Should Researchers Record the Date of AI Use?

Does It Matter When You Used AI, or Is Naming the Tool Enough?

A researcher uses ChatGPT to classify open-ended survey responses in January. The classifications are reviewed and incorporated into the study's analytical dataset.

Six months later, the researcher returns to the same application and repeats the original prompt. Several responses receive different classifications.

The researcher remembers the tool and the instructions but did not record the original interaction date, underlying model, or settings.

Could the difference reflect a model update? Did the original conversation contain additional context? Is there enough information to reconstruct what happened?

Now consider a researcher who uses AI once to correct punctuation in a manuscript. Would recording an exact timestamp provide the same methodological value?

These situations illustrate why the date of AI use can be an important research record, although its significance depends on what the technology contributed to the study.

Record Dates for Consequential AI Activities, Especially When Models May Change

A date provides temporal context but does not establish the exact underlying model or guarantee reproducibility. It should be considered alongside prompts, outputs, model information, settings, and relevant verification records.

Why the Timing of AI Interactions Matters for Research Documentation

Why Should Researchers Record When AI Was Used?

Research documentation should make consequential procedures and decisions traceable.

Recording the date of an AI interaction helps establish when an output was generated and how it relates to subsequent research activities.

For example, a researcher may use AI to generate preliminary qualitative codes before conducting human review and revising the coding framework.

Dates can help distinguish the original AI output from later analytical decisions.

They may also help researchers identify which service configuration or model release was potentially available at the time.

However, the date is contextual evidence rather than proof of the exact technology used.

Why Are AI-Use Dates More Important for Some Activities?

The significance of a date depends on the role of AI in the research.

For example, an AI model used to classify research data may influence the findings directly.

If the provider subsequently changes the model, researchers may need to establish when the original classifications were produced.

By contrast, the exact date of a minor punctuation correction may have limited relevance to evaluating the research.

Researchers should therefore prioritize recording dates for consequential AI-assisted procedures rather than assuming that every interaction requires identical documentation.

How Do AI Model Updates Affect Research Reproducibility?

Commercial AI services may change their underlying models, system instructions, interfaces, or supporting infrastructure over time.

Consequently, a researcher who repeats an earlier prompt may not obtain the same response.

For example, an AI system used to classify student feedback may generate different labels after a service update.

Such differences could influence category frequencies or analytical conclusions.

Recording the original date helps establish the historical context of the procedure.

However, exact reproduction may still be limited by unavailable model versions, hidden configurations, or nondeterministic generation.

Does the Date Identify the Exact AI Model?

No. A date alone does not reliably establish which underlying model generated a response.

AI providers may offer several models simultaneously, and different users or interfaces may have access to different configurations.

A provider may also update a service without changing its public-facing name.

For example, knowing that a researcher used ChatGPT on a particular date does not necessarily identify the exact model selected or routed to the interaction.

Researchers should therefore record the model identifier separately where it is available.

The question of recording AI model versions complements, rather than replaces, date documentation.

What Is the Difference Between a Date and a Timestamp?

A date identifies the calendar day on which an activity occurred.

A timestamp provides a more precise record that may include the time and time-zone offset.

Exact timestamps may be important for repeated experiments, time-sensitive data collection, or procedures where chronological order affects interpretation.

For routine documentation, the calendar date may be sufficient.

Researchers should choose a level of precision justified by the study rather than recording unnecessary detail.

When Are Exact Timestamps Particularly Useful?

Exact timestamps may matter when the timing of AI interactions is itself relevant to the research.

Examples include longitudinal evaluations of AI behavior, repeated benchmarking, real-time experimental interventions, and automated workflows involving sequential processing.

For example, a study comparing an AI model's responses before and after a documented service change may need precise interaction times.

Similarly, an experiment involving AI-generated feedback delivered to participants at specific stages may require timestamps to reconstruct the intervention.

The need for precision depends on the research question and the potential consequences of timing differences.

Should Researchers Record the Time Zone?

Where exact timestamps matter, recording the time zone or UTC offset can prevent ambiguity.

For example, a researcher in the Philippines may record an interaction using Philippine Standard Time, while an external service records activity in Coordinated Universal Time.

Without time-zone information, two records may appear to refer to different calendar days even when they describe the same interaction.

For collaborative or international research, a consistent time standard can improve chronological interpretation.

However, time-zone documentation may be unnecessary for activities where only the calendar date matters.

Researchers should avoid presenting locally recorded times as globally unambiguous without identifying the relevant offset.

What Do Major Publishing Authorities Require?

The International Committee of Medical Journal Editors (ICMJE) recommends disclosure of AI-assisted technologies and an explanation of how they were used.

The Committee on Publication Ethics (COPE) similarly emphasizes transparency concerning AI assistance in research and manuscript preparation.

These positions support accurate documentation of consequential AI use, but they should not be interpreted as imposing a universal requirement to record exact timestamps for every interaction.

