Ownership in AI-Assisted Everyday Tasks
Megan Wei1
Melanie Subbiah1
arXiv:2609.20658v1 [cs.AI] 17 Sep 2026
Dave Edwards4
1
Audrey Lee2
Helen Edwards4
Annya Dahmani3 Ellie Pavlick1
Brown University {meganwei, m.subbiah, ellie_pavlick}@brown.edu 2 Carnegie Mellon University [email protected] 3 University of California, Berkeley [email protected] 4 Artificiality Institute {dave, helen}@artificialityinstitute.org
Abstract When does work done with AI still feel like ours? As AI becomes woven into everyday tasks, we must examine what happens to our sense of ownership and contribution when a machine shares in producing what we make. We report an exploratory qualitative survey in which participants were asked to describe two recent, self-selected tasks completed with AI: one that felt like their own and one that did not. We find that felt ownership depends on the process of collaboration: people disown work when they merely approve AI’s suggestions, but retain ownership when they lead, iterate, or rewrite. Ownership can also extend to settings where people own the vision for a project but not the execution; respondents reported high ownership on tasks they could not have completed without AI. Loss of personal voice and a lack of comprehension of the output both erode ownership. Finally, willingness to disclose AI use is often decoupled from actual pride or ownership, and instead shaped by community norms and fear of credit erasure. We propose several research directions as a result of these findings to promote AI development that supports people’s sense of authorship over their own lives.
1
Introduction
Artificial Intelligence (AI) is rapidly becoming integrated into both our personal and work lives, primarily through Large Language Models (LLMs) used as chatbots or agents [Hosseini and Seilani, 2025]. It is, therefore, imperative to consider the effect this integration will have not just on our work, but on our personal well-being [Sharma et al., 2026, Yang et al., 2026, Luettgau et al., 2025]. The relevant research and media attention have largely focused on the most high-stakes areas of our well-being, involving self-harm, death, and crises of mental health [Guo et al., 2024, Chung et al., 2026, Archiwaranguprok et al., 2025]. However, other important aspects of our well-being that may erode more slowly over time are no less consequential. One such area is our sense of purpose and fulfillment in our life and work, which hinges on some degree of pride or ownership in the decisions and actions we take [McKnight and Kashdan, 2009, Kirk et al., 2015, Polimetla et al., 2026]. In this paper, informed by a qualitative survey, we narrow in on this issue, asking, How does using AI in everyday tasks affect our sense of ownership of our daily life decisions and work? Prior work has defined cognitive offloading [Risko and Gilbert, 2016, Gerlich, 2025, Grinschgl et al., 2021, Guingrich et al., 2026, Yang et al., 2026] as delegating cognitive effort to another entity, such as an AI tool, and thus reducing the user’s effort and involvement in the process and output. Cognitive offloading affects comprehension and learning for the person doing the task, often with minimal gains Preprint.
in efficiency [Kosmyna et al., 2025, Yu et al., 2026]. In our work, we focus on the feelings attached to this phenomenon, developing hypotheses for factors that contribute to people feeling ownership of everyday tasks they complete with AI assistance. We conduct an informal survey1 comparing people’s patterns and feelings when using AI in a task over which they ultimately feel high ownership versus a task over which they feel low ownership. Comparing points along this spectrum is in line with Polimetla et al. [2026]’s work on felt ownership in creative work and allows us to probe differences in the tasks themselves, in interaction patterns with the AI, and in personal reactions to the task and process. Our qualitative analysis of these responses reveals several themes driving ownership: 1) the importance of process over output, 2) a feeling of differentiating one’s personal voice, 3) achieving something beyond one’s own abilities, and 4) comprehension of the output. Additionally, we uncover that people’s willingness to disclose AI use to colleagues or peers is less connected to their actual sense of ownership or pride in their work and more contingent on the norms and stigmas in their communities. Definitely not
Probably not
Might or might not
Probably yes
Definitely yes
Would you be able to do this task without AI?
“Wanted to build this sort of thing for over a decade, now I can code with my AI assistant(s)… It's all still mine, AI is just a translator.” “It is me — minus friction. My language and ideas.”
High ownership
“Personalization and verification (done right) take time and keep my brain connected enough to make me feel like the final product is me.”
Low ownership “It sounded nothing like me… AI emails sound very vanilla.” “In general, topics I process using AI and I know little about do not feel like my own work.”
“I provided the intent and information… I feel more like I approved the output than authored it.”
Figure 1: Responses when asked whether users are able to do the task without AI, across high and low ownership tasks.
2
Survey Design and Methods
We design a survey (all questions shown in Appendix A) with 1) demographics, 2) general AI usage, 3) two task reflections, 4) a checklist of common AI tasks, and 5) comparative questions. For the task reflections [Polimetla et al., 2026], participants described one high-ownership and one low-ownership task, shown in randomized order. They answered whether the task would have been done, and could have been done, without AI, alongside free-text reflections on identity, independent contribution, pride, and disclosure. Participants selected tasks done using AI in the past 30 days from a list adapted from Chatterji et al. [2025]. The comparative questions (Why these tasks? What made the biggest difference in felt ownership?) assessed differences across tasks [Zindulka et al., 2026]. We recruited respondents via professional networks and social media (LinkedIn, Instagram), as well as university mailing lists and Slack channels, collecting responses from August 6-27, 2026. The sample is biased (Appendix C), especially toward mid-to-late career highly educated white adults who heavily use AI and are able to pay out of pocket for it. In total, 117 people began the survey, with 52 complete responses and 64 responses containing at least one task reflection.
