AI Disclosure for Education Research: Student Data, Classroom Materials, and Journal Submissions

AI disclosure for education research starts before the paper

Education research puts [AI disclosure](/how-to-disclose-ai-use-for-neurips-icml-and-acl-submissions/) under pressure.

A researcher may use ChatGPT to draft survey items, Claude to summarize interview memos, Copilot to debug R code, or an image model to create classroom materials for a study. Each use may look small on its own. Together, they can affect the design, data handling, analysis, and wording of a paper.

That means education researchers need more than a one sentence note in the acknowledgments. They need a record.

An [[AI Usage Card](/chatgpt-disclosure-academic-papers/)](/what-are-ai-usage-cards/) gives you that record. It names the tool, the task, the stage of research, the type of input, the human checks, and the place where the AI assistance appears in the manuscript. For education studies, that record also helps you think about student privacy before you paste anything into an external system.

Use the free generator at ai-cards.org while you work, not only after the paper feels done.

Why education research needs a different level of care

Education studies often involve classroom talk, student writing, grades, learning management system traces, teacher notes, parent interviews, or school records. These materials can identify a student even when you remove names.

The U.S. Department of Education describes FERPA personally identifiable information as data in education records that includes direct identifiers, indirect identifiers, and other information that can identify a student through links with other information. Dates of birth, student IDs, small class contexts, rare disability categories, or a detailed writing sample can create risk even without a name. (studentprivacy.ed.gov)

That privacy point changes the AI disclosure task. You should not only state that you used a model. You should say whether you entered student data, de-identified excerpts, synthetic examples, survey instruments, code, or manuscript text.

A disclosure that says "ChatGPT was used for editing" does not tell an editor whether you uploaded classroom transcripts. It does not tell a school partner whether student records left an approved environment. It does not tell a reviewer whether an AI tool shaped the coding scheme.

Education research already shares many disclosure concerns with social science research. The student data context adds another layer: you need to connect AI use to consent, IRB review, data sharing agreements, and school partner rules.

Check journal rules before you write the statement

Journal policies differ.

AERA Open tells authors to reveal AI generated content in a submission, including text, images, or translations. It also tells authors who use generative AI to include a short acknowledgment in the paper and submit a fuller disclosure using Sage's template. The same guidance says tools such as ChatGPT should not appear as authors. (journals.sagepub.com)

Taylor & Francis asks authors to acknowledge generative AI use in a statement that names the tool, version number, how they used it, and why they used it. It also states that generative AI tools cannot take authorship because they cannot take responsibility for the work or manage copyright and licensing agreements. (taylorandfrancis.com)

ICMJE gives a similar rule for medical journals: authors should disclose AI assisted technologies at submission and describe how they used them in the cover letter and manuscript. ICMJE also states that chatbots and other AI tools should not appear as authors because they cannot take responsibility for accuracy, integrity, and originality. (icmje.org)

If your education study sits near health, psychology, learning sciences, disability research, or child development, read the target journal's author instructions before submission. The safe habit: create your own record first, then adapt it to the journal. Our [journal submission guide](/ai-transparency-journal-submission/) shows how to place the statement in the cover letter, methods section, acknowledgments, or supplementary file.

What to record for an education study

A useful record answers seven plain questions.

First, name the tool. Include the model or version when the interface gives you one.

Second, name the task. Say "drafted distractors for a reading comprehension pilot" rather than "writing assistance."

Third, name the research stage. Separate study design, instrument development, recruitment materials, transcription support, coding, statistical analysis, figure drafting, and manuscript editing.

Fourth, describe the input. State whether you entered public text, invented examples, de-identified student work, identifiable records, code, or your own manuscript.

Fifth, describe the output. Did the model produce suggested wording, labels for qualitative codes, R code, summaries, images, or feedback on structure?

Sixth, describe human review. Name the check you performed: source verification, code inspection, inter-coder discussion, re-analysis, or comparison with the approved protocol.

Seventh, say where the AI shaped the submitted work. Point to the survey appendix, interview guide, analytic memo, code repository, figure caption, or prose edits.

The disclosure log template for research tools gives you a place to track these details as they happen. That habit saves you from reconstructing prompts two days before submission.

Common education research cases

Survey and instrument design

AI tools can help draft candidate survey items, simplify instructions, or suggest age appropriate wording. Do not treat these outputs as validated instruments.

Your disclosure should say that you generated candidate items and then revised, piloted, or validated them through your own process. If you used an existing scale, do not ask a model to paraphrase copyrighted items unless you have permission and understand the license.

Example:

"ChatGPT-5 was used in March 2026 to draft candidate wording for six classroom belonging items. The authors revised all items, removed AI generated items that overlapped with existing scales, and piloted the final instrument with the procedure described in Section 3.2."

