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

The DOI matches its Crossref record; 73 of the 73 references it cites were found in a registry or an index; 8 of the 99 sources its claims cite were checked for support.

ABScribe: Rapid Exploration & Organization of Multiple Writing Variations in Human-AI Co-Writing Tasks using Large Language Models

Open access
Reza, Mohi, Laundry, Nathan M, Musabirov, Ilya, Dushniku, Peter, Yu, Zhi Yuan “Michael”, Mittal, Kashish, Grossman, Tovi, Liut, Michael, Kuzminykh, Anastasia, Williams, Joseph Jay · 2024 · Proceedings of the CHI Conference on Human Factors in Computing Systems

Abstract

Exploring alternative ideas by rewriting text is integral to the writing process. State-of-the-art Large Language Models (LLMs) can simplify writing variation generation. However, current interfaces pose challenges for simultaneous consideration of multiple variations: creating new variations without overwriting text can be difficult, and pasting them sequentially can clutter documents, increasing workload and disrupting writers’ flow. To tackle this, we present ABScribe, an interface that supports rapid, yet visually structured, exploration and organization of writing variations in human-AI co-writing tasks. With ABScribe, users can swiftly modify variations using LLM prompts, which are auto-converted into reusable buttons. Variations are stored adjacently within text fields for rapid in-place comparisons using mouse-over interactions on a popup toolbar. Our user study with 12 writers shows that ABScribe significantly reduces task workload (d = 1.20, p < 0.001), enhances user perceptions of the revision process (d = 2.41, p < 0.001) compared to a popular baseline workflow, and provides insights into how writers explore variations using LLMs.

Overview

CoChat Summary

TL;DR

ABScribe is an LLM-based writing interface that stores multiple text variations in Variation Fields with a Popup Toolbar, Variation Sidebar, AI Modifiers, and AI Drafter. In a within-subjects study with 12 writers, it reduced subjective task workload and improved perceptions of the revision process compared with a chat-based baseline.

Key findings

  • ABScribe significantly reduced NASA-TLX subjective task workload versus the baseline (𝑑 = 1.20, 𝑝 < 0.001).
  • ABScribe significantly increased summed Likert agreement on the efficacy of the revision process versus the baseline (𝑑 = 2.41, 𝑝 < 0.001).
  • Qualitative themes: non-linear variation storage reduced clutter and pressure to commit early; the Popup Toolbar supported in-context comparison; AI Modifiers made prompts reusable and encouraged more imperative, generalizable prompt writing.
  • Trade-offs: two participants perceived clutter as unproblematic or beneficial, and some preferred linear side-by-side comparison for large segments, though the Variation Sidebar was seen as mitigating this.

Methods

  • Within-subjects controlled evaluation with 12 writers (5 women, 7 men, aged 18–34), compensated 30 CAD; each completed two guided tasks (LinkedIn post and email to a professor) randomly paired with counterbalanced ABScribe versus baseline.
  • Participants used AI to explore eight variations of three text segments (24 variations total) per task; the baseline was a GPT-4 rich-text editor with a chat-based assistant and direct AI text insertion.
  • Measures included weighted NASA-TLX and 7-point Likert scales; analysis used pairwise one-sided t-tests and reflexive thematic analysis of interview transcripts and task observation notes.

Limitations

  • Stated: the baseline was constructed rather than an existing system, which may have affected ecological validity; future studies could compare ABScribe with real-world AI-assisted writing workflows.
  • Stated: external validity is limited to English writers and a single 1.5-hour session with two guided tasks; AI Drafter and AI Modifiers carry hallucination risk; no logging metrics were collected.
  • Evident: the small sample (N = 12) and lab-based tasks limit generalizability, and the reported trade-offs from two participants suggest individual differences in preference for clutter and linear editing.