3
Results
Process drives ownership. When asked what made the difference in ownership between their two tasks, respondents highlighted that ownership was about who led the process (“ownership comes from owning the decisions made during the processing of the information. If there was no processing of information nor synthesis then it doesn’t feel I own it.”). This finding is in line with Polimetla et al. [2026]’s work on creative ownership which showed process, and particularly control over decisions, to be a differentiating factor in ownership. We also see ownership diminishes when the human simply 1 Participants were recruited through our professional networks, spanning students at research universities and adults in the design and entrepreneurship community, who were generally heavy users of AI tools. We acknowledge this is a biased sample; we report these results as a necessary first step toward surfacing problems and insights in this space, which we encourage future work to validate through more structured research studies. View survey results here: aiownership.github.io
2
approves the AI output (“I reviewed, I looked, I was in the loop, but I wasn’t in charge.”) [Dhillon et al., 2024]. Some users are intentional about protecting their process to preserve ownership (“I augment my own expertise and skills with AI, I don’t hand off the whole thing.”). In collaborative settings, ownership on the shared product shifts based on observations of collaborators’ AI usage that may be misaligned with an individual’s process (“Early on, I was more proud of the project. . . As the deadline approached, I saw my students use AI more and more extensively for filling in gaps.”). Homogeneous voice, personalization, and rejection. For writing tasks, a prominent source of disownership was the voice [Ippolito et al., 2022]: “it sounded nothing like me. . . AI emails sound very vanilla.”; “I didn’t like how ‘artificial’ it sounded.”; and “the sentence structure and big words used did not sound like any human personality” [Chakrabarty et al., 2025], which aligns with findings of homogenization and reduced diversity in AI-generated content [Doshi and Hauser, 2024, Si et al., 2025, Anderson et al., 2024, Sourati et al., 2026]. Respondents who invested in personalization [Wan and Kalman, 2026, Hwang et al., 2025, Shaikh et al., 2025] of models reported retaining ownership (“consistent with my voice after much training of this project”; “My model already has md file for content creation. This shows samples of my work, direct instructions about my tone, guardrails of what not to say or do”). However, there is a ceiling to this personalization [Wang et al., 2025], where the outputs were “over aggregated and without some of my unique POV”. Interestingly, AI’s bad outputs sometimes helped people build ownership by giving them a point of contrast: “It helped me understand what I didn’t want so I could create what I wanted. . . I definitely felt more ownership. . . because AI failed so horribly.”; “I don’t always know what I want. . . until I see where the AI-generated version misses. . . even an imperfect output can be useful because the process of disagreeing with it. . . becomes part of my own thinking.” This phenomenon has also been evidenced in Guo et al. [2025], where writers perceived bad outputs as useful “anti-patterns”: seeing what they don’t want helps them understand what they do want. Encountering and rejecting bad AI outputs can help people define and own their work. Enabling outputs beyond one’s own abilities. Counterintuitively, respondents were less certain they could have done their high-ownership task without AI, where 42% answered “definitely yes” against 63% for low-ownership tasks (Figure 1). Some of the most-owned projects would have been impossible for their creators and unlocked new creative powers. For example, two respondents had longed to produce children’s books and graphic novels but could not illustrate their writing, with one commenting, “my inability to get [my vision] out of my head originally led me to stop writing children’s books.” AI helped them realize their envisioned illustrations with great positive effect: “I feel like I’m truly honoring my younger self” [Fu et al., 2025, Newman et al., 2024, Louie et al., 2020]. When the human directs the project’s vision, AI occupies the role of a hired illustrator or junior teammate (“it felt like directing a human paralegal. That’s my job.”) [Khadpe et al., 2020], which does not threaten ownership [Maier et al., 2026]. People can feel high ownership over tasks they do not possess the skills to complete without AI. Meanwhile, disowned tasks were more often things people could do, but chose to delegate: routine emails, lookups, and formatting. Owned and disowned tasks differ in nature: “a task that I wanted to have ownership of” compared to “a task that I would have rather not have had to do in the first place”, which aligns with Pierce et al. [2001]’s defined motives towards psychological ownership. No ownership without comprehension. One route towards psychological ownership [Pierce et al., 2001] is intimately knowing the task. In our study, ownership failed when respondents could not understand the artifact (“I really don’t understand much of the codebase. . . I wouldn’t consider this my work at all. . . honestly I’ll probably be annoyed if I have to patch it in the future since I don’t even feel responsible for it.”) [Vaithilingam et al., 2022, Prather et al., 2024] or evaluate the output (“I certainly don’t know enough about the topic to judge and evaluate the output — hence does not feel like something I would call ‘mine”’). Without comprehension, people felt little responsibility for maintaining and defending their work [Seo et al., 2026]. Disclosure. Respondents’ willingness to disclose AI use varied widely and often depended on the situation. Individuals most willing to disclose AI use had concerns about correctly attributing the source of the work (“I’d absolutely mention that almost all of it was AI generated, otherwise I’d be taking way more credit than I deserve”) [Fang et al., 2026, Walters and Wilder, 2023, Topaz et al., 3
2026]. People cautiously open to disclosure generally preferred to narrate the process of AI usage rather than a label of AI-generated or not [He et al., 2025]: “I’d be willing to say I used AI for it if I clarified the limited extent to which it was used.” Often, they fear credit erasure [Reif et al., 2025] under a simple label, worrying people would think “that I didn’t do any work. . . as if I don’t think for myself” or that “the code itself was mostly or entirely AI generated, thus discounting my contribution.” This concern has been reflected in previous work, with AI disclosure eroding trust [Schilke and Reimann, 2025]; damaging perceptions of the creator [Rae, 2024]; and impacting marginalized groups, future evaluations, and hiring outcomes [Kadoma et al., 2025]. These concerns drove reluctance for respondents least willing to disclose AI use (“I don’t tend to tell people I use AI for my emails when I do use it”; “I would probably be ashamed to admit using the tools”). Respondents also reported that disclosure depends “somewhat on the audience and the stakes”, as prior work found with AI-written dating profiles [Barkallah and Zytko, 2026]. The surrounding organizations often influence these audiences: from a company that “encourages its use,” to situations where people would disclose to colleagues but not clients, to communities that might assume “delegation rather than oversight.” One lecturer reported that they would not share a failed AI experiment with colleagues who “are in love with AI and won’t hear anything bad against it.” As a result of these complex social pressures, willingness to disclose AI use is not necessarily coupled with feelings of pride or ownership of the work: “AI helped me clarify my thinking. . . [but] I think I would have to be careful disclosing that I used AI. Not that I’m not proud, but there is a stigma.” Disclosure is a distinct dimension: willingness to disclose AI use does not necessarily accompany pride or ownership of the work.
4
Takeaways and Recommendations
From our survey, ownership emerges as a function of the process: iterating with the model, authoring the prompts, personalizing models, and rejecting and learning from bad outputs. AI has made previously unachievable tasks achievable, and it has created situations where people produce output they can’t fully evaluate or explain. As AI becomes more deeply integrated into society [Kobiella et al., 2024], we need to investigate how to preserve people’s creative agency [Kumar et al., 2025, Guo et al., 2025], ownership [Polimetla et al., 2026], and learning across various collaborative, evaluative, and social settings. These findings motivate further research into the themes identified in our results: Focus on process. We encourage study of not just how we can increase human investment in producing AI-assisted work, but also how we can increase process visibility for users and collaborators. For example, does recording and displaying the interaction history (edits, rejections, prompts, decisions, leadership) strengthen felt ownership and calibrate self-attribution? How does visibility of a collaborator’s process reshape ownership in shared projects? Can we measure adverse effects when people do not take ownership of the process, such as increased effort for reviewers of the work? Personalization. We encourage future work on personalization to not be purely driven by commercial incentives but also to promote user well-being and identity, such as by investigating the effects of perceived versus actual personalization. Several respondents felt they had uniquely scaffolded their AI to adapt to them, making it now their AI. Are these personalization methods actually unique and effective? How does informing people that their scaffolding methods are actually similar change their perception of ownership? Can we measure how differentiated a model is in its responses, and does this influence the user’s ownership or identification with the model? Enabling the previously impossible. Our exploration shows that people can feel great ownership over outputs that they could not have produced on their own, like illustrating a graphic novel. However, it is not clear how this sense of ownership coexists with low expertise in some areas of the output. Is this effect primarily present in cases where users may be strong evaluators but not strong producers of the work (e.g., the art critic vs. the artist)? Does revealing a skilled critique from an expert human collaborator change the felt ownership of an accomplishment outside of one’s domain? Promoting honest disclosure of AI use. People can more authentically own their work and learn healthy modes of human-AI collaboration when community cultures encourage honesty around discussing AI use. We encourage not just studying how readers or reviewers of work perceive AI disclosures, but also what factors encourage or discourage the producer from honestly disclosing AI use when it is required. For example, what formats of disclosure are most effective (e.g., categorical vs. narrative)? How does social feedback on output quality affect willingness to disclose use? 4
Acknowledgments and Disclosure of Funding This material is based upon work supported by the U.S. National Science Foundation (NSF) under Cooperative Agreement No. 2433429, "NSF Al Research Institute on Interaction for Al Assistants (ARIA)”, and by Google LLC.
