Canonical text (the exact bytes fingerprinted)

TRUVACE RECORD VERSION
record: TRV-2026-0692
version: 1
kind: certified
reason: Certified as a peer-reviewed Trace on AI-assisted writing and authenticity.
timestamp: 2026-08-08T09:25:02.413591Z
status: published
lens: trace
sector: entertainment
headline: “It Was 80% Me, 20% AI”: Seeking Authenticity in Co-Writing with Large Language Models
dek: 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.
gain_title: Personalized GPT-4 writing suggestions often aligned more closely with participating writers’ styles and helped some writers develop ideas, maintain their writing flow and reduce the work required to revise mismatched suggestions.
problem_title: Participating writers worried that AI assistance could weaken creative control, interrupt their established process or reproduce a fixed version of their style rather than support continued experimentation and growth.
trace_subject: Professional writers and readers judged authenticity in AI-assisted writing differently
gain_reading: Personalized GPT-4 writing suggestions often aligned more closely with participating writers’ styles and helped some writers develop ideas, maintain their writing flow and reduce the work required to revise mismatched suggestions.
gain_evidence: AI as a driver of the writing flow: Having the option to continuously request assistance from AI may help carry on writers’ writing flow and remove writing blocks. | Occasionally, AI suggestions might point to novel directions or ideas that writers would not have conceived of by themselves, and the tool might also offer “jumping points” that allow writers to transition from one idea to another.
problem_reading: Participating writers worried that AI assistance could weaken creative control, interrupt their established process or reproduce a fixed version of their style rather than support continued experimentation and growth.
problem_evidence: On the negative end, participants worried personalization might also lead to writers adopting more suggestions from AI, allowing more influences from the tool. | In our study, professional writers still desired to conduct most of their writing work. | Instead, designers and developers of these tools could target the growth of writers as their ultimate design goal.
quick_read: A mixed-methods study examined how 19 professional writers and 30 avid readers understood authenticity in writing produced with AI assistance. Writers completed short writing tasks using both personalized and non-personalized GPT-4 suggestions, while readers evaluated passages written independently or with either form of AI support.

Writers associated authenticity with personal experience, creative intention, control and the process of developing a piece. Most reacted positively to personalized assistance, but they warned that reproducing an existing writing style could restrict experimentation and creative growth. Readers were less concerned about AI involvement and did not reliably distinguish the short AI-assisted passages from passages written independently.

The result matters because judging only the finished text may miss how AI changes a writer’s sense of authorship and creative control. However, the small exploratory sample, short passages and 2023 data collection limit how broadly the findings can be applied.
limitation: The study does not establish that readers generally cannot detect AI-generated writing or that AI has no effect on writing quality. It examined short, human-directed passages with 19 professional writers and 30 readers rather than complete works generated autonomously by AI.

Participants were recruited through Upwork and online reader communities, and the data was collected from June through October 2023. The findings may not generalize to journalism, academic writing, education or current AI systems and public attitudes. The solo-written passages also did not use the same time constraints as the AI-assisted writing sessions.
tag: Automated dual reading
key_points: Researchers conducted semi-structured interviews with 19 professional writers who used personalized and non-personalized GPT-4 writing assistance in randomized order. | A separate group of 30 avid readers evaluated solo-written, personalized-AI-assisted and non-personalized-AI-assisted passages without initially knowing how each passage was produced. | Writers defined authenticity largely through personal experience, creative intention, control and active participation in the writing process. | Most writers preferred personalized assistance, but some worried that imitation of their prior style could reinforce old habits or restrict their development. | Readers did not reliably distinguish the short AI-assisted passages from independently written passages and reported no statistically significant differences in likeability, enjoyment or creativity across the three conditions.
rundown: Researchers recruited professional writers through Upwork and asked them to complete two short writing sessions using a GPT-4-powered interface. One version generated suggestions using a sample of the participant’s prior writing, while the other provided non-personalized assistance. The order was randomized and participants were not initially told which version was personalized.

Interviews showed that writers did not define authenticity solely by whether the finished text sounded like them. They also considered whether the work reflected their experiences and intentions, whether they retained meaningful control and whether they remained actively responsible for creative decisions. Writers often viewed AI as more acceptable when it supported an existing direction than when it generated a piece from beginning to end.

Thirty avid readers then evaluated passages written independently and with both forms of AI assistance. Readers did not report significant differences in likeability, enjoyment or creativity across the conditions. They also did not reliably identify which passages involved AI. After the production methods were disclosed, readers generally continued to treat the writer as the principal author and viewed AI more as a tool than an equal collaborator.

Most writers reacted favorably to personalization, but stylistic similarity did not resolve every authenticity concern. Some warned that an AI system based on their previous writing could preserve an outdated version of their voice, reinforce existing habits and reduce opportunities for creative development.
sources:
- peer_reviewed | 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, Vol. 9, No. 2, Article CSCW122. | https://doi.org/10.1145/3711020 | 2025-04-01
prev: 0000000000000000000000000000000000000000000000000000000000000000

sha256

87d0d5c56e6bbe72d0526c321ac13cf57508cbe1608543799cce7a1f0f094fbd

previous

0000000000000000000000000000000000000000000000000000000000000000