Qualitative coding and interview analysis

Education researchers often study interviews with teachers, students, school leaders, parents, or tutors. A model can suggest code labels or summarize a memo, but that use can affect interpretation.

If you used AI during coding, say whether the model saw transcripts, excerpts, paraphrases, or memos. If transcripts included students or school staff, name the privacy step. For more detail, pair this article with our guide to AI disclosure for qualitative research.

A weak disclosure says:

"AI helped with qualitative analysis."

A stronger one says:

"Claude Sonnet 4 was used to suggest candidate labels for researcher written analytic memos. The tool did not receive raw interview transcripts. Two authors reviewed the suggestions, rejected labels that introduced unsupported interpretations, and used the final codebook shown in Appendix B."

Student writing and classroom artifacts

Student writing carries voice, context, and personal detail. Even a short excerpt can identify a child in a small school or specialized program.

If you use AI to classify or summarize student writing, explain whether you worked with consented data, de-identified excerpts, local tools, or school approved systems. Do not hide this in a generic methods note.

FERPA generally bars disclosure of personally identifiable information from education records without written consent, though it allows data sharing under specific conditions for studies, audits, evaluations, and other listed purposes. The Department of Education also notes that written agreements often protect student data when institutions share data with vendors or outside parties. (studentprivacy.ed.gov)

AI disclosure cannot replace IRB approval or a data sharing agreement. It can, however, show editors and readers that you treated AI use as part of the research record.

Coding, statistics, and reproducible analysis

Many education papers rely on R, Python, Stata, SPSS, or qualitative analysis software. If Copilot, ChatGPT, Gemini, or another tool helped write code, disclose the role when the tool affected analysis.

You do not need to turn the methods section into a transcript. Name the analytic task and the checks.

Example:

"GitHub Copilot was used to suggest R syntax for data cleaning and visualization. The authors inspected the code, ran unit checks on derived variables, and compared descriptive statistics with the preregistered analysis plan before producing the tables."

If you need example wording, see AI Usage Cards examples and templates.

A short LaTeX disclosure template

You can paste an AI disclosure into the acknowledgments, methods, or a journal specific statement field. If you write in LaTeX or Overleaf, keep the wording close to the manuscript section it supports.

\section*{AI use disclosure}
 
The authors used \textit{[tool name, model/version]} on \textit{[dates or month/year]}
for \textit{[task: e.g., drafting candidate survey items, revising prose, suggesting R syntax]}.
 
The tool received \textit{[type of input: e.g., researcher-written prompts, de-identified excerpts,
synthetic examples, manuscript paragraphs, code]}. It did not receive \textit{[excluded data:
e.g., identifiable student records, raw interview transcripts, school names]}.
 
All AI output was reviewed by the authors. The authors checked \textit{[accuracy, citations,
code output, qualitative interpretations, privacy constraints]} and take responsibility for
the final manuscript.

If your journal wants the statement in the acknowledgments, adapt the same text. If it wants a supplementary file, attach the full AI Usage Card. Our LaTeX tutorial for AI Usage Cards and Overleaf guide show how to include the card in a manuscript workflow.

How AI Usage Cards differ from other documentation

AI Usage Cards document how you used AI in a specific scholarly work. They do not evaluate the model as a product, and they do not document a dataset.

That distinction matters.

A Model Card describes a model's intended use, limits, evaluation data, and performance patterns. A Datasheet for Datasets describes how a dataset was created, collected, and maintained. A System Card can describe a deployed AI system and its safeguards.

An AI Usage Card answers a smaller but sharper question: what did the researcher do with AI while producing this paper?

If your education study also releases a dataset, you may need both a dataset document and an AI Usage Card. If your paper introduces a tutoring model, you may need a Model Card too. Our comparison of AI documentation frameworks explains how these records fit together without turning them into substitutes for each other.

A practical workflow for education researchers

Start with the target journal's rules. Then check the IRB protocol, data management plan, consent language, school partner agreement, and tool terms.

Create an AI Usage Card before you upload data or draft prompts. Treat it like a lab note. If you decide not to enter student data into an AI system, write that down. If you use only synthetic examples, write that down too.

During analysis, keep the AI record near your codebook, analysis scripts, or memo folder. At submission, turn that record into the journal's required statement.

Before you submit, ask one blunt question: could a reviewer tell what the AI tool saw and what it changed?

If the answer is no, revise the disclosure.

Then generate a clean card at ai-cards.org. You can copy the text into your acknowledgments, attach the card as supplementary material, or use it to fill the AI disclosure field in a journal submission system. Education research asks readers to trust how you handled people, classrooms, and evidence. Give them a record they can read.