References Barrett R. Anderson, Jash Hemant Shah, and Max Kreminski. Homogenization Effects of Large Language Models on Human Creative Ideation. In Proceedings of the 16th Conference on Creativity & Cognition, C&C ’24, pages 413–425, New York, NY, USA, June 2024. Association for Computing Machinery. ISBN 9798400704857. doi: 10.1145/3635636.3656204. URL https://doi.org/10.1145/3635636.3656204. Chayapatr Archiwaranguprok, Constanze Albrecht, Pattie Maes, Karrie Karahalios, and Pat Pataranutaporn. Simulating Psychological Risks in Human-AI Interactions: Real-Case Informed Modeling of AI-Induced Addiction, Anorexia, Depression, Homicide, Psychosis, and Suicide. arXiv preprint arXiv:2511.08880, November 2025. URL https://arxiv.org/abs/2511.08880. Meryem Barkallah and Douglas Zytko. “I Wanted Them to Think That I Wrote That”: AI-Generated Self-Presentation on Dating Apps and Implications of Non-Disclosure on Informed Consent. In Proceedings of the 2026 CHI Conference on Human Factors in Computing Systems, CHI ’26, New York, NY, USA, April 2026. Association for Computing Machinery. ISBN 9798400722783. doi: 10.1145/3772318.3791593. URL https://doi.org/10.1145/3772318.3791593. Tuhin Chakrabarty, Philippe Laban, and Chien-Sheng Wu. Can AI writing be salvaged? Mitigating Idiosyncrasies and Improving Human-AI Alignment in the Writing Process through Edits. In Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems, CHI ’25, New York, NY, USA, April 2025. Association for Computing Machinery. ISBN 9798400713941. doi: 10.1145/3706598.3713559. URL https://doi.org/10.1145/3706598.3713559. Aaron Chatterji, Thomas Cunningham, David J. Deming, Zoë Hitzig, Christopher Ong, Carl Yan Shan, and Kevin Wadman. How People Use ChatGPT. Working Paper 34255, National Bureau of Economic Research, September 2025. URL https://www.nber.org/papers/w34255. Van-Han-Alex Chung, Pénélope Bernier, and Alexandre Hudon. Mass Media Narratives of Psychiatric Adverse Events Associated With Generative AI Chatbots: Rapid Scoping Review. JMIR Mental Health, 13:e93040, March 2026. doi: 10.2196/93040. URL https://mental.jmir.org/2026/ 1/e93040. Paramveer S. Dhillon, Somayeh Molaei, Jiaqi Li, Maximilian Golub, Shaochun Zheng, and Lionel Peter Robert. Shaping Human-AI Collaboration: Varied Scaffolding Levels in Cowriting with Language Models. In Proceedings of the 2024 CHI Conference on Human Factors in Computing Systems, CHI ’24, New York, NY, USA, May 2024. Association for Computing Machinery. ISBN 9798400703300. doi: 10.1145/3613904.3642134. URL https: //doi.org/10.1145/3613904.3642134. Anil R. Doshi and Oliver P. Hauser. Generative AI enhances individual creativity but reduces the collective diversity of novel content. Science Advances, 10(28):eadn5290, July 2024. doi: 10.1126/sciadv.adn5290. URL https://www.science.org/doi/10.1126/sciadv.adn5290. Jingchao Fang, Victoria Xiaohan Wen, and Mina Lee. What Influences Readers’ and Writers’ Perceived Necessity of AI Disclosure? In Proceedings of the 2026 ACM Conference on Fairness, Accountability, and Transparency, FAccT ’26, pages 894–920, New York, NY, USA, June 2026. Association for Computing Machinery. ISBN 9798400725968. doi: 10.1145/3805689.3806724. URL https://doi.org/10.1145/3805689.3806724. Yue Fu, Michele Newman, Lewis Going, Qiuzi Feng, and Jin Ha Lee. Exploring the Collaborative Co-Creation Process with AI: A Case Study in Novice Music Production. In Proceedings of the 2025 ACM Designing Interactive Systems Conference, DIS ’25, pages 1298–1312, New York, NY, USA, July 2025. Association for Computing Machinery. ISBN 9798400714856. doi: 10.1145/3715336.3735829. URL https://doi.org/10.1145/3715336.3735829. 5
Michael Gerlich. AI Tools in Society: Impacts on Cognitive Offloading and the Future of Critical Thinking. Societies, 15(1):6, January 2025. doi: 10.3390/soc15010006. URL https://www. mdpi.com/2075-4698/15/1/6. Sandra Grinschgl, Frank Papenmeier, and Hauke S. Meyerhoff. Consequences of cognitive offloading: Boosting performance but diminishing memory. Quarterly Journal of Experimental Psychology, 74(9):1477–1496, September 2021. doi: 10.1177/17470218211008060. URL https://doi. org/10.1177/17470218211008060. Rose E. Guingrich, Dvija Mehta, and Umang Bhatt. Belief Offloading in Human-AI Interaction. arXiv preprint arXiv:2602.08754, February 2026. URL https://arxiv.org/abs/2602.08754. Alicia Guo, Shreya Sathyanarayanan, Leijie Wang, Jeffrey Heer, and Amy X. Zhang. From Pen to Prompt: How Creative Writers Integrate AI into their Writing Practice. In Proceedings of the 2025 Conference on Creativity and Cognition, C&C ’25, pages 527–545, New York, NY, USA, June 2025. Association for Computing Machinery. ISBN 9798400712890. doi: 10.1145/3698061.3726910. URL https://doi.org/10.1145/3698061.3726910. Zhijun Guo, Alvina Lai, Johan H. Thygesen, Joseph Farrington, Thomas Keen, and Kezhi Li. Large Language Models for Mental Health Applications: Systematic Review. JMIR Mental Health, 11:e57400, October 2024. doi: 10.2196/57400. URL https://mental.jmir.org/2024/1/ e57400. Jessica He, Stephanie Houde, and Justin D. Weisz. Which Contributions Deserve Credit? Perceptions of Attribution in Human-AI Co-Creation. In Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems, CHI ’25, New York, NY, USA, April 2025. Association for Computing Machinery. ISBN 9798400713941. doi: 10.1145/3706598.3713522. URL https: //doi.org/10.1145/3706598.3713522. Soodeh Hosseini and Hossein Seilani. The role of agentic AI in shaping a smart future: A systematic review. Array, 26:100399, July 2025. doi: 10.1016/j.array.2025.100399. URL https://www. sciencedirect.com/science/article/pii/S2590005625000268. Angel Hsing-Chi Hwang, Q. Vera Liao, Su Lin Blodgett, Alexandra Olteanu, and Adam