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

Integrity signals

Registry

Crossref · 10.1145/3613904.3641899

Cites

73 of 73 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Figures

Figure 1 The ABScribe Interface: (1) Variation Fields: Multiple variations are stored within text fields that do not break the flow of the draft. (2) Popup Toolbar: Users can swiftly choose and compare multiple variations by hovering over a dynamically placed toolbar located above the selected Variation Field, or clone and edit variations in-place. (3) Variation Sidebar: Users can navigate through multiple variations organized as an accordion-based interface. (4) AI Modifiers: Users can create variations using AI by typing modification instructions auto-converted into reusable buttons that can be applied to other Variation Fields. (5) AI Drafter: Users can draft text from GPT-4 directly into the document by typing '@ai <prompt>' and pressing enter.
Figure 1 The ABScribe Interface: (1) Variation Fields: Multiple variations are stored within text fields that do not break the flow of the draft. (2) Popup Toolbar: Users can swiftly choose and compare multiple variations by hovering over a dynamically placed toolbar located above the selected Variation Field, or clone and edit variations in-place. (3) Variation Sidebar: Users can navigate through multiple variations organized as an accordion-based interface. (4) AI Modifiers: Users can create variations using AI by typing modification instructions auto-converted into reusable buttons that can be applied to other Variation Fields. (5) AI Drafter: Users can draft text from GPT-4 directly into the document by typing '@ai <prompt>' and pressing enter.CC-BY · From the PDF
Image 2
CC-BY · From the PDF
Image 3
CC-BY · From the PDF
Figure 4 AI Modifiers: Variation Fields can also be edited using the AI Modifiers, which let users specify custom alterations. Descriptive labels are automatically generated for each AI Modifier to turn them into reusable buttons.
Figure 4 AI Modifiers: Variation Fields can also be edited using the AI Modifiers, which let users specify custom alterations. Descriptive labels are automatically generated for each AI Modifier to turn them into reusable buttons.CC-BY · From the PDF
Figure 5 AI Drafter: Users can draft LLM-generated text directly into the document, providing tighter integration between the Human and AI generated writing workflow. Users can see the AI generated content in real-time and choose to insert or delete the output, or revise the prompt, giving them control over what is included in their document.
Figure 5 AI Drafter: Users can draft LLM-generated text directly into the document, providing tighter integration between the Human and AI generated writing workflow. Users can see the AI generated content in real-time and choose to insert or delete the output, or revise the prompt, giving them control over what is included in their document.CC-BY · From the PDF
Figure 6 Overall results on NASA TLX subjective task workload
Figure 6 Overall results on NASA TLX subjective task workloadCC-BY · From the PDF
Figure 7 Summed Likert ratings for users' perceptions of the revision process
Figure 7 Summed Likert ratings for users' perceptions of the revision processCC-BY · From the PDF
Image 8
CC-BY · From the PDF
Figure 9 The Baseline Interface: (1) We retained the ability to insert text directly into the document using AI Drafter as some modern AI editors have similar capabilities. (2) The chat-based interface on the left used the same underlying model (GPT-4) as ABScribe. All tangential differences such as font size and rich-text editing capabilities were consistent with ABScribe.
Figure 9 The Baseline Interface: (1) We retained the ability to insert text directly into the document using AI Drafter as some modern AI editors have similar capabilities. (2) The chat-based interface on the left used the same underlying model (GPT-4) as ABScribe. All tangential differences such as font size and rich-text editing capabilities were consistent with ABScribe.CC-BY · From the PDF

Verification

Claims the paper makes

5 Valid · 3 Not valid · 43 Not checked · 57 Own results and interpretations

35 more claims

Create a free account to see all 108 claims and to verify the 91 references still open.

See all claims — free account
  • Probably not a substantial claim. 60
  • It may state two separate claims as one. 21

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?

