designers increase emotional content by 9.81% and com-plexity by 17.1% after the AI launch.9Additionally, we find that the unsuccessful designersin all three categories do not change emotions in theirdesigns (Table 11) and do not change complexity of theirdesigns (Table 12) after the AI launch.The granular model with five emotions also showsthat the unsuccessful designers in all three categories donot change emotional content of their design submis-sions (see Online Appendix F, Table F12).To confirm all the results, we use DiD models withpropensity score matching (PSM-DID1 and PSM-DID2Table 4. Number of Contests Before and After AI for the Three Groups of Successful DesignersDependent variableis number of contests(Ycdt) Lower-tier Cross-tierCross-categoryLower-tier(PSM –DID1)Cross-tier(PSM –DID1)Cross-category(PSM –DID1)Lower-tier(PSM –DID2)Cross-tier(PSM –DID2)Cross-category(PSM –DID2)Aftert (or Aftert ×Treatmentdm for DID)�0.21*** �0.041ns 0.22ns �0.0858ns 0.4ns 0.142ns 0.043ns 0.063*** �0.17***(0.03) (0.043) (0.401) (0.203) (0.258) (0.186) (0.051) (0.0182) (0.012)Constant 2.86*** 3.05*** 2.649*** 2.4*** 2.3*** 2.14*** 3.112*** 3.21*** 3.15***(0.22) (0.305) (0.266) (0.36) (0.047) (0.15) (0.0075) (0.0065) (0.007)Designer fixed effects Yes Yes Yes No No No No No NoNo. of designers 119 103 63 — — — — — —Sample size 18,450 20,730 15,915 27,920 26,852 22,540 15,036 14,468 14,602Note. Subscript t denotes time, subscript d denotes designers, subscript c denotes contests, subscript dm denotes matched designers in treatmentand control groups, and subscript cdt denotes number of contests per designer per day.***p < 0.01; **p < 0.05; *p < 0.1; ns, not significant.Table 5. Number of Contests Before and After AI for the Three Groups of Unsuccessful DesignersDependentvariable isnumber ofcontests (Ycdt) Lower-tier Cross-tierCross-categoryLower-tier(PSM –DID1)Cross-tier(PSM –DID1)Cross-category(PSM –DID1)Lower-tier(PSM –DID2)Cross-tier(PSM –DID2)Cross-category(PSM –DID2)Aftert (or Aftert ×Treatmentdm forDID)0.55*** 0.524*** 0.56*** 0.329** 0.222* 0.4* 0.48*** 0.196*** 0.7445***(0.016) (0.0213) (0.06) (0.167) (0.121) (0.22) (0.025) (0.0272) (0.0292)Constant 3.99*** 3.37*** 3.71*** 2.99*** 2.32*** 3.27*** 3.495 3.62*** 3.63***(0.32) (0.3) (0.399) (0.02) (0.083) (0.065) (0.007) (0.0073) (0.0075)Designer fixedeffectsYes Yes Yes No No No No No NoNo. of designers 477 450 263 — — — — — —Sample size 44,618 29,365 27,158 61,185 58,498 34,937 31,888 18,405 18,033Note. Subscript t denotes time, subscript d denotes designers, subscript c denotes contests, subscript dm denotes matched designers in treatmentand control groups, and subscript cdt denotes number of contests per designer per day.***p < 0.01; **p < 0.05; *p < 0.1; ns, not significant.Lysyakov and Viswanathan: User Responses to the Threat of AIInformation Systems Research, 2023, vol. 34, no. 3, pp. 1191–1210, © 2022 INFORMS 1203Downloaded from informs.org by [47.197.133.180] on 28 October 2025, at 18:19 . For personal use only, all rights reserved.

Prior research on AI limitations indicates that,although modern AI systems with advanced deep learn-ing capabilities are very impressive, humans are stillmore advanced in such qualities as creativity, imagina-tion, and emotions in general (Braga and Logan 2017)and creativity and emotional and social intentions indesign and art specifically (Hertzmann 2018, Mazzoneand Elgammal 2019). Recent advances in generativeadversarial networks (i.e., so-called creative adversarialnetworks) suggest that algorithms can be trained to usethe same distribution of styles used by human artists butat the same time to maximize the differences between anew algorithmically generated art and all prior works,thus making the AI-generated art as novel as possible

The findings of this study also have important practi-cal implications. Platform providers can use our findingsto better evaluate the impacts of AI systems on contestsand designers’ behaviors. Understanding how differentgroups of users respond to the launch of an AI systemcan help market providers to design relevant pricingand marketing strategies to optimize performance andrevenue from both sources: AI and human designers. Amore nuanced understanding of the capabilities andlimitations of AI systems relative to those of experthuman designers can help platform providers recom-mend specific guidelines for contest holders and contestparticipants to improve outcomes. This could also pavethe way for hybrid solutions that leverage the capabil-ities of both the AI system and human experts.

Perhaps, the most interesting set of results pertainto the differences in how successful designers respondto the threat from AI compared with unsuccessful de-signers. In examining how the behaviors of successfuldesigners and the others differ in response to the AI sys-tem launch, we find that in contrast to the unsuccessfulcontestants who increase the number of contests theyparticipate in (by 13%–15%), the successful contestantssubstantially increase the number of submissions (by30%–60%) within a contest compared with the periodbefore the AI system launch. Furthermore, we find thatwith an increase in the number of submissions by suc-cessful designers, there is a concomitant significantincrease in the emotional content and the complexity oftheir designs after the AI system launch. On the other

AI system. To understandwhether and how designers change their design sub-missions in response to the threat from AI, we measurethe designs’ emotional content and complexity becausethose variables have been shown (by prior research inpsychology and marketing) to affect esthetic perceptionof art and design images as described later. We thenexamine whether and how the emotional content andcomplexity of design submissions affect the likelihoodof winning a contest and how these differ for successfuland unsuccessful designers before and after the intro-duction of the AI system.

How do designers respond to the introduction of theAI system for logo design tasks?How do the behaviors of successful designers differfrom the behaviors of other designers in response to theintroduction of the AI system?This study builds on the theory of threat. Specifically, weuse a theoretical lens of the protection motivation theory(PMT; Rogers 1975, 1983) to understand possible responsesof designers to the threat of AI. The protection motivationand protection behaviors depend on whether individualsfeel the threat and whether they have coping abilities todeal with the threat. According to the PMT, individualswho do not perceive a threat will not respond to it. Amongthe others, individuals will take steps to avoid the threat orexhibit adaptive behaviors based on their ability to dealwith the threat.