Trischler. “It was 80% me, 20% AI”: Seeking Authenticity in Co-Writing with Large Language Models. Proceedings of the ACM on Human-Computer Interaction, 9(2), May 2025. doi: 10.1145/3711020. URL https://doi.org/10.1145/3711020. Daphne Ippolito, Ann Yuan, Andy Coenen, and Sehmon Burnam. Creative Writing with an AI-Powered Writing Assistant: Perspectives from Professional Writers. arXiv preprint arXiv:2211.05030, November 2022. URL https://arxiv.org/abs/2211.05030. Kowe Kadoma, Danaé Metaxa, and Mor Naaman. Generative AI and Perceptual Harms: Who’s Suspected of using LLMs? In Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems, CHI ’25, New York, NY, USA, April 2025. Association for Computing Machinery. ISBN 9798400713941. doi: 10.1145/3706598.3713897. URL https://doi.org/ 10.1145/3706598.3713897. Pranav Khadpe, Ranjay Krishna, Li Fei-Fei, Jeffrey T. Hancock, and Michael S. Bernstein. Conceptual Metaphors Impact Perceptions of Human-AI Collaboration. Proceedings of the ACM on HumanComputer Interaction, 4(CSCW2), October 2020. doi: 10.1145/3415234. URL https://doi. org/10.1145/3415234. Colleen P. Kirk, Scott D. Swain, and James E. Gaskin. I’m Proud of It: Consumer Technology Appropriation and Psychological Ownership. Journal of Marketing Theory and Practice, 23(2):166– 184, March 2015. doi: 10.1080/10696679.2015.1002335. URL https://www.tandfonline. com/doi/full/10.1080/10696679.2015.1002335. Charlotte Kobiella, Yarhy Said Flores López, Franz Waltenberger, Fiona Draxler, and Albrecht Schmidt. “If the Machine Is As Good As Me, Then What Use Am I?” – How the Use of ChatGPT Changes Young Professionals’ Perception of Productivity and Accomplishment. In Proceedings of the 2024 CHI Conference on Human Factors in Computing Systems, CHI ’24, New York, NY, USA, May 2024. Association for Computing Machinery. ISBN 9798400703300. doi: 10.1145/3613904.3641964. URL https://doi.org/10.1145/3613904.3641964. 6
Nataliya Kosmyna, Eugene Hauptmann, Ye Tong Yuan, Jessica Situ, Xian-Hao Liao, Ashly Vivian Beresnitzky, Iris Braunstein, and Pattie Maes. Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task. arXiv preprint arXiv:2506.08872, June 2025. URL https://arxiv.org/abs/2506.08872. Harsh Kumar, Jonathan Vincentius, Ewan Jordan, and Ashton Anderson. Human Creativity in the Age of LLMs: Randomized Experiments on Divergent and Convergent Thinking. In Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems, CHI ’25, New York, NY, USA, April 2025. Association for Computing Machinery. ISBN 9798400713941. doi: 10.1145/3706598.3714198. URL https://doi.org/10.1145/3706598.3714198. Ryan Louie, Andy Coenen, Cheng Zhi Huang, Michael Terry, and Carrie J. Cai. Novice-AI Music Co-Creation via AI-Steering Tools for Deep Generative Models. In Proceedings of the 2020 CHI Conference on Human Factors in Computing Systems, CHI ’20, pages 1–13, New York, NY, USA, April 2020. Association for Computing Machinery. ISBN 9781450367080. doi: 10.1145/3313831.3376739. URL https://doi.org/10.1145/3313831.3376739. Lennart Luettgau, Vanessa Cheung, Magda Dubois, Keno Juechems, Jessica Bergs, Luke Symes, Henry Davidson, Bessie O’Dell, Hannah Rose Kirk, Max Rollwage, and Christopher Summerfield. People readily follow personal advice from AI but it does not improve their well-being. arXiv preprint arXiv:2511.15352, November 2025. URL https://arxiv.org/abs/2511.15352. Sebastian Maier, Manuel Schneider, and Stefan Feuerriegel. Partnering with Generative AI: Experimental Evaluation of Model-Led and Human-Led Interaction in Human-AI Co-Creation. In Proceedings of the 2026 CHI Conference on Human Factors in Computing Systems, CHI ’26, New York, NY, USA, April 2026. Association for Computing Machinery. ISBN 9798400722783. doi: 10.1145/3772318.3791185. URL https://doi.org/10.1145/3772318.3791185. Patrick E. McKnight and Todd B. Kashdan. Purpose in Life as a System that Creates and Sustains Health and Well-Being: An Integrative, Testable Theory. Review of General Psychology, 13(3):242– 251, September 2009. doi: 10.1037/a0017152. URL https://doi.org/10.1037/a0017152. Michele Newman, Kaiwen Sun, Ilena B. Dalla Gasperina, Grace Y. Shin, Matthew Kyle Pedraja, Ritesh Kanchi, Maia B. Song, Rannie Li, Jin Ha Lee, and Jason Yip. “I want it to talk like Darth Vader”: Helping Children Construct Creative Self-Efficacy with Generative AI. In Proceedings of the 2024 CHI Conference on Human Factors in Computing Systems, CHI ’24, New York, NY, USA, May 2024. Association for Computing Machinery. ISBN 9798400703300. doi: 10.1145/3613904.3642492. URL https://doi.org/10.1145/3613904.3642492. Jon L. Pierce, Tatiana Kostova, and Kurt T. Dirks. Toward a Theory of Psychological Ownership in Organizations. Academy of Management Review, 26(2):298–310, April 2001. doi: 10.5465/amr. 2001.4378028. URL https://journals.aom.org/doi/10.5465/amr.2001.4378028. Tejaswi Polimetla, Katy Ilonka Gero, and Elena L. Glassman. A Paradigm for Creative Ownership. In Proceedings of the 2026 CHI Conference on Human Factors in Computing Systems, CHI ’26, New York, NY, USA, April 2026. Association for Computing Machinery. ISBN 9798400722783. doi: 10.1145/3772318.3791316. URL https://doi.org/10.1145/3772318.3791316. James Prather, Brent N. Reeves, Juho Leinonen, Stephen MacNeil, Arisoa S. Randrianasolo, Brett A. Becker, Bailey Kimmel, Jared Wright, and Ben Briggs. The Widening Gap: The Benefits and Harms of Generative AI for Novice Programmers. In Proceedings of the 2024 ACM Conference on International Computing Education Research - Volume 1, ICER ’24, pages 469–486, New York, NY, USA, August 2024. Association for Computing Machinery. ISBN 9798400704758. doi: 10.1145/3632620.3671116. URL https://doi.org/10.1145/3632620.3671116. Irene Rae. The Effects of Perceived AI Use On Content Perceptions. In Proceedings of the 2024 CHI Conference on Human Factors in Computing Systems, CHI ’24, New York, NY, USA, May 2024. Association for Computing Machinery. ISBN 9798400703300. doi: 10.1145/3613904.3642076. URL https://doi.org/10.1145/3613904.3642076. Jessica A. Reif, Richard P. Larrick, and Jack B. Soll. Evidence of a social evaluation penalty for using AI. Proceedings of the National Academy of Sciences, 122(19):e2426766122, May 2025. doi: 10.1073/pnas.2426766122. URL https://www.pnas.org/doi/10.1073/pnas.2426766122. 7
