AI Disclosure for HCI and CHI Submissions

HCI papers make AI disclosure messy fast

Human-computer interaction research rarely uses AI in one clean place.

You may use an LLM to polish the introduction. You may use it again to draft study materials, simulate user tasks, generate interface copy, cluster interview codes, or build the prototype that participants test. Those uses do not carry the same weight.

That is the problem.

A CHI-style paper does not need a vague sentence that says "we used AI tools." It needs a record that tells reviewers and readers where AI entered the work, what the authors checked, and whether the tool shaped the evidence.

If you want a short way to create that record, generate an [[[AI Usage Card](/ai-disclosure-for-social-science-research/)](/chatgpt-disclosure-academic-papers/)](/) before submission. Treat it as a disclosure note for the paper file, the appendix, or your internal project folder.

Start with the ACM baseline, then add HCI detail

Many HCI venues publish through ACM, so ACM's authorship and disclosure rules give HCI authors a useful starting point. ACM updated its Policy on Authorship on May 14, 2026. The policy says it applies across ACM publication venues, including journals, conferences, ICPS conferences, magazines, newsletters, books, and other ACM publication venues. (acm.org)

ACM also states that authors may use generative AI tools to prepare a manuscript if the tools do not plagiarize, misrepresent, or falsify content, and if the authors take responsibility for the whole work. (acm.org)

The harder line concerns research use. ACM says that when authors use AI to conduct research, they must describe the tool use in detail in the methods section. ACM names research design, data source creation or selection, experiments, data collection, coding, model implementation, simulations, analysis, testing, validation, deployment, and archiving as examples. (acm.org)

That matters for HCI because AI often appears inside the study, not just around the manuscript.

If you used ChatGPT to revise a paragraph, the disclosure question may sit in the acknowledgments or may depend on the venue's current instructions. If you used ChatGPT to generate interview probes, create participant tasks, sort open-ended responses, or build the artifact you evaluated, you should describe that in the method.

For a broader view of ACM-oriented disclosure, see our guide to [[AI disclosure](/how-to-disclose-ai-use-for-neurips-icml-and-acl-submissions/) for ACM FAccT submissions](/ai-disclosure-for-acm-facct-submissions/). For field-level computer science guidance, see AI disclosure for computer science research papers.

Separate writing help from research help

Reviewers care about the difference between prose support and evidence production.

Writing help changes the paper's presentation. Research help may change the study itself.

For example, an HCI team might ask an LLM to shorten a related work paragraph. That affects wording. The same team might ask an LLM to generate personas, design tasks, score participant comments, or suggest themes from interview transcripts. Those uses can affect the findings.

Write those differences down while you work. Do not wait until camera-ready week. Nobody remembers the model version, prompt style, and checks six months later.

A basic record should answer four questions:

1. What tool did you use?
2. Where did you use it?
3. What did the tool produce?
4. What did a human author review, change, reject, or verify?

If that feels like extra work, make the record small. A short AI Usage Card can hold the same facts without turning your appendix into a diary.

HCI use cases that need method-level disclosure

Some AI use belongs near the end of the paper. Some belongs in the method.

If AI shaped your materials, data, analysis, artifact, or results, put it where readers can judge the research design.

Study materials

Disclose AI use when you generate or revise consent text, onboarding scripts, interview questions, scenario descriptions, usability tasks, chatbot prompts, or tutorial content that participants see.

Readers need to know who wrote the stimulus. They also need to know whether the tool introduced tone, assumptions, examples, or cultural cues that may affect participant behavior.

A clear method note might say:

"To draft the initial usability tasks, the authors used Claude Sonnet 4 with prompts written by the first author. The authors rewrote the tasks to remove tool-generated examples, checked the reading level, and piloted the final version with two lab members. Participants saw only the revised tasks."

That sentence gives reviewers something to inspect. It names the tool, the purpose, the output, and the human check.

Prototype content

HCI papers often evaluate an artifact. If AI generated part of that artifact, readers need to know.

Maybe you used Midjourney for placeholder images. Maybe an LLM generated interface labels. Maybe a code assistant wrote part of a Figma plugin or web prototype. Maybe your "AI system" relied on a commercial model through an API.

Do not hide that inside a broad system description. Say where the AI content appears and whether participants interacted with it.

If the AI component affects the claims, treat it as part of the research object. Describe prompts, model names, version dates if available, filtering steps, and author edits.

The same logic applies to generated images. If your HCI paper includes AI-generated visual material, pair this guide with how to disclose AI-generated images in academic papers.

Qualitative analysis

AI-assisted qualitative analysis needs careful wording.

If you used an LLM to suggest initial codes, cluster excerpts, write memos, summarize transcripts, or compare themes, say so in the analysis section. Explain what the tool saw. Explain what the human coders did after that.

Do not write "AI was used for thematic analysis" and stop. That sentence raises more questions than it answers.

A better version:

"After human transcription and de-identification, the authors used ChatGPT to suggest preliminary code labels for 20 pilot excerpts. Two authors reviewed the suggestions, discarded labels that did not match the codebook, and coded the full dataset manually in Dovetail. The LLM did not receive participant names or contact details."

