How Should AI-Assisted Translation Be Documented in Research?

What Should Researchers Record When AI Helps Translate Research Materials?

You have used generative AI to translate a questionnaire, interview guide, participant information sheet, or research manuscript. You reviewed the output, corrected several expressions, and incorporated the translated material into your study.

Months later, a reviewer asks how the translation was performed. Which AI system did you use? Were the original instructions preserved? Did a bilingual expert review the text? Were culturally specific expressions changed? How did you determine that the translated questionnaire retained its intended meaning?

If the translation process was not documented, answering these questions may depend on incomplete recollection.

AI-assisted translation creates a research record that extends beyond the final translated document. Researchers need to distinguish the machine-generated draft from the reviewed version and explain how consequential decisions were made.

Document the Translation Process, Not Just the AI Tool

What Belongs in an AI-Assisted Research Translation Record?

Start by Identifying What Was Translated and Why

Translation documentation should begin with the research material itself. A record stating only that ChatGPT or another AI system was used is insufficient to explain the translation process.

Researchers should identify the source language, target language, document type, original version, and intended audience.

The purpose matters because translating a published abstract for general communication differs from translating an informed consent form or a validated psychological questionnaire.

For example, a questionnaire translation may require evidence of conceptual equivalence and instrument adaptation, whereas a translated interview transcript may require attention to participant meaning and interpretive context.

Where relevant, the record should also identify regional language varieties. Filipino, for instance, should not automatically be treated as interchangeable with every Philippine language or regional variety.

Record the AI Tool, Model, and Relevant Settings

AI systems can change over time. A commercial product may use different underlying models across versions, subscriptions, or service configurations.

Researchers should record the information available when translation occurred, including the tool or service, model identifier where disclosed, translation date, and relevant configuration details.

If the interface does not reveal an exact model identifier, researchers should state what is known rather than guessing.

Where translation settings, terminology constraints, or other options materially affect the output, those details may also belong in the record.

Recording the model does not guarantee that the same translation can later be regenerated. However, it supports provenance and helps explain which system contributed to the research workflow.

Preserve Significant Prompts and Translation Instructions

AI translation is influenced by the instructions researchers provide. A request to translate literally may produce different wording from a request to preserve conceptual equivalence or adapt language for a particular audience.

Consequential instructions should therefore be documented.

For example, a researcher might instruct the system to retain technical terminology, preserve uncertainty, avoid adding explanations, and flag culturally specific expressions without direct equivalents.

Not every exploratory prompt necessarily needs to appear in the published manuscript. However, prompts that materially influenced the final translation or validation decisions may be important to the internal research record.

Researchers should also avoid preserving confidential information unnecessarily in prompt logs or shared documentation.

Keep the Source, AI-Generated Draft, and Approved Translation Distinguishable

One of the most useful documentation practices is to preserve identifiable versions of the translated material.

The original source, initial AI-generated translation, reviewed draft, and final approved version should not be confused with one another.

Version control, descriptive filenames, or controlled document histories can help maintain these distinctions.

For research instruments, the exact version administered to participants should be identifiable. A translation revised after data collection should not be presented as though participants received the revised wording.

Document Human Review and Its Consequences

Researchers should record whether a bilingual reviewer, disciplinary specialist, or translation committee evaluated the material.

Useful details include reviewer qualifications, the scope of review, identified discrepancies, revisions, and the process used to resolve disagreements.

A statement such as "the translation was checked by experts" provides limited methodological information if readers cannot determine what the experts examined.

For consequential materials, qualified bilingual expert review may examine conceptual fidelity, terminology, cultural suitability, and intended audience comprehension.

Documentation should explain the review procedure without unnecessarily disclosing personal information about reviewers.

Record Decisions About Technical and Cultural Terminology

Some translation decisions are straightforward. Others require researchers to choose among several plausible expressions or retain a source-language term.

These decisions may affect how constructs are understood, how participant responses are coded, or how findings are communicated.

A terminology log can record the source expression, proposed translation, approved equivalent, definition, and reason for the decision.

Such records are particularly useful when researchers must preserve specialized scientific terminology or explain culturally specific expressions that do not have simple equivalents.

Not every synonym requires a formal entry. Documentation should prioritize decisions that could materially influence interpretation or measurement.

Report How Translation Accuracy Was Evaluated

Translation documentation should distinguish the procedures performed from the conclusions researchers draw about accuracy.