Evan F. Risko and Sam J. Gilbert. Cognitive Offloading. Trends in Cognitive Sciences, 20(9):676–688, September 2016. doi: 10.1016/j.tics.2016.07.002. URL https://doi.org/10.1016/j.tics. 2016.07.002. Oliver Schilke and Martin Reimann. The transparency dilemma: How AI disclosure erodes trust. Organizational Behavior and Human Decision Processes, 188:104405, May 2025. doi: 10.1016/j.obhdp.2025.104405. URL https://www.sciencedirect.com/science/article/ pii/S0749597825000172. Jwawon Seo, Elmira Deldari, and Helena M. Mentis. Whose Code Is It? How AI Autonomy Reshapes Ownership, Responsibility, and Disclosure in AI-Assisted Programming. In Proceedings of the 31st International Conference on Intelligent User Interfaces, IUI ’26, pages 393–425, New York, NY, USA, March 2026. Association for Computing Machinery. ISBN 9798400719844. doi: 10.1145/3742413.3789121. URL https://doi.org/10.1145/3742413.3789121. Omar Shaikh, Michelle S. Lam, Joey Hejna, Yijia Shao, Hyundong Justin Cho, Michael S. Bernstein, and Diyi Yang. Aligning Language Models with Demonstrated Feedback. In The Thirteenth International Conference on Learning Representations, April 2025. URL https://openreview. net/forum?id=1qGkuxI9UX. Mrinank Sharma, Miles McCain, Raymond Douglas, and David Duvenaud. Who’s in Charge? Disempowerment Patterns in Real-World LLM Usage. In Forty-third International Conference on Machine Learning, July 2026. URL https://openreview.net/forum?id=dhzRnzD9jR. Chenglei Si, Diyi Yang, and Tatsunori Hashimoto. Can LLMs Generate Novel Research Ideas? A Large-Scale Human Study with 100+ NLP Researchers. In The Thirteenth International Conference on Learning Representations, April 2025. URL https://openreview.net/forum? id=M23dTGWCZy. Zhivar Sourati, Farzan Karimi-Malekabadi, Meltem Ozcan, Colin McDaniel, Alireza Ziabari, Jackson Trager, Ala N. Tak, Meng Chen, Fred Morstatter, and Morteza Dehghani. The shrinking landscape of linguistic diversity in the age of large language models. Nature Human Behaviour, August 2026. doi: 10.1038/s41562-026-02550-0. URL https://www.nature.com/articles/ s41562-026-02550-0. Maxim Topaz, Nir Roguin, Pallavi Gupta, Zhihong Zhang, and Laura-Maria Peltonen. Fabricated citations: an audit across 2·5 million biomedical papers. The Lancet, 407(10541):1779–1781, May 2026. doi: 10.1016/S0140-6736(26)00603-3. URL https://www.thelancet.com/journals/ lancet/article/PIIS0140-6736(26)00603-3/fulltext. Priyan Vaithilingam, Tianyi Zhang, and Elena L. Glassman. Expectation vs. Experience: Evaluating the Usability of Code Generation Tools Powered by Large Language Models. In Extended Abstracts of the 2022 CHI Conference on Human Factors in Computing Systems, CHI EA ’22, New York, NY, USA, April 2022. Association for Computing Machinery. ISBN 9781450391566. doi: 10.1145/3491101.3519665. URL https://doi.org/10.1145/3491101.3519665. William H. Walters and Esther Isabelle Wilder. Fabrication and errors in the bibliographic citations generated by ChatGPT. Scientific Reports, 13(1):14045, September 2023. doi: 10.1038/ s41598-023-41032-5. URL https://www.nature.com/articles/s41598-023-41032-5. Yun Wan and Yoram M. Kalman. Diverse AI personas can mitigate the homogenization effect in human-AI collaborative ideation. Computers in Human Behavior: Artificial Humans, 8:100289, May 2026. doi: 10.1016/j.chbah.2026.100289. URL https://www.sciencedirect.com/ science/article/pii/S294988212600040X. Zhengxiang Wang, Nafis Irtiza Tripto, Solha Park, Zhenzhen Li, and Jiawei Zhou. Catch Me If You Can? Not Yet: LLMs Still Struggle to Imitate the Implicit Writing Styles of Everyday Authors. In Findings of the Association for Computational Linguistics: EMNLP 2025, pages 10040–10055, Suzhou, China, November 2025. Association for Computational Linguistics. doi: 10.18653/v1/ 2025.findings-emnlp.532. URL https://aclanthology.org/2025.findings-emnlp.532/. 8
Mick Yang, Stephen Casper, Jonathan Stray, Jasmine Li, Cameron Jones, Anna Gausen, Natasha Jaques, Brian Christian, Bálint Gyevnár, Hannah Kirk, Zhonghao He, Dan Zhao, Siao Si Looi, Joshua Levy, Kobi Hackenburg, Elizabeth Seger, Matt Kowal, Michelle Malonza, Luke Hewitt, Hause Lin, Maarten Sap, Dylan Hadfield-Menell, Thomas Costello, Reihaneh Rabbany, JeanFrançois Godbout, David Rand, Atoosa Kasirzadeh, Gordon Pennycook, Yoshua Bengio, and Kellin Pelrine. AI Epistemic Risks: Emerging Mechanisms & Evidence. SSRN preprint, June 2026. URL https://ssrn.com/abstract=6873005. Sunny Yu, Myra Cheng, Ahmad Jabbar, Ilia Sucholutsky, Katherine M. Collins, Dan Jurafsky, and Robert D. Hawkins. Cognitive offloading and the speedup illusion in human-AI interaction. In Proceedings of the 48th Annual Meeting of the Cognitive Science Society, Rio de Janeiro, Brazil, July 2026. URL https://arxiv.org/abs/2605.23177. Tim Zindulka, Sven Goller, Daniela Fernandes, Robin Welsch, and Daniel Buschek. The AI Memory Gap: Users Misremember What They Created With AI or Without. In Proceedings of the 2026 CHI Conference on Human Factors in Computing Systems, CHI ’26, New York, NY, USA, April 2026. Association for Computing Machinery. ISBN 9798400722783. doi: 10.1145/3772318.3791494. URL https://doi.org/10.1145/3772318.3791494.
