How Do Human Creators Embrace Human-AI Co-Creation? A Perspective on Human Agency of Screenwriters
Open accessPublished version of a preprint
The DOI matches its Crossref record; 71 of the 80 references it cites were found in a registry or an index; 11 of the 20 sources its claims cite were checked for support.
Abstract
Generative AI has greatly transformed creative work in various domains, such as screenwriting. To understand this transformation, prior research often focused on capturing a snapshot of human-AI co-creation practice at a specific moment, with less attention to how humans mobilize, regulate, and reflect to form the practice gradually. Motivated by Bandura’s theory of human agency, we conducted a two-week study with 19 professional screenwriters to investigate how they embraced AI in their creation process. Our findings revealed that screenwriters not only mindfully planned, foresaw, and responded to AI usage, but, more importantly, through reflections on practice, they developed themselves and human-AI co-creation paradigms, such as cognition, strategies, and workflows. They also expressed various expectations for how future AI should better support their agency. Based on our findings, we conclude this paper with extensive discussion and actionable suggestions to screenwriters, tool developers, and researchers for sustainable human-AI co-creation.
Overview
CoChat Summary
TL;DR
This two-week qualitative study with 19 professional screenwriters uses Bandura’s human agency theory to examine how screenwriters plan, anticipate, regulate, and reflect on human-AI co-creation. It finds that screenwriters actively steer AI, develop strategies, workflows, and cognition through reflection, and call for future systems that support intentionality, forethought, self-reactiveness, and self-reflectiveness.
Key findings
- Screenwriters exercised intentionality by assigning AI four major tasks: content expansion, fragment consolidation, external knowledge acquisition, and story reconfiguration; they evaluated AI outputs along narrative alignment, structural conformity, logical consistency, and content accuracy.
- Immediate reactions to AI outputs fell into four evaluation types—intended/supportive, intended/disruptive, unintended/supportive, and unintended/disruptive—and three responses: adoption, adjustment, and avoidance.
- Through reflection, participants developed evolving strategies (e.g., transforming instruction failures into long-term success, selective withdrawal), enhanced confidence, reorganized workflows, reported concerns about skill erosion, transformed cognition, and reconsidered AI roles as dependable colleague, credible creative competitor, and emotional support partner.
- For future AI, they wanted a plan-making partner, proactive and farsighted thinker, responsible gatekeeper, and reflective mentor; some participants reported disconnected effects without evolving strategies.
Methods
- Three-stage qualitative study over two weeks: co-creation of a five-minute screenplay with AI, retrospective think-aloud, and semi-structured interview; 19 professional screenwriters were recruited via snowball sampling (6 male, 13 female; ages 20–39, mean age 29; average around 8 years of screenwriting experience).
- Participants used CloseChat with DeepSeek Reasoner under default settings (temperature 0.8, context window limit 10, max iterations 3 of 10), logged interactions in Excel, and produced annotated daily logs plus approximately 2,850 minutes of transcribed audio.
- Thematic analysis followed Braun and Clarke’s six-phase method; findings were organized by RQ1 (intentionality, forethought, self-reactiveness), RQ2 (self-reflectiveness), and RQ3 (future expectations).
Limitations
- Stated: the five-minute short-film task within two weeks may not generalize to feature-length films or television series; screenplays were not staged or produced, so performability, production constraints, and collaboration with other stakeholders were not addressed.
- Stated: snowball sampling introduces potential selection bias, and the sample shared relatively similar professional and cultural backgrounds; participants also had prior AI co-creation experience, aligning with the study’s developmental goal but limiting representativeness.
- Evident from the text: the sample is small (19) with self-reported demographics, and the study does not measure long-term skill atrophy or behavioral change beyond the two-week period and retrospective accounts.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Figures



Verification
Claims the paper makes
7 Valid · 2 Not valid · 6 Not checked · 185 Own results and interpretations
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“However, these accounts focus on isolated stages of use and therefore provide only snapshots of human-AI collaboration.”
- Also stated in: § 2.1 AI in Screenwriting · plain
“Despite these contributions, existing empirical studies provide only snapshots of how writers leverage AI at specific moments.”
What the sources say
- Partly supportedcites [19] [19] Katy Ilonka Gero, Tao Long, and Lydia B Chilton. 2023. Social Dynamics of AI Support in Creative Writing. In Proceedings of the 2023 CHI Conference on Human Factors in Computing Systems (CHI '23) . ACM, Article 245, 15 pages. doi:10.1145/3544548.3580782 - sourceExperimental — accuracy not yet measured
The Methodology (§3.1; page unavailable) describes one-hour interviews with creative writers, so the study offers a snapshot rather than tracking how AI-use practices develop over time. But Related Work (§2.1; page unavailable) says the interviews were designed to cover a range of writing processes, not an isolated stage. The participants were writers across genres, not a study of screenwriters’ AI use specifically. (LLM, conf=0.85)
- Partly supportedcites [60] [60] Yuying Tang, Haotian Li, Minghe Lan, Xiaojuan Ma, and Huamin Qu. 2025. Understanding Screenwriters' Practices, Attitudes, and Future Expectations in Human-AI Co-Creation. In Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems (CHI '25) . 1-18. doi:10.1145/3706598.3714120 - sourceExperimental — accuracy not yet measured
Section 2.2 (p. 3) says existing studies often focus on isolated stages and overlook the complete screenwriting workflow, supporting the description of limited snapshots. It does not say those studies fail to track how screenwriters’ strategies and practices develop over time; its stated gap concerns coverage of workflow stages, not longitudinal change. (LLM, conf=0.84)
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Also stated in: § 2.1 AI in Screenwriting · plain
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“Prior research has often conceptualized agency from a static standpoint, emphasizing passive influence or limiting analysis to dimensions of control and decision support [2].”
