Student or educator? Sign up with your school email and get 6 months of Pro free.
CoChat

The DOI matches its Crossref record; 61 of the 61 references it cites were found in a registry or an index; 17 of the 61 sources its claims cite were checked for support.

'It was 80% me, 20% AI': Seeking Authenticity in Co-Writing with Large Language Models

Open access
Hwang, Angel Hsing-Chi, Liao, Q. Vera, Blodgett, Su Lin, Olteanu, Alexandra, Trischler, Adam · 2025 · Proceedings of the ACM on Human-Computer Interaction

Abstract

Given the rising proliferation and diversity of AI writing assistance tools, especially those powered by large language models (LLMs), both writers and readers may have concerns about the impact of these tools on the authenticity of writing work. We examine whether and how writers want to preserve their authentic voice when co-writing with AI tools and whether personalization of AI writing support could help achieve this goal. We conducted semi-structured interviews with 19 professional writers, during which they co-wrote with both personalized and non-personalized AI writing-support tools. We supplemented writers' perspectives with opinions from 30 avid readers about the written work co-produced with AI collected through an online survey. Our findings illuminate conceptions of authenticity in human-AI co-creation, which focus more on the process and experience of constructing creators' authentic selves. While writers reacted positively to personalized AI writing tools, they believed the form of personalization needs to target writers' growth and go beyond the phase of text production. Overall, readers' responses showed less concern about human-AI co-writing. Readers could not distinguish AI-assisted work, personalized or not, from writers' solo-written work and showed positive attitudes toward writers experimenting with new technology for creative writing.

Overview

CoChat Summary

TL;DR

This study examines how professional writers and avid readers conceptualize authenticity in human–AI co-writing, and whether personalizing AI writing assistance helps preserve writers' authentic voices. Writers' authenticity concerns centered on internal experience and process rather than just output, and while writers preferred personalized AI, readers could not reliably distinguish AI-assisted from solo-written work and reacted positively to it.

Key findings

  • Writers conceptualized authenticity through three themes: source authenticity (who produced the content), authentic self (internal experiences, identity, and lived experience grounding the work), and content authenticity (whether the output resembles the writer's signature style); the "authentic self" dimension extended beyond prior frameworks of category, source, and value.
  • Writers subjectively preferred personalized AI (personalized via in-context learning from their own writing samples) and felt it better preserved their voice, but behavioral logs showed no significant difference in requesting AI assistance (β = 0.13, SE = 1.08, t = 0.12, p = 0.904) or accepting suggestions (β = -0.02, SE = 0.07, t = -0.27, p = 0.790) between personalized and non-personalized tools.
  • Readers rated solo human work as more likely written by a human than either AI-assisted version (solo vs. personalized: β = -0.73, SE = 0.25, t = -2.94, p = 0.004; solo vs. non-personalized: β = -0.64, SE = 0.24, t = -2.57, p = 0.011), but showed no significant differences in enjoyment (F = 0.63, p = 0.536), likeability (F = 1.23, p = 0.298), or creativity (F = 1.19, p = 0.311) across the three conditions, and could not precisely identify AI-generated portions.
  • Readers expressed significantly positive interest in reading AI-assisted writing (Z = 5.82, p < 0.001), viewed it as authentic writing (Z = 4.21, p < 0.001), and rated the writing (Z = 3.70, p < 0.001) and the writer (Z = 4.58, p < 0.001) positively after learning of AI involvement; no significant trend on whether humans or AI deserved more credit (β = 0.18, SE = 0.19, t = 0.97, p = 0.339).

Methods

  • Part 1: semi-structured interviews with 19 professional writers (recruited via Upwork) across genres; each completed a pre-interview survey, then two ~20–25 minute co-writing sessions (~200 words each) using the CoAuthor interface with GPT-4, one personalized via in-context learning from the writer's own sample and one non-personalized, in randomized order, followed by reflection interviews.
  • Part 2: online survey with 30 avid readers (recruited from Reddit, screened for AI-generated responses), who read three passages from six Part 1 writers (solo, personalized-AI, non-personalized-AI) in randomized order and rated likeability, enjoyment, creativity, likelihood of human writing, voice preservation, and credit/authorship.
  • Analysis: thematic coding and affinity diagrams for qualitative data; linear mixed-effects models (controlling for participant ID and order) for within-subject comparisons; Wilcoxon one-sample tests against the scale midpoint for single measures.