A
Survey
We include our Qualtrics survey questions. Response formats are marked MC (single choice), SA (select all that apply), DD (dropdown), and FF (free text). Conditional display logic and the taskreflection section are noted where they apply. Before beginning the survey, participants reviewed a consent page: “I am 18 or older and I consent to take part in this survey. [MC: Yes, I consent / No].” A.1
Demographic information • What is your age? [DD: Under 18; 18–24; 25–34; 35–44; 45–54; 55–64; 65 or older] • In which country do you currently live? [DD: country list] • How do you describe your gender? [MC: Woman; Man; Non-binary; Prefer to self-describe (FF); Prefer not to say] • How do you describe your race or ethnicity? [SA: Native American; Asian; African American; Hispanic or Latino; Middle Eastern; Pacific Islander; Caucasian; Other (FF); Prefer not to say] • What is the highest level of education you have completed? [DD: Less than high school; High school / GED; Some college; Associate’s degree; Bachelor’s degree; Master’s degree; Doctorate / Professional degree; Prefer not to say] • What best describes your current employment status? [DD: Employed full-time; Employed part-time; Self-employed; Student; Not currently employed; Retired; Prefer not to say] • Which best describes the industry you work in? (If not currently working, select “Not applicable.”) [DD: Healthcare; Education; Technology / software; Finance / insurance; Retail / hospitality; Manufacturing / trades; Government / public sector; Media / arts / design; Legal; Science / engineering; Transportation / logistics; Other; Not applicable] • What is your job title or main role? [FF]
A.2
AI usage • In the past 3 months, how often have you used any AI chatbot or assistant (e.g., ChatGPT, Gemini, Claude, Copilot, or an AI voice assistant)? [DD: Never; Less than monthly; Monthly; Weekly; Several times a week; Daily; Many times a day] • Which AI tools do you use? [SA: ChatGPT (OpenAI); Google Gemini; Claude (Anthropic); Grok (xAI); Microsoft Copilot; Meta AI; Perplexity; DeepSeek; An AI image generator/editor (e.g., Midjourney, Nano Banana); An AI video generator/editor (e.g., Google Veo, Runway); An AI music or audio generator/editor (e.g., Suno); An AI voice assistant or voice mode (e.g., ChatGPT voice, Gemini Live, Siri, Alexa); An AI coding assistant (e.g., 9
Cursor, Copilot); An AI companion/character app (e.g., Character.AI, Replika); An AI chat embedded in a different tool/website (please specify what tool) (FF); Another AI tool (FF)] • When you use AI, how do you interact with the AI? [SA: Typing; Voice; Image/video upload; Document/file upload; Tool call/app integrations/MCPs; Other (FF)] • When did you start using AI tools regularly? [DD: I don’t use them regularly; In the last 3 months; 3–6 months ago; 6–12 months ago; 1–2 years ago; More than 2 years ago] • If you are employed, how much do you personally use AI in your job? [DD: Not at all; Rarely; Sometimes; Often; Constantly; Not applicable] • For the AI tools you use, who pays? [SA: I only use free versions; I pay out of my own pocket; My employer or company provides/pays; My school or university provides/pays; Someone else (e.g., family) pays; I get a student or other discount; Prefer not to say] • About how much do you personally pay for AI tools per month, in total? [DD: Less than $20; $20–29; $30–49; $50–99; $100–199; $200–499; $500 or more; Prefer not to say] (shown only if “I pay out of my own pocket” selected) • About how long have you been paying (out of pocket) for an AI tool? [DD: Less than 3 months; 3–6 months; 6–12 months; 1–2 years; More than 2 years; Prefer not to say] (shown only if “I pay out of my own pocket” selected) A.3
Task reflections (two blocks, randomized order)
Preface shown before the blocks: “In this next section, we’ll ask about two recent tasks where you used AI — one where the result DID feel like yours, and one where the result DID NOT feel like yours. These will appear in a random order, so please check each section header before answering.” Each block began: “For the next questions, ‘AI’ means tools like ChatGPT, Gemini, Claude, Copilot, or other chatbots and assistants that generate text, answers, images, or recommendations. Please tell us about a task that you recently used AI to complete and where the result [felt very much like yours | did not really feel like yours]. Think about whether the result felt like yours, how much you shaped the task direction, and how independently you completed the task. Consider describing a creative project, something you planned for another person, a chore or errand, a decision you made, a work or school project, or anything else that comes to mind. There are no right or wrong answers — we’re just as interested in small, everyday tasks as in big ones.” • Briefly, what was the task? [FF] • Why did you need to do this task, and how did you use AI to complete it? If you used any other resources to complete the task, please describe them. [FF] • In what ways, if any, does the result of this task reflect who you are — your values, personality, or identity? If it doesn’t feel like a reflection of you, why not? [FF] • Reflect on other similar tasks you have done. How much does this task feel like your own work? Consider how you did/did not shape the direction of the task, any decisions you made, or how much you relied on others. [FF] • Think about how you feel about the outcome of this work. Are you proud of the final result and would you be comfortable telling others you used AI for it? [FF] • Was using AI worth it to you? Consider why you chose to use AI for this task and how the result might feel or be different if you hadn’t used AI. [FF] • Would you still have done this task without AI? [MC: Definitely yes; Probably yes; Might or might not; Probably not; Definitely not] • Would you be able to do this task without AI? [MC: Definitely yes; Probably yes; Might or might not; Probably not; Definitely not] A.4
Common tasks with AI
Below are common tasks done using AI. Select any you have used AI to do in the past 30 days (1 month). [SA] 10
• Looking up or checking specific information; Learning about a topic or how something works; How-to help or step-by-step guidance; Summarizing, explaining, or translating a document, article, or video; Editing or improving something I’d already written; Writing messages, emails, or other communication; Creative writing (stories, poems, fiction); Brainstorming or coming up with ideas; Making or editing images, audio, or video; Cooking, recipes, or meal planning; Health, fitness, beauty, or self-care; Medical or health-related questions; Shopping or product research; Planning something (a trip, an event, a gift, a schedule); Advice on a personal problem, relationship, or how I feel; Coding, math, or data analysis; Casual conversation, games, or roleplay; Something else (please specify) (FF) A.5
Comparison and closing • As a reminder, here are the tasks you described using AI for earlier. [Task 1; Task 2 inserted in.] Why did you choose to talk about these tasks and not any of the other tasks you selected above? [FF] • Between the two tasks you selected, what made the biggest difference in how much ownership you felt? If they felt similar, tell us that too. [FF] • Is there anything else about how you use AI — or about this survey — you’d like to share? [FF]
B
Survey Participation
In Table 1, we report the number of participants who completed each section of our survey. Since all questions were optional, the number of participants n who completed each section varies. The main source of attrition comes from the task reflections section. In our figures and analyses, we report all results available and the number of participants n involved. The median completion time for the survey was 24 minutes. Table 1: Number of participants who completed each section of the survey. Stage
