AI Disclosure for Economics Research: Data, Code, Models, and Manuscript Writing

Economics papers need AI disclosure before the replication package exists

Economics researchers already know what documentation feels like.

You document data sources. You archive code. You explain sample restrictions. You track survey instruments, preregistration records, IRB approvals, and conflict disclosures.

AI use now belongs in that same folder.

If you used ChatGPT, Claude, Gemini, Copilot, Perplexity, Elicit, or another AI tool while preparing an economics paper, you should record what the tool did and what you checked yourself. That record does not need to turn your paper into a diary. It should help an editor, reviewer, coauthor, or future reader understand how AI entered the work.

An [AI Usage Card](/) gives you a short, structured way to do that. You can attach it to a submission, paste it into an acknowledgment, or keep it with your data and code archive.

Why economics has a special AI disclosure problem

Economics papers often combine prose, data work, code, theory, and policy interpretation.

AI can touch each part.

A tool might summarize papers for a literature review. It might draft Stata, R, Python, Julia, or MATLAB code. It might help debug a LaTeX table. It might suggest wording for an identification strategy. It might translate field survey materials or produce synthetic examples for teaching.

Those uses differ. A grammar check does not raise the same concern as AI generated code that changes a regression specification.

The American Economic Association says it publishes papers only when authors clearly document the data and code used in the analysis and when access to those materials does not belong only to the authors. The AEA also tells authors to disclose potential conflicts of interest and, when relevant, IRB approval information. [[AI disclosure](/ai-disclosure-in-systematic-reviews-and-meta-analyses/)](/how-to-disclose-ai-use-for-neurips-icml-and-acl-submissions/) should sit beside these existing research records, not replace them. (aeaweb.org)

If your economics paper uses AI in ways that affect data, code, model selection, or interpretation, you should treat that use as part of the research workflow.

For broader field guidance, see AI Disclosure for Social Science Research and AI Transparency Requirements for Journal Submissions.

Start with the use, not the tool name

A weak disclosure says:

"ChatGPT was used during manuscript preparation."

That tells the editor almost nothing.

A useful disclosure says what you asked the tool to do, where you used the output, and how you checked it. Tool names matter, but the role matters more.

You can group AI use into four buckets.

First, writing support. This includes grammar edits, sentence shortening, abstract drafts, title options, and reviewer response wording.

Second, research support. This includes literature search help, paper summaries, citation leads, and concept explanations.

Third, computational support. This includes code generation, code debugging, data cleaning suggestions, table formatting, simulation scripts, and replication checks.

Fourth, analytical support. This includes help comparing model specifications, explaining mechanisms, interpreting regression output, or drafting policy implications.

The last two buckets need more detail because they can affect the results.

If you used an AI coding assistant, say whether a human author reviewed each code change, ran the script, inspected output, and compared the result against the intended analysis plan. If you used AI to interpret output, say whether the authors accepted, revised, or rejected that interpretation.

The guide AI tools for research: a disclosure log template for papers and theses can help you keep this record while the project is still moving.

What to record in an economics AI disclosure log

You do not need to save every prompt forever.

You do need enough information to reconstruct the role AI played.

For an economics paper, record:

• Tool name and version if available
• Date range of use
• Task type
• Input type, such as public text, draft prose, code, summary statistics, or confidential data
• Output type, such as revised prose, code suggestions, literature notes, or table formatting
• Human review step
• Location in the paper or repository where the output mattered

This matters most when AI touches empirical work.

If you pasted restricted data, confidential survey responses, proprietary firm data, referee reports, or identifiable participant information into an AI service, you may have a privacy, contract, or ethics issue. Do not hide it in vague wording. Talk to your coauthors, data provider, IRB, supervisor, or editor before submission.

The AEA Data Editor guidance tells authors to prepare replication packages with data citations and provenance information. If AI helped create or modify code in that package, your disclosure log can help you explain provenance for code changes too. (aeadataeditor.github.io)

How AI disclosure fits with data and code archives

AI disclosure does not replace a replication package.

A replication package lets another researcher run the analysis. An AI Usage Card explains how AI assistance entered the work.

Those two records answer different questions.

Your data and code archive might include:

• Raw data access notes
• Cleaned data scripts
• Analysis scripts
• Simulation files
• Table and figure code
• README files

Your AI Usage Card can sit next to that archive and explain whether AI helped write, edit, debug, or check those materials.

For example, if GitHub Copilot suggested R code for a bootstrap routine, record that the authors reviewed and tested the code before using it. If ChatGPT helped rewrite a README, record that the authors checked the file paths and commands.

The AEA policy connects publication to clear documentation of data and code. AI disclosure adds a layer that tells readers where machine assistance entered the documentation and analysis process. (aeaweb.org)

If you want examples of short and long disclosure formats, use [AI Usage Cards Examples and Templates](/ai-usage-cards-examples/).

