Projects that disclose AI use raise less money on Kickstarter, but the penalty depends on what AI made: admitting AI-created images carries a smaller funding hit than admitting AI-created text, a gap that widens when AI produces the final product and narrows when AI is used for marketing.

The Transparency Penalty Unraveled: How AIGC Modality Moderates Consumer Responses to AI Disclosure in Crowdfunding

Wenhao Bai, Zhong Yao, Wuhuan Xu · July 05, 2026 · Journal of the Association for Information Systems

openalex correlational medium evidence 7/10 relevance Summary only summary available; pdf_status=paywall Source PDF

Using LLM-assisted classification and entropy balancing on 41,073 Kickstarter projects, the study finds that disclosing AI use is associated with lower funding performance overall, with smaller penalties for AI-generated visual content than for AI-generated text and heterogeneity by whether AI was used for final products versus marketing materials.

Governments and platforms worldwide increasingly mandate disclosure of AI-generated content (AIGC), yet how such transparency shapes market outcomes remains underexplored. Situated in reward-based crowdfunding, this study examines the association between AI-use disclosure and funding performance, and whether this association differs across disclosed AIGC modalities (visual vs. textual) and application stages (final product vs. marketing material). Grounded in signaling theory and legitimacy theory, we analyze 41,073 Kickstarter projects using an LLM-assisted text classification approach and entropy balancing. The results indicate that AI-use disclosure is associated with a significant decline in funding performance. However, this association exhibits systematic heterogeneity: disclosing AI involvement in visual content creation is associated with a weaker penalty than disclosing AI involvement in textual content creation, and this modality gap widens for final products yet narrows for marketing materials. These findings illuminate the modality-level heterogeneity and application-stage boundary conditions of consumer responses to AI-use disclosure in crowdfunding.

Summary

Main Finding

AI-use disclosure on Kickstarter is associated with a statistically significant decline in funding performance. The penalty is heterogeneous by modality and application stage: disclosing AI involvement in visual content creation incurs a smaller funding penalty than disclosing AI involvement in textual content, and this modality gap is larger when AI is used in final products but smaller when AI is used in marketing materials.

Key Points

Sample: 41,073 Kickstarter (reward-based crowdfunding) projects.
Overall effect: Projects that disclose AI use raise less funding / perform worse than projects that do not disclose AI use.
Modality heterogeneity:
- Visual AIGC disclosure → weaker negative effect on funding.
- Textual AIGC disclosure → stronger negative effect on funding.
Application-stage heterogeneity:
- For final products, the difference between visual vs textual disclosure penalties widens (textual disclosure especially costly).
- For marketing materials, the modality gap narrows (visual vs textual disclosures produce more similar effects).
Theoretical framing: findings interpreted through signaling theory (disclosure as a signal of unobserved product quality or capabilities) and legitimacy theory (disclosure affecting perceived social/market acceptability).

Data & Methods

Data: 41,073 Kickstarter project listings (text and metadata).
Identification of AI disclosure and classification by modality (visual vs textual) and stage (final product vs marketing material) achieved using an LLM-assisted text classification pipeline.
Estimation approach: entropy balancing to create covariate-balanced comparisons between disclosing and non-disclosing projects, controlling for observable project and creator characteristics (e.g., category, funding goal, prior creator success, timing).
Robustness: heterogeneity analyses by modality and application stage; results reported as associations (observational design with balancing to reduce confounding).

Implications for AI Economics

Market signaling and adoption incentives:
- Disclosure can reduce demand — creators face a trade-off between regulatory/compliance transparency and reduced funding prospects.
- Differential penalties by modality create uneven incentives for AI adoption across sectors (text-intensive products may see slower adoption than visual-intensive ones).
Platform and regulatory design:
- Mandates to disclose AI use may have unintended distributional effects across creators and product types; policy design should consider modality- and stage-specific impacts.
- Platforms might need to accompany disclosure rules with consumer education or credibility-enhancing mechanisms (third-party verification, authenticity badges) to mitigate penalties.
Labor and product-market effects:
- Greater consumer sensitivity to textual AIGC could protect or reward human-generated textual labor relative to visual content where AI is less penalized.
- Innovation strategies may shift toward using AI more in visual/marketing roles and less in core textual content if disclosure persists.
Measurement and research methods:
- LLM-assisted text classification combined with entropy balancing is a viable approach for large-scale observational studies of AIGC disclosure and market outcomes.
Welfare considerations:
- While transparency supports informed choice, disclosure-induced demand reductions can lower creator surplus and potentially slow diffusion of productive AI uses; weighing transparency benefits against economic costs requires further study.

Assessment

Paper Typecorrelational Evidence Strengthmedium — Large sample and use of an LLM classifier plus entropy balancing strengthen internal consistency and reduce measured confounding, but the study is still observational with nonrandom disclosure, potential unobserved confounders, possible measurement error in classifying disclosures and modalities, and therefore cannot unequivocally establish causality. Methods Rigormedium — The analysis leverages modern tools (LLM-assisted text classification) and a principled weighting approach (entropy balancing) appropriate for observational data, indicating careful empirical work; however, reliance on self-disclosed text, possible classifier misclassification, limited information about robustness to alternative specifications, and the inherent limits of balancing (vs. identification strategies like instruments or experiments) constrain methodological rigor. Sample41,073 reward-based crowdfunding projects on Kickstarter (project text used for classification of AI-use, modality: visual vs textual, and application stage: final product vs marketing material); funding outcomes (e.g., amount raised / success) are analyzed; timeframe and geographic coverage not specified in the summary. Themesadoption human_ai_collab IdentificationObservational association estimated on 41,073 Kickstarter projects where AI-use disclosure (and modality/stage) is identified via an LLM-assisted text classification pipeline; covariate balance is enforced using entropy balancing to adjust for observed confounders and estimate weighted associations between disclosure and funding outcomes (no randomized assignment). GeneralizabilityLimited to reward-based crowdfunding on Kickstarter and may not extend to other platforms (e.g., Indiegogo) or offline markets, Findings apply only to projects where AI use is disclosed in project text and may not generalize to undisclosed AI use or to contexts with different disclosure norms, Cultural and temporal variation in attitudes toward AI may limit transferability across countries and over time as perceptions evolve, Results concern consumer/backs patron responses and may not generalize to employer hiring, B2B sales, or other economic outcomes, Potential classifier errors and platform-specific dynamics (category composition, platform reputation) constrain external validity

Claims (5)

Claim	Direction	Outcome	Confidence & Evidence	Details
AI-use disclosure is associated with a significant decline in funding performance for Kickstarter projects. Firm Revenue	negative	funding performance	Reading fidelity high Study strength medium	n=41073 0.3
Disclosing AI involvement in visual content creation is associated with a weaker funding penalty than disclosing AI involvement in textual content creation. Firm Revenue	negative	funding performance	Reading fidelity high Study strength medium	n=41073 0.3
The modality gap (weaker penalty for visual vs. textual AI-use disclosure) widens when AI is used in final products but narrows when AI is used in marketing materials. Firm Revenue	mixed	funding performance	Reading fidelity high Study strength medium	n=41073 0.3
This study analyzes Kickstarter projects using an LLM-assisted text classification approach combined with entropy balancing. Other	null_result	methodological approach (LLM-assisted classification and entropy balancing)	Reading fidelity high Study strength high	n=41073 0.5
The study is grounded theoretically in signaling theory and legitimacy theory to interpret consumer responses to AI-use disclosure. Governance And Regulation	null_result	theoretical framing	Reading fidelity high Study strength speculative	not reported 0.05