n
Opened the survey and consented Continued past the demographics page Continued past the AI-usage page (reached the task reflections) Described at least one task (high-ownership 57; low-ownership 56; both 49) Completed the common-tasks checklist Answered “why these tasks” / “biggest difference” Reached the end of the survey
C
117 108 104 64 53 50 / 48 52
Demographics
We collect personal demographics as well as AI usage preferences, patterns, and payments to build additional context to ground our study. In Table 2, we report the personal demographics of our participants. Due to our sourcing channels of university mailing lists, Slack channels, professional networks, and social media, our sample is biased with a bimodal distribution of young adults and senior professionals, who are predominantly Caucasian and US-centric, well-educated, and experienced with AI in the technology or education sector. This sample is a first step towards understanding trends in ownership in AI-assisted tasks. We encourage future work to apply our framework to a broader and more representative population. In Table 3, we report AI usage frequency and familiarity from the respondents. Our participants lean towards active users of AI tools and have been using these tools for a while. Figure 2 highlights the AI tools participants use. Figure 3 shows the modes of interaction with AI tools. Table 4 reveals payment patterns of AI tools. 11
C.1
Personal Demographics
Table 2: Demographics of survey respondents. All questions were optional, so percentages are calculated based on the number of respondents for each question. For select-all-that-apply items, percentages do not sum to 100%. Other includes Argentina, New Zealand, Cyprus, Germany, and the Netherlands, each represented by 1 respondent. Demographic
Age (n=62)
Country of Residence (n=60)
Gender (n=64)
Race/Ethnicity (select all) (n=64)
Education (n=64)
Employment Status (n=63)
Work Industry (n=63)
Response
n
%
18–24 25–34 35–44 45–54 55–64 65 or older
13 3 5 21 13 7
21.0% 4.8% 8.1% 33.9% 21.0% 11.3%
United States Australia Canada India United Kingdom Other*
43 4 4 2 2 5
71.7% 6.7% 6.7% 3.3% 3.3% 8.3%
Man Woman Other Prefer not to say
32 27 4 1
50.0% 42.2% 6.2% 1.6%
American Indian or Alaska Native Asian Black or African American Hispanic or Latino/a/e Middle Eastern or North African Native Hawaiian or Other Pacific Islander White or Caucasian Other Prefer not to say
0 7 0 4 0 0 51 1 5
0.0% 10.9% 0.0% 6.2% 0.0% 0.0% 79.7% 1.6% 7.8%
Less than high school High school / GED Some college Associate’s degree Bachelor’s degree Master’s degree Doctorate / Professional degree Prefer not to say
0 2 14 0 17 18 12 1
0.0% 3.1% 21.9% 0.0% 26.6% 28.1% 18.8% 1.6%
Employed full-time Employed part-time Self-employed Student Not currently employed Retired
25 3 18 11 1 5
39.7% 4.8% 28.6% 17.5% 1.6% 7.9%
Education Finance / insurance Government / public sector Healthcare Legal Media / arts / design Retail / hospitality Science / engineering Technology / software Other Not applicable
10 1 2 3 3 3 1 4 20 8 8
15.9% 1.6% 3.2% 4.8% 4.8% 4.8% 1.6% 6.3% 31.7% 12.7% 12.7%
12
C.2
AI Usage
Table 3: Self-reported AI usage among survey respondents. Percentages are calculated based on the number of respondents for each question, as not all participants answered every question. Question
Usage Frequency (Past 3 Months) (n=54)
Length of AI Use (n=64)
AI Use in Job (n=57)
Response
n
%
Many times a day Daily Several times a week Weekly Monthly Less than monthly
35 5 8 2 2 2
64.8% 9.3% 14.8% 3.7% 3.7% 3.7%
More than 2 years ago 1–2 years ago 6–12 months ago 3–6 months ago In the last 3 months I don’t use them regularly
30 12 8 2 8 4
46.9% 18.8% 12.5% 3.1% 12.5% 6.2%
Constantly Often Sometimes Rarely Not at all Not applicable
28 12 6 4 2 5
49.1% 21.1% 10.5% 7.0% 3.5% 8.8%
Which AI tools did you use? (n=64) Claude ChatGPT Google Gemini AI coding assistant AI voice assistant Microsoft Copilot AI image generator/editor Embedded AI chat Other Perplexity AI audio generator/editor AI video generator/editor DeepSeek Grok AI companion/character app Meta AI
14 (22%) 12 (19%) 12 (19%) 11 (17%) 11 (17%) 7 (11%) 5 (8%) 3 (5%) 1 (2%)
0
10
24 (38%) 21 (33%) 19 (30%) 19 (30%)
20
57 (89%)
45 (70%) 43 (67%)
30 40 Number of respondents
50
60
Figure 2: AI tools used by survey respondents (n = 64; select all that apply). Bar labels show the number of respondents who selected each tool, with the percentage of respondents to this question in parentheses. Because participants could select multiple tools, percentages do not sum to 100%. “Other” includes Kimi (x2) and Codex (x2) and 12 other tools mentioned once. How do you interact with AI? (n=64) Typing Document/file upload Image/video upload Voice Tool call/app integration/MCP Other
64 (100%) 50 (78%) 34 (53%) 32 (50%) 25 (39%) 5 (8%)
0
10
20
30 40 Number of respondents
50
60
70
Figure 3: Ways that survey respondents interact with AI tools (n = 64; select all that apply). Bar labels show the number of respondents who selected each tool, with the percentage of respondents to this question in parentheses. Because participants could select multiple tools, percentages do not sum to 100%. “Other” includes API calls (x2), HTML document reviews (x1), and Git (x1). 13
C.3
Paid AI Usage
Table 4: Self-reported paid AI usage among survey respondents. Percentages are calculated based on the number of respondents for each question, as not all participants answered every question. For select-all-that-apply items, percentages do not sum to 100%. The Monthly Spend and Length of Paid Use questions are restricted to respondents who reported paying for AI usage out of pocket. Question Who Pays (select all) (n=63)
Monthly Spend ($) (n=36)
Length of Paid Use (n=36)
Response
n
%
I only use free versions I pay out of my own pocket I get a student or other discount My employer or company provides/pays My school or university provides/pays
17 36 4 33 6
27.0% 57.1% 6.3% 52.4% 9.5%
Less than $20 $20-29 $30-49 $50-99 $100-199 $200-499 $500 or more Prefer not to say
6 2 8 5 7 6 1 1
16.7% 5.6% 22.2% 13.9% 19.4% 16.7% 2.8% 2.8%
More than 2 years 1–2 years 6–12 months 3–6 months Less than 3 months
14 8 10 2 2
38.9% 22.2% 27.8% 5.6% 5.6%
D
Survey Results
D.1
Interest and Ability in Performing Tasks without AI
In Figure 1, we highlight the difference between participants being able to do the task without AI across high and low ownership tasks, which reveals a difference among those who responded “definitely yes”. When compared to whether they would have done the task without AI, there is less of a difference between high and low ownership tasks. Definitely not
Probably not
Might or might not
Probably yes
Definitely yes
Bar labels: % of respondents
Would you still have done this task without AI? High ownership n = 55
Low ownership n = 56
11
16
9
20
9
24 11
40
14
46
Would you be able to do this task without AI? High ownership
11
n = 55
Low ownership n = 56
4
13 13
7 13
27 9
42 63
Figure 4: Responses to questions about task completion without AI: whether respondents would have done the task (top) and whether they would have been able to do the task (bottom). Within each subplot, responses are split by high ownership and low ownership. Bars show the proportion of respondents in each group selecting each of the five Likert options (Definitely yes, Probably yes, Might or might not, Probably not, Definitely not). 14
D.2
Pride and Disclosure
When asked about whether they felt proud of the result, respondents are generally proud of high ownership tasks and mixed for low ownership tasks. Disclosure of AI usage remains similar across high and low ownership tasks, though 10% are uncomfortable disclosing AI use for low ownership tasks. When pooled across high and low ownership tasks, comfort with disclosure is loosely tied to pride. Proud of the result?