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 67 references

Research works printed but not linked: 20

  1. 10.7551/mitpress · doi:10.7551/mitpress
  2. Virginia Braun and Victoria Clarke. 2019. Reflecting on reflexive thematic analysis. Qualitative research in sport, exercise and health 11, 4 (2019), 589–597.
  3. Bill Buxton. 2010. Sketching user experiences: getting the design right and the right design. Morgan kaufmann, Boston.
  4. John Joon Young Chung, Wooseok Kim, Kang Min Yoo, Hwaran Lee, Eytan Adar, and Minsuk Chang. 2022. TaleBrush: Sketching stories with generative pretrained language models. In Proceedings of the 2022 CHI Conference on Human Factors in Computing Systems. Association for Computing Machinery, New York, NY, USA, 1–19.
  5. Robert Dale. 2021. GPT-3: What’s it good for?Natural Language Engineering 27, 1 (2021), 113–118.
  6. 10.18653/v1 · doi:10.18653/v1
  7. Jill Fitzgerald. 1987. Research on revision in writing. Review of educational research 57, 4 (1987), 481–506.
  8. 10.58680/ccc198115885 · DOI · Paper page
  9. Han L Han. 2020. Designing Representations for Digital Documents. In Adjunct Proceedings of the 33rd Annual ACM Symposium on User Interface Software and Technology. Association for Computing Machinery, New York, NY, USA, 174–178.
  10. 10.1145/3313831.3376804 · DOI
  11. Iikka Hauhio, Anna Kantosalo, Simo Linkola, and Hannu Toivonen. 2023. The Spectrum of Unpredictability and its Relation to Creative Autonomy. In Proceedings of the 14th International Conference on Computational Creativity. Association for Computational Creativity (ACC), Spain, 148–152. https://computationalcreativity.net/iccc23/ International Conference on Computational Creativity, ICCC ; Conference date: 19-06-2023 Through 23-06-2023.
  12. Ernest Hemingway. 2014. Moveable feast: the restored edition. Simon and Schuster, New York, NY, USA.
  13. David G Jansson and Steven M Smith. 1991. Design fixation. Design studies 12, 1 (1991), 3–11.
  14. Charles A MacArthur. 2018. Evaluation and revision. Best practices in writing instruction 287 (2018), 23 pages.
  15. Thomas Mahatody, Mouldi Sagar, and Christophe Kolski. 2010. State of the art on the cognitive walkthrough method, its variants and evolutions. Intl. Journal of Human–Computer Interaction 26, 8 (2010), 741–785.
  16. James P. Purdy. 2014. What Can Design Thinking Offer Writing Studies?College Composition and Communication 65, 4 (2014), 612–641. http://www.jstor.org/stable/43490875
  17. Horst Rittel. 1967. Wicked problems. Management Science,(December 1967) 4, 14 (1967), B141–B142.
  18. Anthony Seow. 2002. The writing process and process writing. Methodology in language teaching: An anthology of current practice 315 (2002), 320.
  19. Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Ł ukasz Kaiser, and Illia Polosukhin. 2017. Attention is All you Need. In Advances in Neural Information Processing Systems, I. Guyon, U. Von Luxburg, S. Bengio, H. Wallach, R. Fergus, S. Vishwanathan, and R. Garnett (Eds.). Vol. 30. Curran Associates, Inc., Long Beach, CA. https://proceedings.neurips.cc/paper_files/paper/2017/file/3f5ee243547dee91fbd053c1c4a845aa-Paper.pdf
  20. William Knowlton Zinsser. 2001. On writing well: The classic guide to writing nonfiction. Quill/A Harper Collins Books, 195 Broadway New York, NY, USA.

Cited by

See all 101

Metadata

Authors
Reza, M., Laundry, N. M., Musabirov, I., Dushniku, P., Yu, Z. Y. “., Mittal, K., Grossman, T., Liut, M., Kuzminykh, A., & Williams, J. J.
Affiliations
  • University of Toronto (CA)
  • Computer Science, University of Toronto, Canada
Venue
Proceedings of the CHI Conference on Human Factors in Computing Systems
Publisher
ACM
Published
2024-05-11
Pages
1-18
OpenAlex
W4396832282
Cited by
52
References
67
Citations by year
2026: 16 · 2025: 34 · 2024: 2
FWCI
16.76
Type
conference paper
Language
en
Open access
Locations
  • CHI Conference on Human Factors in Computing Systems (publishedVersion)
  • arXiv (Cornell University) (submittedVersion, cc-by)
  • arXiv (Cornell University) (cc-by)
Funders
  • NSF (National Science Foundation) (2209819)
  • Natural Sciences and Engineering Research Council of Canada (RGPIN-2019-06968)
  • Office of Naval Research (N00014-21-1-2576)
Topics
  • Physical Sciences › Computer Science › Human-Computer Interaction › Usability and User Interface Design
  • Physical Sciences › Computer Science › Human-Computer Interaction › Innovative Human-Technology Interaction
  • Physical Sciences › Computer Science › Computer Graphics and Computer-Aided Design › Data Visualization and Analytics
Keywords
rewriting, large language models, textual revision, LLM prompts, text structure

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 2026-10-04, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-04 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Cited sources are checked with a language model’s help (qwen3.8-omni-flash).

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: ABScribe: Rapid Exploration & Organization of Multiple Writing Variations in Human-AI Co-Writing Tasks using Large Language Models. https://app.cochat.ai/papers/10.1145/3613904.3641899
The paper
Reza, M., Laundry, N. M., Musabirov, I., Dushniku, P., Yu, Z. Y. “., Mittal, K., Grossman, T., Liut, M., Kuzminykh, A., & Williams, J. J. (2024). ABScribe: Rapid Exploration & Organization of Multiple Writing Variations in Human-AI Co-Writing Tasks using Large Language Models. Proceedings of the CHI Conference on Human Factors in Computing Systems. https://doi.org/10.1145/3613904.3641899

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.