Individual journals, institutions, funders, or research agreements may establish more specific requirements.

Researchers should distinguish general research-recordkeeping principles from mandatory date-reporting instructions.

Should AI-Use Dates Be Included in the Published Manuscript?

Not necessarily. Recording a date internally is different from reporting it in a publication.

For example, the date of AI-assisted grammar correction may have little relevance to readers.

However, when an AI model is central to the research methodology, the date of access or data collection may be important for understanding the procedure.

A study evaluating an AI model's responses may need to identify when those responses were collected.

Similarly, repeated AI-assisted classifications may require documentation of the periods during which the system was used.

The appropriate reporting depends on the journal's instructions and the methodological significance of the timing.

Where Should AI-Use Dates Be Recorded?

Researchers may record dates in an AI-use log, electronic laboratory notebook, computational experiment record, or another approved documentation system.

The date should be linked to the corresponding AI activity and relevant output.

For example, an entry may identify the date on which a language model generated preliminary classifications and reference the saved coding matrix.

Where AI interactions form part of a formal research procedure, dates may also appear in methods or supplementary materials.

The broader structure of a research AI-use log can help connect dates with tools, prompts, outputs, and human decisions.

Should Researchers Record the Date of Every Prompt?

Not universally. The appropriate granularity depends on the activity.

For a single analytical session using one model configuration, recording the session date and preserving the relevant conversation may be sufficient.

For a study involving repeated model evaluations, each prompt-response interaction may require its own timestamp.

For incidental manuscript editing, recording every prompt time may provide little additional methodological value.

Researchers should therefore distinguish session-level documentation from interaction-level documentation.

The selected approach should be consistent with the study's design and applicable recordkeeping requirements.

What If AI Use Extends Across Several Days?

A single date may not adequately describe a procedure conducted over an extended period.

For example, a researcher may use AI to classify survey responses in several batches over two weeks.

If the service changes during that period, different batches may have been processed under different conditions.

Researchers should consider recording the date associated with each consequential batch or analytical session.

The record should also identify which outputs belong to which session.

Where the same model and settings were used throughout, that information may be documented as part of the overall procedure.

The purpose is to preserve meaningful chronology rather than create unnecessary timestamps.

Should Researchers Record Dates for AI-Assisted Literature Searching?

Dates are particularly important when AI contributes to formal evidence retrieval or study selection.

Literature databases and discovery systems change as new publications are indexed and existing records are updated.

PRISMA 2020 requires reporting the dates on which information sources were last searched.

When AI-assisted searching forms part of a systematic review, researchers should document the relevant search dates and procedures.

However, the date of an exploratory conversation about possible search terms may have different methodological significance.

The requirements concerning AI-assisted literature searching should be considered alongside established evidence-synthesis reporting standards.

Should Researchers Record Dates for AI-Assisted Qualitative Analysis?

Dates may help reconstruct the sequence of AI-generated coding, human review, and subsequent theme development.

For example, a researcher may generate preliminary codes in one session, revise the coding framework after examining the data, and repeat the classification later.

Recording when each stage occurred can help distinguish the original outputs from subsequent analytical decisions.

However, the chronology alone does not establish the credibility of the analysis.

Researchers must still explain the coding procedures, methodological assumptions, and verification undertaken.

Dates support the audit trail but do not replace analytical documentation.

Should Researchers Record Dates for AI-Generated Statistical Code?

Recording the date of code generation may help document the development of a computational workflow.

However, the final tested code and its execution environment are generally more directly relevant to reproducing the statistical results.

For example, a researcher may generate an initial R script in January, revise it in February, and execute the final analysis in March.

The code-generation date and execution date describe different events.

Researchers should avoid treating the date of the original AI response as though it establishes when the final analysis was performed.

Where appropriate, version-control records and execution logs can provide more precise evidence of computational development.

Should Researchers Distinguish Generation Dates From Verification Dates?

Yes, when the distinction is consequential.

An AI output may be generated on one date and verified or incorporated into the research later.

For example, a language model may generate preliminary statistical code in April, while the researcher tests and approves the final implementation in May.

Recording only the generation date would not establish when verification occurred.

A useful research log may therefore distinguish the date of AI interaction, the date of human review, and the date the output was incorporated into the final workflow.

Not every activity requires all these fields, but the distinction can be valuable when reconstructing important research decisions.

Can AI-Use Dates Help Establish an Audit Trail?

Potentially. Dates can connect AI interactions to manuscript drafts, analytical files, research decisions, and verification records.

For example, a saved AI response dated before a particular coding revision may help establish the sequence in which the analysis developed.

However, dates alone do not prove that a procedure occurred as described.

A manually entered date may be inaccurate, and some metadata can be modified.

Researchers should therefore consider dates alongside the content, provenance, and consistency of supporting records.

