CoChat

The AI Ghostwriter Effect: When Users do not Perceive Ownership of AI-Generated Text but Self-Declare as Authors

Open access
Draxler, Fiona, Werner, Anna, Lehmann, Florian, Hoppe, Matthias, Schmidt, Albrecht, Buschek, Daniel, Welsch, Robin · 2024 · ACM Transactions on Computer-Human Interaction

The DOI matches its Crossref record; 119 of the 119 references it cites were found in a registry or an index; 8 of the 16 sources its claims cite were checked for support.

Abstract

Human-AI interaction in text production increases complexity in authorship. In two empirical studies (n1 = 30 & n2 = 96), we investigate authorship and ownership in human-AI collaboration for personalized language generation. We show an AI Ghostwriter Effect : Users do not consider themselves the owners and authors of AI-generated text but refrain from publicly declaring AI authorship. Personalization of AI-generated texts did not impact the AI Ghostwriter Effect , and higher levels of participants’ influence on texts increased their sense of ownership. Participants were more likely to attribute ownership to supposedly human ghostwriters than AI ghostwriters, resulting in a higher ownership-authorship discrepancy for human ghostwriters. Rationalizations for authorship in AI ghostwriters and human ghostwriters were similar. We discuss how our findings relate to psychological ownership and human-AI interaction to lay the foundations for adapting authorship frameworks and user interfaces in AI in text-generation tasks.

Overview

CoChat Summary

TL;DR

In two studies (n=30, n=96), the paper identifies an “AI Ghostwriter Effect”: users often do not feel ownership of AI-generated text yet still self-declare as authors and omit AI credit. Personalization did not affect this effect, while greater perceived control/leadership increased ownership, and AI ghostwriters were credited less than a human ghostwriter.

Key findings

  • AI Ghostwriter Effect: In Study 1, participants attributed ownership to the AI for AI-supported postcards, but fewer than one-fifth declared AI as author; only 6–7 participants per AI Interaction Method mentioned AI (n=30). Study 2 replicated this: AI was mentioned as author in 71/96 cases (73.9%) versus 79/96 (82.3%) for a human ghostwriter, significantly less often, χ²(1)=31.92, p<.001.
  • Personalization did not affect ownership or authorship: Study 1 found no Personalization effect on ownership, F(1,28)=0.52, p=.477, η²p=.02; the Bayesian model assuming no difference was 2.38 times more likely. True personalization texts had higher cosine similarity (M=.77, SD=.04) than placebo-personalized texts (M=.72, SD=.06), t(25.72)=-2.89, p=.008, d=-1.06.
  • Greater influence/control increased ownership: Interaction Method affected ownership, F(2.02,56.58)=55.75, p<.001, η²p=.67; sense of control predicted ownership, β=0.21, 95% CI [0.05,0.37], t(109)=2.57, p=.012; sense of leadership also predicted ownership, β=0.21, 95% CI [0.04,0.37], t(109)=2.46, p=.015.
  • Human vs. AI ghostwriter: ownership was lower for the human ghostwriter than the AI ghostwriter (Welch t(95)=-2.05, p=.043, d=-0.21; Wilcoxon W=1095, p=.018, r=.10); the link between ownership and declaring authorship was weak for AI (r_s=.24, p=.017) and human ghostwriters (r_s=.26, p=.011).

Methods

  • Study 1: n=30 final; 4 Interaction Methods (Writing, Editing, Choosing, Getting) × 2 personalization (true fine-tuned GPT-3 Davinci vs. placebo-personalization), postcard writing/upload task; measures included visual analog scales for ownership/control/leadership, free-text author declaration, and AttrakDiff; analysis used mixed ANOVA, linear mixed models, Bayes factors, and Cochran’s Q.
  • Study 2: pre-registered, n=96 final; Wizard-of-Oz placebo personalization; one non-personalized AI-generated postcard presented as written by either an AI or a human ghostwriter named “Sasha”; author attribution chosen from a predefined list; analysis used paired t-tests/Wilcoxon tests, McNemar χ², and Spearman correlations.
  • Total across both studies: 126 participants.