Proud
High ownership
Mixed / satisfied
74
n = 43
Low ownership
25
n = 44
43
Comfortable
Depends
Uncomfortable
81
n = 48
Low ownership
17
80
n = 41
10
Comfortable telling others, by pride (pooled)
Comfortable
Proud
Depends
Mixed or not proud
10
13
73
n = 33
2
Uncomfortable
85
n = 39
2
32
Comfortable telling others you used AI? High ownership
Not proud
23
18
3 9
Bar labels: % of answers that addressed the question
Figure 5: Answers to “Are you proud of the final result and would you be comfortable telling others you used AI for it?”, coded by fixed keyword rules (proud / mixed or merely satisfied / not proud; comfortable / depends on audience, stakes or being able to explain the extent / uncomfortable). Bottom panel: both tasks pooled, comparing pride and disclosure trends (n = 72). D.3
Common Tasks and AI Memory Gap
Used AI for this in the past 30 days (% of respondents, n = 53)
Tasks chosen for reflection (% of the 107 tasks)
What people use AI for, and what they chose to reflect on Learning about a topic
3%
Looking up specific information
3%
Editing something already written
94% 87% 9%
How-to / step-by-step guidance
79%
3%
Summarizing, explaining, translating
70% 12%
Brainstorming ideas
70%
5%
70%
Coding, math, data analysis
24%
Medical or health questions
1%
60%
Writing messages, emails, communication
26%
Shopping or product research
1%
Planning (trip, event, gift, schedule)
1%
Making / editing images, audio, video Health, fitness, beauty, self-care
62% 57% 51% 49%
5%
45%
0%
40%
Advice on a personal problem
1%
Cooking, recipes, meal planning
0%
Casual conversation, games, roleplay
0%
Creative writing
28% 25% 25%
2%
Something else
5%
23% 15%
0%
25%
50%
75%
100%
Figure 6: Everyday AI use versus the tasks chosen for reflection. Grey: percent of checklist respondents (n = 53) who used AI for each category in the past 30 days. Blue: percent of the 107 reflected tasks (mapped to checklist category via rule-based coding and LLM judging). 15
On the common tasks checklist (Appendix A.4), respondents (n = 53) reported a median of 9 of 17 task categories in the past 30 days. The most common were learning about a topic (94%) and looking up specific information (87%). Of the tasks respondents chose to reflect on, only a handful were lookup- or learning-related tasks. Tasks respondents chose to reflect on were often writing and technical tasks. Respondents’ explanations for their choices (Appendix A.5) were due to recency bias, salience, and having a clear result to judge ownership of. Ownership reflections often involved tasks that produced a significant artifact rather than informational use from day-to-day interaction. This is consistent with the literature on people underestimating their AI usage [Yu et al., 2026] and forgetting what they created with AI [Zindulka et al., 2026]. D.4
Anecdotes on Tasks
Table 5: Pairs of high and low ownership tasks described by the same respondent, with the respondent’s own explanation of the difference. The tag under each task is that respondent’s answer to “Would you be able to do this task without AI?”. Respondent
High ownership
Low ownership
What’s the difference?
Tech executive
Graphic novels adapted from their own novels
Negotiating an Amazon refund through an AI agent
Solo-practice attorney
An AI pipeline that files scanned documents to the right matter
“I didn’t draw the art, but I feel the same as if I would have hired a human artist” “I rewrite all communication generated by AI so I own it”
definitely not
definitely yes
A memorandum supporting an order to show cause definitely yes
definitely not
Research mentor
A team of agents building speech-recognition software
IT analyst and digital artist
A 3-D multichannel audio-synthesis environment
definitely not
definitely not
Directing a student paper that students revised with AI
“Who’s doing the prompting, who’s defending the claims”
definitely yes
An adventure game about personal issues, built hands-off
“Task 2 was intended to be as hands off as possible, so I would not call the results mine”
definitely not
Artisan
Diagnosing a car problem while steering the search probably not
Diagnosing a car problem while passively receiving the answer
“how I engaged in the direction of the search”
probably not
Retiree
Planning an overseas trip through extended back-and-forth
Identifying a bird from a photo might or might not
“a long process with much back and forth. I think that made me feel more ownership”
probably yes
Student (research intern)
Debugging cluster software by talking through the errors
A school-club website probably not
probably yes
Student
Personal blog posts probably yes
A GIS data-processing pipeline probably not
Postdoctoral researcher
Streaming camera data from an embedded device for robot training
Engineer
A map of locations for a website
A conference website definitely yes
probably yes
probably yes
Clinical informatics specialist QA analyst
Deciding which jobs to target in a career change
A comedic letter about a friend, read at a party definitely yes
A LinkedIn community post definitely yes
might or might not
Test-coverage tracking and documentation
Formatting test cases definitely yes
definitely yes
16
“Claude was as confused as me, so when the problem finally did get solved I felt like I actually played a role” “The % of text created that was typed by me rather than copy-and-pasted” “The first task I did on my own volition and self-interest, the second task I did because it was a responsibility” “The strength of my initial vision for what the result should be, and the amount of existing work I had before I used the tool” “the AI helped with the pattern recognition that I wasn’t seeing in my own experience and choices” “the ownership felt the same. It was work that I did with the use of the tool”