over-represented in the cluster about what individual users share.
So the two halves are a methodological argument about where to look, and a measurement showing the field’s incentives point elsewhere.
For this record it is also a worked instance of DP-005 — adoption is not
evidence — in an unusual direction: the corpus studied is extraordinarily
well-adopted, at 180 citations per paper and 745 policy documents, and
that is the reason to look at it rather than a reason to trust it.
What was measured
295 articles from Science, Nature and PNAS plus their transfer journals, found via OpenAlex, written by 1210 authors. For each author, any year of funding from, collaboration with, or employment by Meta, X, Google or Microsoft — every funding and employment tie manually validated against sources independent of OpenAlex, and counted against each journal’s own competing-interest window (three years for Nature, four for PNAS, five for Science).
- 49% of papers have a disclosable tie, and the majority of those ties are not disclosed in the published work.
- Only 21% of authors have one. So this is not the field engaging broadly with industry; it is a small group with durable relationships.
- By firm: Meta 14%, Google 8%, Microsoft 6%, X 1%.
- Editors and reviewers carry undisclosed ties too, found by the same detection plus public CVs.
- Industry-tied work draws more attention — in citations, in policy documents, in news and on social media.
- Topical skew. Bibliographic coupling gives five communities: misinformation sharing, platform dynamics, mental health, social network analysis, political behavior. Three are at the corpus average. Industry ties are over-abundant in misinformation sharing and sparse in platform dynamics — the industry-linked cluster concentrating on experiments about people sharing false headlines, the independent one on network structure and dynamics.
The corpus’s reach, which is why any of this matters: 53,120 citations, 15,708 news articles, 100,863 social media mentions, 418 Wikipedia mentions, 745 policy documents.
Conditions
The topical finding is correlational and the authors say so — the paper states it cannot definitively show that industry funding causes the skew. Researchers interested in user-level questions may attract industry funding rather than industry funding directing them there, and the design cannot separate those.
“Disclosable tie” is the paper’s own construct, defined to be detectable and to match journal policy windows rather than to capture everything a reader might count as a competing interest. It is deliberately narrower than “conflict of interest” and the paper is explicit about the distinction.
Four firms, seven journals, one field. Meta, X, Google and Microsoft; Science, Nature, PNAS and their transfer venues. A tie to a firm outside that set, or a paper outside those journals, is invisible here.
Detection is asymmetric. Ties found are ties that left a public trace — OpenAlex records, announced RFPs, fellowships, CVs. Undetected ties bias the 49% downward, which is the conservative direction, and it means the figure is a floor rather than an estimate.
Same first author as LIT-482. The two papers are filed as a pair and argue complementary halves of one position. They are not independent sources and the record should not count them as two groups agreeing.