Firms gain when outside scientists build on their papers: papers that spur follow-on research lead firms to raise R&D spending, generate stronger patent outcomes, and hold on to staff—especially for firms with complementary assets or working in nascent fields.

Main Finding

Firms’ public scientific publications generate measurable private returns when external academics build on them. Using U.S. public firms’ publications (1990–2012) and a novel instrument exploiting quasi-random journal-issue assignment, the paper shows that greater accumulation of external follow-on research (multi‑generation citations by unaffiliated scholars) causes: increased subsequent firm investment in science, improved patenting outcomes (both directly and indirectly via follow-ons), higher retention of researcher employees, and more hiring/collaboration with academics. Effects are larger for firms with complementary assets, internal research capabilities, IP rights, and in nascent research fields.

Key Points

• Motivation: Publication creates imitation risks and may facilitate employee exit, but it also mobilizes the broader scientific community. The paper highlights this latter managerial benefit.
• Follow-on research is defined as scientific work outside the firm that cites and builds on a firm’s published findings (measured across citation generations).
• Descriptive evidence: only 7% of firm publications are directly cited by the firm’s own patents, but observing external follow-ons raises the share of firm papers that are effectively reabsorbed by later firm patents by ~33 percentage points.
• Mechanisms:
  • Input channel: academics develop downstream applications or complementary knowledge that firms can recombine into inventions.
  • Validation/signal channel: external attention functions as quality validation, reducing uncertainty about research trajectories and justifying retention and further internal investment.
• Moderators: Benefits concentrate where firms can capture value (IP ownership, complementary assets), where internal capabilities exist to absorb academic outputs, and in emerging fields or domains with accessible public funding.
• Example: IBM’s Bednorz & Müller (1986) paper—massive external follow-on activity produced many later IBM patents that cited those follow-ons; large implied private value beyond patents that directly cited the original paper.

Data & Methods

• Data:
  • DISCERN database of scientific publications and patents for U.S. publicly listed firms (1990–2012).
  • Matched to Microsoft Academic Graph (MAG), Dimensions.ai, American Men and Women of Science (AMWS), and other complementary sources.
  • Outcomes: firm-level subsequent scientific publications, patenting outcomes (counts and textual links), researcher retention, hires and collaborations with academics.
  • Follow-on research measured via multi-generation citation counts (academics’ citations to firm papers, then downstream citations).
• Identification / Empirical Strategy:
  • Endogeneity challenge: high-quality/interesting firm findings both attract follow-on research and elicit greater internal investment—bi-directional causality.
  • Instrument: quasi-random assignment of papers to journal issues. Some issues draw more attention than others because they include prominent external authors.
  • Constructed instrument: within journal-year, sum of the H-indexes of the top two prominent (non-focal) authors in the issue (captures exogenous attention to other issue contents). The argument: issues with more prominent co-authors attract more readers and citations for all papers in that issue.
  • Estimation: two-stage least squares (2SLS) with standard controls (paper, firm, and time covariates implied), using the instrument to isolate exogenous variation in follow-on research and then estimating its effect on firm-level outcomes.
• Supporting analyses:
  • Textual similarity analyses show citing academics’ work is more topically similar to the firm’s focal papers than their own prior work—consistent with influence (directional effect).
  • Examination of citation generations and patent citations to distinguish direct vs. indirect reabsorption channels.

Implications for AI Economics

• For AI firms and platforms:
  • Publishing upstream research (papers, preprints, benchmarks, model cards) can be a strategic choice: it may mobilize external research that later becomes a productive input or validates internal directions, increasing R&D returns and employee retention.
  • The value of open disclosure in AI depends on appropriation mechanisms and complementary assets (compute, datasets, deployment channels). Firms with strong capture mechanisms are more likely to benefit from inducing external follow-ons.
  • In fast-moving, nascent subfields of AI (e.g., foundation models, alignment), external follow-on activity may be particularly valuable because it accelerates downstream application development and signals promising directions.
• For researchers and managers:
  • Managerial evaluation and promotion criteria that incorporate external validation (citations, follow-on developments) can improve allocation of R&D resources under uncertainty.
  • Publishing policies should be considered jointly with hiring/career incentives: publication can increase researcher visibility but—contrary to some concerns—may raise retention if external follow-ons increase researchers’ productivity/value within the firm.
• For policy and ecosystem design:
  • A responsive public scientific ecosystem (active academic follow-on work) is a productive public good that raises private returns to firms’ upstream disclosure—strengthening arguments for public support of basic research and open dissemination channels.
  • Anticipated policy implication: supporting broad, accessible dissemination (funding, open access, conference support) increases downstream spillovers that can benefit both public and private innovation.
• Caveats for AI context / external validity:
  • Norms and channels in AI differ from the paper’s historical setting (1990–2012). AI research now heavily uses arXiv, conference proceedings, code/model releases, and online consumption—journal-issue placement instruments are generally not applicable to preprints/conference-dominated dissemination.
  • The rapid pace and commercial sensitivity of some AI research (large model capabilities, datasets) may change the tradeoffs between disclosure and imitation compared with traditional science; code/model/data releases and compute constraints matter as additional strategic levers.
  • Measuring follow-on work in AI should include artifacts beyond papers: code forks, model fine-tunes, datasets, benchmark usage, and GitHub activity, as these are important channels for external contribution and reabsorption.
• Directions for AI-focused research:
  • Replicate the paper’s logic using AI-era dissemination channels: exploit exogenous variation in conference session assignment, workshop scheduling, or visibility shocks (e.g., spotlight slots, keynote timing) as instruments for external attention.
  • Study the relative value of different disclosure modalities (paper vs. code vs. pretrained weights vs. datasets) for mobilizing external follow-ons and for firm outcomes (patents, products, talent retention).
  • Analyze heterogeneity across AI firms: startups vs. incumbents, those with deployment channels vs. research labs, and the role of compute/dataset exclusivity in moderating benefits from external follow-on research.

Summary takeaway: Beyond imitation risks, corporate scientific publication can pay off by mobilizing external follow-on research that firms can absorb or use as a quality signal. For AI firms, this suggests publication (and other forms of disclosure) can be a deliberate strategy to shape the external research environment—though the practical costs/benefits hinge on capture mechanisms, disclosure modality, and the peculiarities of the AI dissemination ecosystem.