What the sources say
- Partly supportedcites [2] [2] Joshua Ashkinaze, Julia Mendelsohn, Li Qiwei, Ceren Budak, and Eric Gilbert. 2025. How AI Ideas Affect the Creativity, Diversity, and Evolution of Human Ideas: Evidence From a Large, Dynamic Experiment. In Proceedings of the ACM Collective Intelligence Conference (CI '25) . ACM, New York, NY, USA, 198-213. doi:10.1145/3715928.3737481 - sourceExperimental — accuracy not yet measured
Related Work says most experiments do not capture evolving collective dynamics, and §6.1 says static experimental designs may miss changes over time. That supports the contrast with static research, but the paper does not characterize prior research as conceptualizing agency through passive influence, control, or decision support. The paper studies passive exposure to AI ideas rather than presenting it as a prevailing conception of agency. (LLM, conf=0.82)
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Second, screenwriters employed pre-calibration for specific tasks by front-loading precise constraints, success criteria, and exemplars to stabilize response quality.”
plainstates a causeSingle case- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Our findings showed that screenwriters actively formulated plans to achieve their intentions, evaluated and regulated AI outputs, and creatively transformed disruptions into leverage for exploring new creative possibilities.”
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“To examine these expectations, we analyzed participants' explanations of why they perceived AI outputs as supportive or disruptive, revealing four recurrent dimensions that structured their evaluative criteria in human-AI co-creation.”
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
103 more claims
Create a free account to see all 200 claims and every source verdict.
- It may state two separate claims as one. 61
- Probably not a substantial claim. 53
- The quote only partly states this claim. 3
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
Each contradiction between two passages listed here was confirmed by an independent check.
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ 5.4 Screenwriters' Concerns of Skill Sets
“P16 confessed that once AI could 'simplify and extract key points,' he became 'so lazy that I didn't want to carefully read all those texts myself. '”
§ 3.2 Procedure, Apparatus, and Materials
“| 16 | 39 | Female | 15 | 5 |”
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ 5.3 Screenwriters' Reorganization of Workflows
“P8 explained that his role was to define 'the creative framework and main narrative direction' while AI 'filled in some details within the framework.'”
§ 3.2 Procedure, Apparatus, and Materials
“| 8 | 23 | Female | 4 | 2 |”
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ 5.3 Screenwriters' Reorganization of Workflows
“P7 noted that in the past, after investing significant time in one idea, she would 'just execute the previous idea,' even if it was unsatisfying.”
§ 3.2 Procedure, Apparatus, and Materials
“| 7 | 39 | Male | 20 | 3 |”
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ 5.3 Screenwriters' Reorganization of Workflows
“P1 planned to delegate summarization so he could 'focus more on weaving the plot.'”
§ 3.2 Procedure, Apparatus, and Materials
“| 1 | 29 | Female | 5 | 4 |”
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ 3 Method
“we conducted a three-stage qualitative study, consisting of a co-creation stage, a retrospective think-aloud stage, and a semi-structured interview, conducted over the span of two weeks”
§ 3.2 Procedure, Apparatus, and Materials
“Except for two participants who postponed their think-aloud sessions for personal reasons (after 9 and 16 days), all others participated within 5 days.”
4 more findings
Create a free account to see every finding of this check.
Citations
Cites
Using thematic analysis in psychology
Braun, V., & Clarke, V. (2006) · Qualitative Research in Psychology · doi:10.1191/1478088706qp063oa · Cited by 194,414
Goodman, L. A. (1961) · The Annals of Mathematical Statistics · doi:10.1214/aoms/1177705148 · Cited by 3,606
Toward a Psychology of Human Agency
Bandura, A. (2006) · Perspectives on Psychological Science · doi:10.1111/j.1745-6916.2006.00011.x · Cited by 3,459
World Medical Association (2025) · JAMA · doi:10.1001/jama.2024.21972 · Cited by 1,695
Proceedings of the 2024 CHI Conference on Human Factors in Computing Systems
(2024) · doi:10.1145/3613904 · Cited by 896
Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems
(2025) · doi:10.1145/3706598 · Cited by 534
Agency plus automation: Designing artificial intelligence into interactive systems
Heer, J. (2019) · Proceedings of the National Academy of Sciences · doi:10.1073/pnas.1807184115 · Cited by 279 · cited 2×
“In human–AI collaboration, Heer [28] examined how human con-trol is balanced with automation, while Méndez et al. [45] described agency as the degree to which the user or the system drives visualization creation.”