Limitations

  • Stated: the postcard scenario may be artificial/simple, postcards were not sent and did not describe an actual trip; personalization was only applied to half of Study 1 and none of Study 2, with fewer fine-tuning prompt-completion pairs than OpenAI recommends; the data are purely quantitative; the focus on (placebo-)personalized AI may bias results toward self-authorship; the studies cannot prescribe how people should declare authorship.
  • Stated: author declaration differed strongly by interface—free-text in Study 1 yielded about 20% AI mention, while the dropdown in Study 2 yielded about 70%—so interface design may affect disclosure.
  • Evident from the text: Study 1’s small n=30 limits precise estimation of relative frequencies, and its within-subject design with all Interaction Methods may have caused carryover, which the paper notes and addresses in Study 2.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1145/3637875

Cites

119 of 119 references verified0 citation errors · 0 not checkable

Statistics

11 tests recomputed · 0 with the reported and recomputed p on opposite sides of .05

Verification

Claims the paper makes

7 Valid · 0 Not valid · 7 Not checked · 102 Own results and interpretations

58 more claims

Create a free account to see all 116 claims and every source verdict.

See all claims — free account
  • Probably not a substantial claim. 24
  • It may state two separate claims as one. 31
  • The quote only partly states this claim. 3

Create a free account to see these claims.

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

9 more claims about this paper

Create a free account to see every claim other papers make about this one.

See all claims — free account

Statistics that don’t add up

Each reported p-value recomputed from its test statistic and degrees of freedom.

1 of 1

Statistics recomputed

Every reported test result, recomputed from its test statistic and degrees of freedom. A result shows as not adding up only once an independent check has confirmed it.

  • Doesn’t add upF(1, 101.55) = 6.47reported p = .02 · recomputed p = .012

    “F (1, 101.55) = 6.47, p = .02”

  • Consistentt(25.72) = -2.89reported p = .008 · recomputed p = .008

    “t (25.72) = -2.89,p = .008”

  • ConsistentQ(3) = 18.54reported p < .001 · recomputed p < .001

    “Q (3) = 18.54, p < .001”

  • ConsistentF(3, 86.02) = 28reported p < .001 · recomputed p < .001

    “F (3, 86.02) = 28.00,p < .001”

  • Consistentt(109) = 2.57reported p = .012 · recomputed p = .012

    “t (109) = 2.57, p = .012”

5 of 11 · the rest aren’t loaded

6 more recomputed tests

Create a free account to see every test the Stats check recomputed.

See all tests — free account

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.

1 of 1

Citations

Cites

Show all 119 references

Cited by

Metadata

Authors
Draxler, F., Werner, A., Lehmann, F., Hoppe, M., Schmidt, A., Buschek, D., & Welsch, R.
Affiliations
  • Ludwig-Maximilians-Universität München (DE)
  • University of Bayreuth (DE)
  • Aalto University (FI)
Venue
ACM Transactions on Computer-Human Interaction
Published
2024-02-05
Volume
31
Issue
2
Pages
1-40
ISSN
  • 1073-0516 (ISSN-L)
  • 1557-7325
OpenAlex
W4389913593
Cited by
186
References
116
Citations by year
2026: 105 · 2025: 61 · 2024: 19
FWCI
54.89
Type
journal article
Language
en
Open access
Locations
  • ACM Transactions on Computer-Human Interaction (publishedVersion, cc-by)
  • arXiv (Cornell University) (submittedVersion, cc-by-sa)
  • Aaltodoc (Aalto University) (submittedVersion, cc-by)
  • arXiv (Cornell University)
Funders
  • Bavarian State Ministry of Science and the Arts with the Bavarian Research Institute for Digital Transformation
  • HumanE AI Network from the European Union’s Horizon 2020 (952026)
Topics
  • Social Sciences › Psychology › Social Psychology › Social Robot Interaction and HRI
  • Physical Sciences › Computer Science › Artificial Intelligence › Authorship Attribution and Profiling
  • Physical Sciences › Computer Science › Artificial Intelligence › Ethics and Social Impacts of AI
Keywords
authorship, ownership attribution, AI-generated text, human-AI collaboration, psychological ownership, human-AI interaction

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 2026-10-12, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-12 (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

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.