Should AI Use Be Planned Before a Research Project Begins?

Should Researchers Decide How They Will Use AI Before Collecting Data?

A research team develops a study protocol involving student surveys and interviews. The protocol describes the research questions, sampling procedures, instruments, and planned analyses but does not mention artificial intelligence.

After collecting the data, the researchers discover that a generative AI system can classify interview responses and generate analytical code. They decide to incorporate the system into their workflow.

The new approach appears efficient, but it raises questions that were not considered during the original planning process. Is the proposed data processing authorized? Does the AI-assisted classification fit the qualitative methodology? Should the analytical change be documented as a departure from the original plan?

Must every possible AI use be specified before a study begins, or can researchers adopt AI responsibly as new needs and opportunities emerge?

Plan Consequential AI Use Early, but Allow Documented and Authorized Changes

Why AI Planning Matters Before and During Research

Why Should AI Be Considered During Research Design?

Research design establishes how evidence will be collected, processed, analyzed, and interpreted.

When AI contributes to these procedures, it may introduce additional assumptions, dependencies, and risks.

For example, using AI to classify survey responses creates questions about the classification framework, model performance, output variability, and validation.

Using AI to process confidential interviews raises questions about authorized data handling and participant expectations.

Considering these issues during planning allows researchers to design appropriate safeguards rather than attempting to reconstruct them after the analysis is complete.

Early planning is particularly valuable when AI outputs will become primary research measurements or materially influence scientific conclusions.

Which AI Uses Should Be Specified Before the Study Begins?

The need for advance planning depends on the contribution and its consequences.

These categories are illustrative rather than universal approval classifications. Institutional and journal requirements may differ.

Planning AI Use Is Not the Same as Committing to AI

A research protocol may identify AI as a possible analytical option without requiring its eventual use.

For example, a team may plan to evaluate whether an AI classifier performs adequately on a reference dataset before deciding whether to use it for the full analysis.

If validation is unsatisfactory, the team may retain an established alternative.

This conditional approach can be scientifically preferable to assuming that AI will be appropriate before sufficient evidence exists.

The protocol should distinguish proposed evaluation from authorized deployment, especially when participant information or consequential decisions are involved.

Why Early Planning Helps Protect Participant Information

Researchers may collect information under consent and confidentiality arrangements that restrict subsequent processing.

Introducing an external AI service after data collection may create a use that participants or the responsible institution did not authorize.

For example, an interview transcript may contain identifiable information even after names are removed.

Researchers should therefore establish which systems may process the data, what information will be supplied, how records will be retained, and whether the proposed activity is compatible with existing obligations.

Planning before collection may allow researchers to design appropriate consent language and secure processing arrangements where needed.

However, consent alone does not necessarily satisfy all legal, ethical, or institutional requirements.

When Does AI Use Require Ethics Review?

AI use may require ethics review when it changes approved research procedures, introduces new participant risks, or materially affects how participant information is processed.

Requirements depend on the jurisdiction, institution, and research context.

For research governed by applicable U.S. Common Rule provisions, changes to approved research generally require prior institutional review board approval, except where necessary to eliminate apparent immediate hazards to participants.

Other regulatory and institutional systems may apply different procedures.

Researchers should consult their responsible ethics authority rather than assume that a general AI policy determines whether a protocol amendment is required.

Approval should be obtained before implementing changes that require it.

Why AI-Assisted Measurement Should Be Planned in Advance

When AI generates categories, scores, or other research variables, it becomes part of the measurement process.

Measurement procedures influence what the study can legitimately conclude.

For example, an AI system used to classify student engagement may produce different labels depending on the prompt, model version, and contextual information supplied.

Researchers should define the intended construct, classification rules, validation procedure, and handling of uncertain cases.

These decisions are particularly important in longitudinal studies where measurements from different periods must be comparable.

Planning can help prevent changes to classification procedures from being mistaken for changes in the underlying phenomenon.

Should Researchers Prespecify AI-Assisted Analyses?

Prespecification can strengthen transparency when a study tests planned hypotheses or estimates predefined outcomes.

Researchers may specify which analytical procedures will be used, how variables will be constructed, and what validation criteria must be satisfied.

This can reduce the risk of choosing methods after examining results in ways that favor a preferred conclusion.