Example disclosure for an empirical economics paper

Here is a practical disclosure for a paper that uses AI in writing and code support.

\section*{AI usage disclosure}
 
The authors used ChatGPT and GitHub Copilot during preparation of this manuscript.
 
ChatGPT was used to improve the clarity of draft prose in the introduction and conclusion. The authors reviewed and edited all suggested text.
 
GitHub Copilot was used for code completion while preparing R scripts for data cleaning and table generation. The authors reviewed the suggested code, ran the scripts locally, inspected the output, and retained responsibility for all analysis decisions.
 
No AI tool was used to generate data, select the empirical specification, interpret confidential data, or write the final conclusions without author review.

This disclosure gives an editor a useful map. It names the tools. It names the tasks. It separates writing help from analysis decisions.

You can generate a cleaner version at ai-cards.org and use the card as a submission attachment or as text in your manuscript.

Example disclosure for a theory paper

Theory papers can use AI too. The disclosure may look different.

AI might help polish exposition, check notation consistency, draft examples, or convert notes into LaTeX. If it suggests proofs, assumptions, or lemmas, you need to explain how the authors verified them.

\section*{AI usage disclosure}
 
The authors used Claude to revise explanatory prose and to check consistency in notation across draft sections.
 
Claude also suggested wording for one informal example in Section 3. The authors rewrote the example and checked it against the model assumptions.
 
No AI tool was used as an author, and no AI output was accepted as a proof without independent verification by the authors.

This matters because AI tools can produce mathematical text that sounds plausible and fails under inspection. A disclosure should not create trust by force. It should give readers enough context to judge the work.

For authorship questions, see Can AI Be a Co-Author on a Research Paper?. COPE states that AI tools cannot be listed as authors because they cannot take responsibility for a paper. (publicationethics.org)

Example disclosure for literature review and policy work

Economics papers often use literature summaries to build context. AI tools can help, but they can also invent citations, flatten disputes, or miss identification details.

If you used AI for literature work, say whether the tool found sources, summarized sources, or drafted prose. Then say how you verified the sources.

\section*{AI usage disclosure}
 
The authors used Perplexity AI and ChatGPT to identify candidate papers for the literature review and to summarize abstracts during early screening.
 
All cited sources were checked by the authors against the original publications. The authors did not cite any source based only on an AI-generated summary.

If you used an AI paper finder, connect the disclosure to your search process. The guide How to Use AI Paper Finders and Disclose Them in Research gives more detail.

Where to put the AI disclosure

Journal rules differ. Check the instructions before submission.

Some publishers ask for a separate AI declaration near the acknowledgments or before the references. Elsevier states that authors who use generative AI or AI assisted technologies in the writing process must disclose that use in a statement at the end of the manuscript, before the reference list. (sciencedirect.com)

Taylor & Francis tells authors that they remain accountable for originality, validity, integrity, and references when they choose to use generative AI tools. (taylorandfrancis.com)

For economics journals, you may need to satisfy several layers at once: journal AI rules, publisher rules, society rules, data and code policies, funder rules, and university rules.

A good workflow looks like this:

1. Keep an AI disclosure log during the project.
2. Generate an AI Usage Card before submission.
3. Add a short statement to the manuscript if the journal asks for one.
4. Place the card with your supplementary materials or internal project records.
5. Update the card after revision if AI use changes.

If you work in Overleaf or LaTeX, see the LaTeX Tutorial for AI Usage Cards and How to Use AI Usage Cards in Overleaf.

What editors and reviewers want to know

Editors do not need theater. They need a clear answer to a few questions.

Did AI write text that appears in the paper?

Did AI help produce code, tables, figures, simulations, or appendices?

Did any author paste restricted, confidential, or personal data into a tool?

Did the authors verify citations and results?

Did humans make the final research decisions?

A short AI Usage Card answers these questions without making the manuscript clumsy. It also helps coauthors align before submission. That alone can save trouble, especially when one coauthor used an AI coding assistant and another assumed no AI touched the analysis.

If you are unsure whether your use needs disclosure, read Do I Need to Disclose AI Usage in My Paper?. When AI affects claims, analysis, interpretation, code, or reader trust, disclose it.

A practical rule for economics researchers

Ask one question:

Would a reader, reviewer, replication team, or editor understand the paper differently if they knew how AI helped?

If yes, document it.

A grammar edit may need only a short note, or no note if the journal exempts basic language checking. AI generated code, literature summaries, analysis suggestions, translated survey materials, or rewritten policy claims need more detail.

Do the work while you still remember it. Open ai-cards.org, generate an AI Usage Card, and save it with your manuscript and replication files. Then, if the editor asks how AI entered the paper, you will not need to reconstruct the answer from memory.