That kind of note helps readers assess privacy, bias, and interpretive control.

For a closer look at interviews, coding, and field notes, see AI disclosure for qualitative research.

Data cleaning and logs

HCI teams often collect messy interaction logs. AI can help label errors, normalize event names, detect duplicates, or summarize free-text bug reports.

If those steps affect which data you include, disclose them.

Readers should know whether an AI tool filtered participants, classified sessions, labeled dropouts, or marked behavior as "successful." In many HCI papers, those choices shape the results more than a polished introduction ever could.

Peer review brings a separate duty

Do not treat peer review like ordinary writing.

ACM's FAQ warns that authors remain responsible for problematic content no matter where it came from, and ACM says content integrity problems tied to AI use can lead to rejection before publication or retraction after publication. (acm.org)

Reviewers carry a different burden: confidentiality. If you review a CHI-style paper, do not paste the manuscript, figures, transcripts, or supplementary material into a public AI tool unless the venue allows that workflow and the tool contract protects confidential material.

The safest habit is simple. If the review system or chair instructions do not allow it, do not use it.

For editor and reviewer guidance, read AI disclosure in peer review. If your conference paper used agents to inspect code, run checks, or draft review responses, our page on AI agent use in conference papers may also help.

Where to place the disclosure

Use the paper structure to decide placement.

If AI shaped the research, put the disclosure in the method. If AI helped write or edit text, follow the venue's instruction for acknowledgments, author notes, submission forms, or appendices. If AI generated figures, data, code, or interface assets, describe that near the relevant artifact and include enough detail for readers to understand the role.

ACM's FAQ now draws a distinction between AI used in research and AI used to assist writing. It says AI used in research must appear in the methods section, while ACM no longer requires disclosure for writing assistance alone under that FAQ framing. (acm.org)

Still, a venue may ask for more. A track chair may ask for a declaration in the submission form. A journal may require a separate statement. Your coauthors may also prefer disclosure for writing support, even when the publisher does not require it.

I would rather include a plain sentence than argue later about whether a use was "only writing." HCI papers already ask readers to trust many design choices. A short AI note costs little.

A sample disclosure for a CHI-style paper

Use this as a starting point, not as a template you paste without thought.

\section*{AI Use Disclosure}
 
The authors used OpenAI ChatGPT-5 on 2026-08-14 to draft alternative wording for two usability task descriptions. The authors rewrote the final task text, checked it against the study protocol, and piloted it with two researchers who did not join the analysis.
 
The authors also used GitHub Copilot in Visual Studio Code while implementing the study prototype. Copilot suggested React component code for non-experimental interface elements, including layout containers and form validation. The authors reviewed, edited, and tested all generated code before deployment.
 
No AI tool analyzed participant transcripts, assigned codes, excluded participants, or generated statistical results.

This example does three useful things.

It names the tools. It separates participant-facing material from prototype code. It also says what the team did not use AI for. That last sentence helps because HCI reviewers may wonder whether AI touched transcripts or outcomes.

If you want to include the same information in an appendix, you can add an AI Usage Card as a short table. See AI Usage Cards examples and templates and the LaTeX tutorial for AI Usage Cards for formats that fit conference papers.

A shorter acknowledgment version

Some papers need a compact disclosure. Use this when AI helped with writing or light materials work, and when the venue allows acknowledgment-style disclosure.

\section*{Acknowledgments}
 
The authors used DeepL Write and ChatGPT-5 to revise sentence-level wording in the manuscript. The authors reviewed all changes and take responsibility for the final text. The tools did not generate research questions, participant data, analysis code, figures, or results.

This wording avoids drama. It tells the reader what happened.

If you write in Overleaf, you can keep the disclosure with your source files and export it with the submission PDF. See how to use AI Usage Cards in Overleaf.

What to record before submission

For HCI projects, record AI use as soon as the tool touches the work.

Write down the tool name, model if visible, date, task, inputs, outputs kept, outputs rejected, and human checks. Add a note about sensitive data. If the tool saw participant information, explain what information and why your ethics approval allowed that use.

Most teams can track this in a shared document. The point is not paperwork. The point is memory.

A good record saves you from the worst version of disclosure writing: four authors in a shared chat, three days before the deadline, trying to remember whether the interview guide came from a prompt or a whiteboard.

Generate an AI Usage Card when you start the paper, update it before submission, and copy the relevant text into the method, acknowledgment, or appendix. If your paper later goes to a journal, use the same card to answer the journal's AI declaration questions. For that step, see AI transparency requirements for journal submissions.

Final check before you submit

Before you upload a CHI-style paper, ask one plain question:

Would a reviewer change their reading of the method if they knew where AI entered the work?

If yes, disclose it in the method. If no, check the venue rules and consider a short acknowledgment anyway.

HCI research studies people, tools, settings, and interpretation. AI can touch all four. Your disclosure should show readers exactly where.

Generate your AI Usage Card at ai-cards.org, save it with your submission files, and use it to write the AI disclosure your reviewers can trust.