Novice-AI Music Co-Creation via AI-Steering Tools for Deep Generative Models
Louie, R., Coenen, A., Huang, C. Z., Terry, M., & Cai, C. J. (2020) · Proceedings of the 2020 CHI Conference on Human Factors in Computing Systems · doi:10.1145/3313831.3376739 · Cited by 271 · cited 1×
“…decision-making:inmusiccomposition,“steering”toolshelped novicespartitioncomplextasksandstrengthenedtheirsenseof ownershipandcompetence[41,58];instoryauthoring,AItools enableddesignerstofocusonnarrativechoiceswhileretaining controlovertheoutput[29,52].”
The experience of agency in human-computer interactions: a review
Limerick, H., Coyle, D., & Moore, J. W. (2014) · Frontiers in Human Neuroscience · doi:10.3389/fnhum.2014.00643 · Cited by 241 · cited 1×
“In interaction design, for example, Coyle et al. [14] measured sense of agency using reaction time as an indicator of perceived control over input devices, a framing later adopted in studies of user experience and feedback [38, 67].”
Mirowski, P., Mathewson, K. W., Pittman, J., & Evans, R. (2023) · Proceedings of the 2023 CHI Conference on Human Factors in Computing Systems · doi:10.1145/3544548.3581225 · Cited by 240 · cited 2×
“The increasing integration of AI in screenwriting transforms not only the quality and style of final outputs [43, 46] but also the practices through which story ideas are conceived and developed [31, 32, 54, 63].”
Cited by
Beyond Compliance: How AI Could Help Creative Writers by Refusing Them
Qin, H. X., Zhu, G., Fan, M., & Hui, P. (2026) · Creativity and Cognition · doi:10.1145/3803784.3807571 · Cited by 1
Farah, Y. A., & Dorneich, M. C. (2026) · Proceedings of the Human Factors and Ergonomics Society Annual Meeting · doi:10.1177/10711813261485922
AI for Creative Visual Content Generation Editing and Understanding
Wei, Z., Tang, Y., Tang, M., & Rao, A. (2026) · ACM SIGGRAPH Conference (SIGGRAPH) · doi:10.1145/3799820.3812481 · Cited by 1
Self-citationYuying, T. A. N. G., Qiu, W., Zhang, Y., ZHANG, B., Zhang, W., Ma, X., & Qu, H. (2026) · Creativity and Cognition · doi:10.1145/3803784.3816817
Self-citation(2000) · Time to knit · doi:10.1145/3772318.3808940
Metadata
- Authors
- Tang, Y., Zhou, J., Li, H., Xie, X., Ma, X., & Qu, H.
- Affiliations
- Hong Kong University of Science and Technology (HK)
- Microsoft Research Asia (China) (CN)
- Venue
- Proceedings of the 2026 CHI Conference on Human Factors in Computing Systems
- Publisher
- ACM
- Published
- 2026-04-13
- Pages
- 1-19
- OpenAlex
- W7128380869
- Cited by
- 5
- References
- 73
- Citations by year
- 2026: 5
- FWCI
- 360.11
- Type
- conference paper
- Language
- en
- Open access
- hybrid
- doi.org
- Licence
- cc-by-nc-nd
- Locations
- CHI Conference on Human Factors in Computing Systems (publishedVersion, cc-by-nc-nd)
- arXiv (Cornell University) (submittedVersion)
- Rare & Special e-Zone (The Hong Kong University of Science and Technology) (submittedVersion)
- arXiv (Cornell University)
- Topics
- Social Sciences › Arts and Humanities › Visual Arts and Performing Arts › Art, Technology, and Culture
- Physical Sciences › Computer Science › Computer Graphics and Computer-Aided Design › Digital Media and Visual Art
- Social Sciences › Arts and Humanities › Visual Arts and Performing Arts › Artistic and Creative Research
- Keywords
- human-AI co-creation, screenwriters, human agency, generative AI, creative work, reflective practice, AI-assisted writing
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
The claim verdicts this report counts are experimental: their accuracy is not yet measured.
A language model drew the claims from the paper on 2026-09-30, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-09-30 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Cited sources are checked with a language model’s help (gpt-6-sol).
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
How to cite
Cite CoChat for the checks on this page, and the paper for its own findings.
- This page
- CoChat. (2026). Citation check: How Do Human Creators Embrace Human-AI Co-Creation? A Perspective on Human Agency of Screenwriters. https://app.cochat.ai/papers/10.1145/3772318.3790300
- The paper
- Tang, Y., Zhou, J., Li, H., Xie, X., Ma, X., & Qu, H. (2026). How Do Human Creators Embrace Human-AI Co-Creation? A Perspective on Human Agency of Screenwriters. Proceedings of the 2026 CHI Conference on Human Factors in Computing Systems. https://doi.org/10.1145/3772318.3790300
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.