Possible procedures include source-target comparison, independent bilingual review, back-translation, cognitive interviewing, pilot testing, and psychometric evaluation.

These procedures provide different forms of evidence.

For example, back-translation may help identify discrepancies between source and target wording, but it does not independently establish that respondents understand an instrument as intended.

Researchers should describe the procedures actually conducted, relevant findings, and unresolved limitations. They should not claim that an instrument was validated merely because an AI system or bilingual reviewer found its wording acceptable.

Documentation Is Especially Important for Research Instruments

When AI translates a questionnaire, scale, or assessment, the translation may influence the properties of the resulting measurements.

Researchers should record the original instrument version, translation procedure, item revisions, response-option adaptations, pretesting, and relevant measurement analyses.

If different language versions are compared, the research report may also need evidence concerning measurement invariance or differential item functioning.

The possibility of measurement nonequivalence introduced by translation makes it particularly important to preserve item-level changes.

Documentation allows later reviewers to examine whether an unexpected finding might be associated with wording differences rather than the underlying construct.

Qualitative Translation Records Should Preserve Interpretive Decisions

In qualitative research, translation may occur before coding, during analysis, or when preparing quotations for publication.

Each approach can influence how meaning is represented.

Researchers should document which materials were translated, when translation occurred, who reviewed the translated text, and how ambiguities were resolved.

Where ethically permissible, preserving relevant source-language excerpts alongside translated quotations can help support interpretive transparency.

The record should also distinguish a participant's original expression from a researcher's explanatory paraphrase.

Researchers should not assume that a polished AI translation is a neutral representation of participant speech.

Internal Documentation and Published Disclosure Serve Different Purposes

The complete translation record may contain detailed prompts, draft versions, reviewer comments, and validation results.

A published manuscript usually requires a more concise methodological account.

Researchers should consult the target journal's current AI-use and disclosure policies. Requirements differ across publishers, journals, and disciplines.

Disclosure of an AI tool does not replace the need to describe translation procedures when those procedures affect the study's methodology.

Protect Confidential and Sensitive Translation Records

Translation records may contain unpublished findings, identifiable interview responses, consent materials, or confidential institutional information.

Researchers should determine whether the AI service is authorized to process the material before uploading it.

They should also consider access controls, retention periods, participant consent, institutional data-governance requirements, and restrictions on sharing original transcripts or prompt histories.

Transparency does not require publicly releasing confidential research material. Where necessary, researchers can provide methodological descriptions, redacted examples, or controlled-access records.

Documentation Should Support Auditability Without Becoming Unmanageable

An exhaustive transcript of every interaction with an AI system may be unnecessary for many translation tasks.

The more useful objective is to preserve a defensible account of the process, particularly the decisions that influenced the final research material.

Researchers can use a standardized translation log to record consequential information consistently across documents and languages.

The appropriate level of detail depends on the translation's methodological significance and the requirements governing the study.

Documenting AI Translation of an Online Learning Questionnaire

A concise methodological description might explain that generative AI assisted with the initial translation, that bilingual reviewers examined conceptual and linguistic equivalence, and that the final version underwent appropriate respondent-based evaluation.

The complete internal record would retain the detailed versions and decisions supporting that description.

This documentation does not prove that every item is equivalent across languages. It establishes what was done and allows the evidence for translation quality to be evaluated.

Common Mistakes in Documenting AI-Assisted Translation

How Much AI Translation Documentation Does Your Study Need?

Documentation should be proportionate to the role translation plays in the research.

Where translation affects participant rights, research measurements, or the interpretation of evidence, researchers should maintain sufficiently detailed records to explain and defend consequential decisions.

A practical approach is to create a translation log before translation begins. This reduces reliance on memory and encourages researchers to identify consequential decisions as they occur.

The objective is not documentation for its own sake. It is to make the translation process sufficiently transparent that its influence on the research can be evaluated.

Before Finalizing AI-Assisted Research Translation Records

Frequently Asked Questions About Documenting AI Translation

Document the Decisions That Connect AI Translation to Research Evidence

How to Cite This Guide

This guide is intended to be read, shared, and used in research, teaching, and academic work. If you draw on its ideas, explanations, or other content, please acknowledge the source by citing the guide. Doing so gives appropriate credit and helps your readers locate the original resource.