Research project
Beyond AI Share
Beyond AI Share is a preregistered survey and paired vignette study examining how people perceive control, authorship and authenticity when generative AI becomes part of creative work.
Study 1 is complete and the preprint is public. This independent research project is not currently recruiting or collecting data.
What the study asked
A single estimate of how much AI contributed to a finished work compresses very different creative processes into one number. A person may set the direction, compare alternatives, reject outputs, revise extensively and decide the final form - or accept a nearly finished result after a short brief.
The study therefore examined declared AI share, perceived control, authorship and authenticity as related but distinct judgments. It asked about participants' perceptions and experiences; it did not attempt to define legal authorship.
The study did not directly test whether control matters more than AI share. Its results do not support that comparative claim.
Study design
Study 1 used a cross-sectional online survey administered in Polish, with a paired within-person vignette component. Main-study data were collected from 15 June to 5 July 2026. The study was preregistered on OSF before the first recorded main-study response.
- Total analytic sample
- 429 completed, consenting adult responses
- Creator AI-user subgroup
- 164 participants for H1-H3
- Vignette comparison
- 428 complete paired cases
- Declared AI share
- Self-report scale from 1 (none or almost none) to 5 (very large)
The vignette task contrasted a human-directed process with acceptance of a near-final generated output. The scenarios differed at the same time in initial intention, selection, editing and final authority. The comparison was therefore a bundled judgment task, not a factorial experiment isolating direction, selection or revision.
Key findings
Declared AI share and authorship
The preregistered negative association was not supported
Result. Among creator AI users, r = .141, 95% CI [-.012, .288], p = .071, N = 164.
Interpretation. The observed direction was positive and the confidence interval included zero.
Boundary. This is not evidence that greater AI involvement increases authorship; the authorship measure also had low reliability.
Perceived control and authorship
A modest positive association supported the prediction
Result. r = .234, 95% CI [.084, .374], p = .003, N = 164. The simple regression gave b = .275, SE = .090 and R² = .055.
Interpretation. Participants who reported more control also tended to report more authorship.
Boundary. The correlational design cannot show which judgment comes first, and both short measures had low reliability.
Control as a moderator of authenticity
The predicted buffering interaction was not supported
Result. The AI share by control interaction was b = .043, SE = .073, 95% CI [-.102, .187], p = .559, N = 164.
Interpretation. The model did not establish that control weakens a harmful association between AI share and authenticity.
Boundary. A positive main association for control cannot substitute for the preregistered interaction.
Expressive orientation and concern about AI
The pattern was cautiously consistent with H4
Result. Expressive orientation correlated with creative identity threat, r = .341, 95% CI [.254, .422], p < .001, N = 429, and with concern about a typical AI aesthetic, r = .504, p < .001. It was not associated with the general authenticity score, r = .011, p = .827.
Interpretation. Stronger expressive values accompanied some, but not all, AI-related concerns.
Boundary. Mixed indicators, limited reliability and multiple related outcomes rule out a broad claim about authenticity.
Paired vignette judgments
The human-directed workflow received higher combined ratings
Result. Human-directed use: M = 3.501, SD = 1.081. Near-final generator use: M = 2.665, SD = 1.132. Mean paired difference = .836, 95% CI [.715, .956], t(427) = 13.628, p < .001, Cohen's dz = .659, N = 428.
Interpretation. Participants clearly distinguished the two process descriptions across authorship, authenticity and control.
Boundary. Because several features changed together, the result concerns the package of process differences and does not isolate a causal component.
What the results do not establish
- a causal effect of control, AI share or any workflow feature;
- a universal threshold at which AI involvement changes authorship;
- a legal definition of who counts as an author;
- that control always matters more than declared AI share;
- the independent effect of direction, selection, revision or veto in the bundled vignettes; or
- population estimates for all creators, AI users or adults.
Limitations and research integrity
The sample was recruited online through non-probability outreach and should be understood as a convenience sample. Reliable counts of all survey views, starts and incomplete submissions were not retained, so response, participation and completion rates could not be calculated.
- The design was cross-sectional and primarily self-report.
- Declared AI share was a subjective five-level estimate, not an audit of artifacts or production histories.
- Reliability was low for the short authorship (alpha = .519), authenticity (.521) and control (.452) composites, and limited for identity threat (.646).
- The fixed-order vignette contrast bundled multiple features and also differed in protagonist gender and task context.
- Completion-time sensitivity and multiple related outcomes add further uncertainty.
- The preprint has not been peer-reviewed.
The study did not receive prospective research ethics committee approval or a formal exemption. This is an important limitation and should be considered when interpreting or reusing the findings.
Public record and materials
Zenodo preprint
Beyond AI Share: A Preregistered Survey and Vignette Study of Perceived Control, Authorship, and Authenticity in AI-Assisted Creative Practice
- Author
- Feliks Mamczur
- Published
- Version
- 1.0
- Status
- Preprint; not peer-reviewed
- License
- CC BY 4.0
OSF preregistration
Is It Still My Work? Authorship, Authenticity and Control in AI-Assisted Creative Practice
- Registered
- Registration status
- Accepted
- Access
- Publicly accessible when verified on
Other public materials
Appendix and preregistration materials
The Zenodo record includes the preprint and Appendix A. The public OSF registration includes preregistration materials and a codebook. The de-identified quantitative dataset and analysis scripts were not public at the time of the preprint.
How to cite the preprint
Mamczur, F. (2026). Beyond AI Share: A Preregistered Survey and Vignette Study of Perceived Control, Authorship, and Authenticity in AI-Assisted Creative Practice [Preprint]. Zenodo. https://doi.org/10.5281/zenodo.21705721
Next research
The project identifies questions for prospective research rather than a settled standard. A future study should separate direction, selection, revision, the ability to reject an output and final decision authority instead of changing them together.
Any future recruitment would require an ethics procedure, approval or formal exemption to be established in advance. No new recruitment or data collection has started.