AI disclosure for ACM FAccT submissions

FAccT authors need a clear AI usage statement

FAccT papers sit in a strange place.

Many submissions study automated systems, data harms, audits, policy, platforms, labor, fairness, accountability, or public impact. That means AI use in the research process can affect both the paper and the thing the paper studies.

For the 2027 cycle, the ACM FAccT Author Guide says authors are expected to include a "Generative AI usage statement" in the Endmatter section. The statement should say whether and how the authors used generative AI in writing the publication. The guide also points authors to its section on authorship and AI writing tools. (facctconference.org)

That creates a practical question.

What do you actually write?

This guide gives you a working answer. It pairs the FAccT statement with a fuller AI Usage Card, so your paper gives readers a short disclosure and your project files keep the detail you may need later.

If you want the short path, generate an AI Usage Card before submission. Then copy a concise version into your FAccT Endmatter and save the card with your reproducibility materials, internal records, or supplementary files.

Separate writing help from research help

ACM draws a line between AI used for writing and AI used for research.

The ACM FAQ says authors may use generative AI tools to prepare a manuscript, but the work must represent the authors' own research and intellectual contribution. Authors also accept responsibility for the truth and correctness of the whole submission, including material produced with AI tools. (acm.org)

The same FAQ says ACM no longer requires disclosure when authors use AI only to assist with writing an ACM submission. But FAccT asks authors to include a generative AI usage statement in Endmatter for the 2027 cycle, so FAccT authors should follow the venue instruction. (acm.org) (facctconference.org)

Use this rule of thumb.

If the tool helped you express your work, describe that in the FAccT Endmatter.

If the tool helped produce evidence, code, data, analysis, figures, labels, measures, prompts, simulations, or results, describe that in the methods as well.

This matters for FAccT because many papers use AI systems as study objects, research instruments, or data sources. A disclosure that says "AI was used for editing" does not cover an LLM used to generate personas for a user study or classify posts in a dataset.

For field specific examples, see [[[AI disclosure](/ai-disclosure-in-systematic-reviews-and-meta-analyses/)](/how-to-disclose-ai-use-for-neurips-icml-and-acl-submissions/) for social science research](/ai-disclosure-for-social-science-research/) and AI disclosure for NLP research papers. FAccT papers often borrow from both traditions.

Write the Endmatter statement for readers, not lawyers

A weak statement sounds like this:

"Generative AI tools were used during preparation of this manuscript."

That sentence tells the reader almost nothing.

A useful FAccT statement names the tool, the task, the part of the paper affected, and the human check. Keep it short. Readers need enough detail to judge scope, not a diary of every prompt.

A better statement sounds like this:

"The authors used ChatGPT 5 to revise grammar and shorten portions of the introduction and related work. The authors reviewed and edited all suggested text and take responsibility for the final manuscript."

That works because it answers the reader's first question: did the AI help write the paper, or did it shape the research?

If you used AI in the research process, the Endmatter statement should point readers to the methods section:

"The authors used Claude Sonnet 4.5 to draft Python code for preprocessing interview transcripts and to suggest candidate labels during early qualitative coding. Section 3.4 describes this use. The authors reviewed the code, revised the labels, and performed the final analysis."

That sentence does not replace method detail. It helps the reader find it.

The broader principle matches the guidance in Do I need to disclose AI usage in my paper?: disclose use that could affect how readers understand the work, reproduce the work, or judge author responsibility.

Put research relevant AI use in the methods

ACM's FAQ asks authors to describe AI use in detail in the methods when AI contributes to the research life cycle. ACM lists examples that include study design, data source creation or selection, experiment design, data generation, coding, model implementation, simulations, analysis, testing, validation, software deployment, and archiving data or code for reproducibility. (acm.org)

That list covers many FAccT workflows.

If you used an LLM to generate survey items, say so where you describe instrument design.

If you used a model to classify appeals, posts, moderation decisions, or court records, say so where you describe coding and validation.

If you used AI to create synthetic participants or vignettes, say so where you describe materials.

If you used an AI coding assistant to write analysis scripts, say how you tested the scripts.

FAccT reviewers care about methods. They will read a vague AI statement as a missing method detail, not as a transparency win.

You can adapt the disclosure patterns from AI tools for research if you need a project log before you write the paper. A log helps you remember dates, model versions, prompts, and checks after the submission deadline turns everything blurry.

Do not name an AI tool as an author

ACM does not allow generative AI tools to appear as authors on ACM works. The ACM policy also says authors must be identifiable human beings, must make a substantial intellectual contribution, and must accept accountability for the work. (acm.org) (acm.org)

This part stays simple.

Do not list ChatGPT, Claude, Gemini, Copilot, Midjourney, or another tool in the author line.

Mention the tool in the Endmatter statement, methods section, acknowledgments, or supplementary disclosure record, depending on how you used it.