Limitations

  • Stated: data collection occurred June–October 2023, a period of rapid change in AI tools and public sentiment, so opinions may shift; responses may be affected by 2023 media coverage; expert writers may focus more on ideation than structure/composition; Reddit readers may have distinctive traits (e.g., high Openness); the study examined only passive, leisurely reading rather than readers seeking a specific writer; solo-written pieces were produced without time constraints; other stakeholders (e.g., clients) and monetization impacts were not captured.
  • Evident from the text: the writer sample (N = 19) and reader sample (N = 30) are small and non-representative, and the reader sample required screening out 7 likely AI-generated responses, which the authors note prompted a second data-collection round.

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

Integrity signals

Registry

Crossref · 10.1145/3711020

Cites

61 of 61 references verified0 citation errors · 0 not checkable

Statistics

Verification

Claims the paper makes

13 Valid · 0 Not valid · 13 Not checked · 190 Own results and interpretations

87 more claims

Create a free account to see all 216 claims and to verify the 6 references still open.

See all claims — free account
  • It may state two separate claims as one. 54
  • Probably not a substantial claim. 87
  • The quote only partly states this claim. 7

Create a free account to see these claims.

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.

  • Consistentz = 5.82reported p < .001 · recomputed p < .001

    “(Z = 5.82, p < .001”

  • Consistentz = 4.21reported p < .001 · recomputed p < .001

    “(Z = 4.21, p < .001”

  • Consistentz = -0.3reported p = .763 · recomputed p = .764

    “(Z = -0.30,p = .763”

  • Consistentz = 3.7reported p < .001 · recomputed p < .001

    “(Z = 3.70, p < .001”

  • Consistentz = 4.58reported p < .001 · recomputed p < .001

    “(Z = 4.58, p < .001”

5 of 5

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.

3 of 3

Citations

Cites

Show all 52 references

Research works printed but not linked: 9

  1. 10.1007/978--94-007--7844--3_4 · doi:10.1007/978--94-007--7844--3_4
  2. 10.1111/j.1747--9991.2008 · doi:10.1111/j.1747--9991.2008
  3. 10.1038/s41598-023--30938--9 · doi:10.1038/s41598-023--30938--9
  4. 10.1145/3544548.3581196 · DOI · Paper page
  5. 10.1145/3290605.3300469 · DOI · Paper page
  6. 10.1145/3491102.3517731 · DOI · Paper page
  7. Michigan Journal of Economics. 2024. Is AI taking over the job market? https://sites.lsa.umich.edu/mje/2024/01/03/isai- taking-over-the-job-market/
  8. 10.1145/3544549.3585893 · DOI · Paper page
  9. 10.1007/978--3-030--83527--9_1 · doi:10.1007/978--3-030--83527--9_1

Cited by

See all 42

Metadata

Authors
Hwang, A. H.-C., Liao, Q. V., Blodgett, S. L., Olteanu, A., & Trischler, A.
Affiliations
  • University of Southern California (US)
  • Microsoft Research Montréal (Canada) (CA)
  • Independent Researcher, Toronto, ON, Canada
Venue
Proceedings of the ACM on Human-Computer Interaction
Published
2025-05-02
Volume
9
Issue
2
Pages
1-41
ISSN
2573-0142 (ISSN-L)
OpenAlex
W4410049808
Cited by
42
References
52
Citations by year
2026: 35 · 2025: 7
FWCI
43.69
Type
journal article
Language
en
Open access
Locations
  • Proceedings of the ACM on Human-Computer Interaction (publishedVersion, cc-by)
  • arXiv (Cornell University) (submittedVersion)
  • arXiv (Cornell University)
Topics
  • Physical Sciences › Computer Science › Artificial Intelligence › Ethics and Social Impacts of AI
  • Social Sciences › Arts and Humanities › Visual Arts and Performing Arts › Art, Technology, and Culture
  • Physical Sciences › Computer Science › Human-Computer Interaction › Innovative Human-Technology Interaction
Keywords
large language models, AI-assisted writing, human-AI co-creation, authenticity in writing, professional writers, reader perception, creative writing, authorial self

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: 'It was 80% me, 20% AI': Seeking Authenticity in Co-Writing with Large Language Models. https://app.cochat.ai/papers/10.1145/3711020
The paper
Hwang, A. H.-C., Liao, Q. V., Blodgett, S. L., Olteanu, A., & Trischler, A. (2025). 'It was 80% me, 20% AI': Seeking Authenticity in Co-Writing with Large Language Models. Proceedings of the ACM on Human-Computer Interaction. https://doi.org/10.1145/3711020

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.