A credible audit trail depends on the overall reliability of the documentation rather than the presence of timestamps alone.

Can File Metadata Establish When AI Was Used?

File metadata may provide useful supporting information, but it should not automatically be treated as definitive evidence.

Creation and modification dates can change when files are copied, exported, converted, or edited.

Similarly, a screenshot timestamp may indicate when the image was saved rather than when the original AI response was generated.

Researchers should distinguish the date of an AI interaction from the date of a subsequent export or archival action.

Where precise chronology matters, contemporaneous records from multiple sources may provide stronger support than a single metadata field.

What If the AI Provider Displays a Different Date?

Differences may arise from time-zone settings, export conventions, or the way the provider records activity.

For example, a conversation occurring shortly after midnight in the Philippines may have a previous-day timestamp in UTC.

Researchers should examine whether the records refer to the same moment using different time standards.

They should avoid silently changing dates to make records appear consistent.

Where a discrepancy matters, the documentation should explain the time-zone conversion or other verified cause.

If the cause cannot be established, the uncertainty should be acknowledged.

What If the Original Date Was Not Recorded?

Researchers may sometimes recover timing information from saved conversations, file histories, code repositories, laboratory notebooks, or other contemporaneous records.

However, the reliability of reconstructed dates depends on the available evidence.

A researcher should not invent an exact date or timestamp based on a vague recollection.

Where only an approximate period can be established, that limitation should be documented.

For example, a record may indicate that an interaction occurred during a particular month rather than assigning an unsupported day.

Accurate uncertainty is preferable to false precision.

Should Researchers Use the Current Date When Reconstructing an Old AI Record?

No. The date on which a retrospective record is created is different from the date of the original AI interaction.

For example, a researcher may reconstruct an AI-assisted coding activity in October using files created several months earlier.

The log should distinguish the reconstruction date from the estimated or documented interaction date.

Otherwise, the record may misleadingly suggest that the AI activity occurred when the retrospective entry was prepared.

Researchers should identify the supporting evidence and acknowledge uncertainty where necessary.

Can Researchers Rely on AI to Remember When It Was Used?

Not reliably without supporting records.

A generative AI system may not have access to the relevant historical interaction or its original metadata.

Even when an application displays previous conversations, the model's generated response about their timing should not be treated as independent evidence of the date.

Researchers should consult actual records rather than ask the model to infer when an earlier interaction occurred.

Where timing information is unavailable, the limitation should be acknowledged.

How Should Dates Be Formatted in an AI-Use Log?

A consistent and unambiguous format is preferable.

For example, ISO 8601 uses the year-month-day sequence for calendar dates and supports timestamps with time-zone information.

This reduces ambiguity between date conventions used in different countries.

These examples illustrate formatting conventions, not actual research events.

Researchers should record the time standard used and avoid implying greater precision than the available evidence supports.

Should AI-Use Dates Be Preserved With the Research Archive?

Where dates are necessary to interpret consequential AI interactions, they should generally remain connected to the corresponding research records.

For example, archived prompts and outputs may be difficult to interpret if their sequence and timing are unknown.

However, retention periods and storage requirements depend on institutional policies, funder conditions, and the role of the records.

Not every incidental AI timestamp must be retained indefinitely.

The broader question of retaining AI-use records with the research archive concerns preservation of relevant chronology alongside other supporting documentation.

Does Recording Dates Prove That AI Use Was Responsible?

No. A timestamp establishes, at most, information about when an interaction or record occurred.

It does not demonstrate that the AI output was accurate, that the research method was appropriate, or that confidential data were processed lawfully.

A researcher may maintain precise timestamps while accepting incorrect classifications or unsupported interpretations.

Dates should therefore be treated as one component of research documentation rather than evidence of methodological validity.

Responsible AI-assisted research requires appropriate procedures, verification, ethical compliance, and accurate reporting.

Why Recording AI-Use Dates Matters When Analysis Is Repeated

The example illustrates why the timing of AI use can matter when services change or procedures are repeated.

However, it does not establish that a model update caused the differences. Prompt context, settings, sampling variation, or other factors may also contribute.

Dates help researchers investigate such possibilities, but they must be interpreted alongside the broader research record.

Misconceptions About Recording AI-Use Dates

How Should You Record Dates for AI-Assisted Research?

Researchers should choose the level of temporal detail that is necessary to reconstruct consequential activities and satisfy applicable policies.

Dates should be linked to the corresponding prompts, outputs, and research decisions rather than recorded as isolated entries.

Researchers should use a consistent format and record model information separately when it is available and relevant.

The appropriate standard is accurate, proportionate chronological documentation that supports the broader research record without creating false precision.

What to Verify When Recording AI-Use Dates

Frequently Asked Questions About Recording AI-Use Dates

Record When AI Was Used, but Do Not Mistake a Date for a Complete Research Record

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.