For a wider discussion of authorship, see Can AI be a co-author on a research paper?. For examples that focus on ChatGPT wording, see How to disclose ChatGPT usage in academic papers.

Use an AI Usage Card as your private record

The FAccT Endmatter statement should stay short. Your lab record can hold more detail.

An AI Usage Card gives you a structured record of tool use. It helps you capture what you used, where you used it, why you used it, and how you checked the output.

That record helps when a reviewer asks for clarification. It also helps when co-authors disagree about what belongs in the paper. Instead of relying on memory, you can open the card and decide what the manuscript needs.

For FAccT, I would keep the card in the same folder as your submission files. If you archive code or materials, include the card there if it does not expose confidential data or break anonymization rules.

You can also use the card to draft the Endmatter text. Generate the card at ai-cards.org, copy the short disclosure text from the page, and adapt it to FAccT's Endmatter wording.

For examples of card language, see [AI Usage Cards examples and templates](/ai-usage-cards-examples/). If you work in LaTeX, pair this guide with the LaTeX tutorial for AI Usage Cards.

LaTeX example for a FAccT Endmatter statement

If your ACM template uses an acknowledgments or Endmatter block, place the statement where the venue asks for it. Keep the language plain.

\begin{acks}
Generative AI usage statement. The authors used ChatGPT 5 to revise grammar
and shorten portions of the introduction and related work. The authors reviewed
and edited all suggested text and take responsibility for the final manuscript.
 
An AI Usage Card documenting this use was generated with ai-cards.org and stored
with the project files.
\end{acks}

If AI shaped the research, point to the methods section.

\begin{acks}
Generative AI usage statement. The authors used Claude Sonnet 4.5 to draft
Python code for preprocessing public comment data and to suggest candidate
labels during exploratory coding. Section~\ref{sec:ai-use} describes this use,
including validation checks. The authors reviewed the code, revised the labels,
and performed the final analysis.
 
An AI Usage Card documenting this use was generated with ai-cards.org and stored
with the replication package.
\end{acks}

Then add a methods subsection.

\subsection{Use of generative AI in preprocessing}
\label{sec:ai-use}
 
We used Claude Sonnet 4.5 to draft Python scripts for removing duplicate
records, normalizing date fields, and flagging empty comments. Two authors
reviewed the scripts before running them. We tested the scripts on a random
sample of 200 records and compared the output with manual checks. We did not
use the tool to code the final dataset or draw conclusions.

The exact wording will change by project. The structure should not.

Say what the tool did. Say what humans checked. Say what the tool did not do if that boundary could confuse readers.

Examples for common FAccT use cases

For language editing only:

"The authors used Grammarly and ChatGPT 5 to identify grammar issues and shorten selected paragraphs. The authors accepted, rejected, and revised suggestions manually. The tools did not generate research questions, data, analysis, citations, or findings."

For literature search support:

"The authors used Elicit to identify candidate papers during early literature search. The authors verified all included sources through publisher pages and citation databases. The tool did not decide inclusion or exclusion."

For qualitative coding support:

"The authors used Claude Sonnet 4.5 to suggest provisional codes for 50 pilot excerpts. Two authors revised the codebook and coded the final dataset. The methods section describes coder training, disagreement resolution, and validation."

For synthetic material generation:

"The authors used GPT-5 to draft initial versions of four fictional moderation scenarios used in the survey. The authors edited the scenarios for study aims and checked them for demographic stereotypes before deployment. The final survey instrument appears in the appendix."

For code assistance:

"The authors used GitHub Copilot to draft portions of the analysis scripts. The authors reviewed all code, added unit tests for preprocessing functions, and reran the analysis from raw data."

These examples share one habit: they do not make the tool sound like a co-researcher. Humans chose, checked, revised, and remain accountable.

Watch anonymization

FAccT uses anonymous review, and many ACM venues have strict rules about revealing author identity during review. If your AI Usage Card names your lab, institution, grant, private dataset, or internal repository, do not include that full card in an anonymous submission unless the venue allows it.

Use two versions if needed.

Keep a full internal card for your records. Create a submission safe version that removes author identifying details. The Endmatter statement can stay general during review and become more specific in the camera ready version if the venue permits or requests it.

This issue also comes up in conference planning. See AI conference deadlines and disclosure planning for a workflow that separates draft records, anonymous submission text, and final publication text.

A short checklist before submission

Before you submit to FAccT, ask these questions.

Did you include a generative AI usage statement in Endmatter?

Did you describe research relevant AI use in the methods?

Did you name the tool and task?

Did you explain the human check?

Did you avoid listing an AI tool as an author?

Did you save an AI Usage Card with your project files?

If you answer yes, your disclosure will give reviewers what they need without turning the paper into a policy memo.

FAccT readers do not need theatrics. They need a clean account of how the work came together.

Generate an AI Usage Card at ai-cards.org, copy the short disclosure into your FAccT Endmatter, and keep the full card with your research records.