However, not every research design requires complete prespecification of all analytical decisions.

Exploratory and interpretive approaches may legitimately evolve as researchers engage with evidence.

The important requirement is to report accurately which analyses were planned, which emerged later, and why consequential changes occurred.

What if AI Suggests a Better Method After Data Collection?

An AI recommendation may identify a genuine weakness in the original analytical plan.

For example, it may draw attention to clustering, missing-data mechanisms, or measurement problems that the team had not adequately considered.

Researchers should evaluate the suggestion against the research design and methodological literature.

If a change is justified, it may be appropriate to revise the analysis.

However, the team should document the original plan, the reason for the change, and whether the revised analysis was developed after inspecting the data or results.

A defensible methodological correction should not be concealed to create the appearance that the revised procedure was always planned.

How Should Researchers Distinguish Planned and Exploratory AI Use?

Both can be legitimate.

The distinction becomes important when AI changes measurements, analytical decisions, or the interpretation of results.

Researchers should avoid presenting exploratory analyses as prespecified confirmatory tests.

Where a new AI procedure changes approved participant-related activities, the necessary authorization should be established before implementation.

Why Model Selection Should Be Considered Early

AI models differ in their capabilities, documentation, deployment arrangements, and stability.

A model may be suitable for generating exploratory suggestions but inappropriate for a study requiring fixed and reproducible classifications.

Researchers should consider whether the selected provider offers sufficient model identification, version control, output retention, and validation information.

For long-term studies, changes in model behavior may complicate comparisons across time.

Planning can establish how model updates will be handled and whether a stable alternative is needed.

Researchers should avoid assuming that a familiar product name guarantees a fixed underlying model.

What Should an AI-Use Plan Contain?

The plan should be proportionate to the intended research activities.

A plan does not need to be a separate lengthy document for every minor task. Relevant provisions may be incorporated into the research protocol, data management plan, analysis plan, or institutional documentation.

Should AI Planning Include Alternatives?

Yes, particularly when AI performs a consequential analytical function.

Researchers should identify what will happen if the selected model becomes unavailable, fails validation, or changes during the project.

For example, a team planning AI-assisted literature screening may establish a procedure for independent human screening if model performance is inadequate.

A quantitative study may preserve validated statistical code that does not depend on regenerating scripts through an AI service.

Contingency planning reduces dependence on a particular provider and helps maintain methodological continuity.

How Can Researchers Avoid Excessive Planning?

Planning should be proportionate to scientific and ethical consequences.

Requiring formal approval for every minor wording suggestion may create unnecessary administrative burden.

By contrast, using AI to process confidential participant data or generate primary research measurements may warrant substantial advance review.

The appropriate approach is to establish clear categories of permitted use, with additional review for activities that introduce new risks or materially affect the study.

Researchers should not confuse a large volume of documentation with effective governance.

The objective is to ensure that consequential decisions are anticipated and managed, while legitimate research flexibility remains possible.

What if AI Use Changes During a Long Research Project?

Changes may occur because new tools become available, existing models are updated, or the research team identifies additional analytical needs.

The team should assess whether a proposed change affects the original methodology, participant risks, information processing, or reproducibility.

Minor changes may be manageable within the existing plan and institutional policies.

Material changes may require updated documentation, additional validation, or formal approval.

Researchers should preserve the distinction between the original procedure and the revised one, especially when results from different periods are compared.

When Should Researchers Decide Not to Use AI?

Advance planning may reveal that AI is unsuitable for a proposed task.

The system may lack adequate validation, require unauthorized access to sensitive data, introduce unacceptable participant risks, or undermine essential reproducibility.

A simpler validated procedure may also perform the task with fewer dependencies.

In such circumstances, excluding AI from the workflow may be the more defensible decision.

Planning is valuable partly because it allows these limitations to be identified before they affect the research record.

Planning AI Use Before Collecting Student Interview Data

Early planning does not establish that AI-assisted coding will necessarily improve the analysis.

It ensures that the researchers can evaluate the proposed contribution before sensitive information is processed and before the resulting outputs influence the study.

Misconceptions About Planning AI Use Before Research

Specify Consequential AI Use Before It Shapes the Evidence

Before beginning a project, identify whether AI will influence participant information, measurement, analysis, or other consequential research procedures.

Researchers should establish which AI uses require approval before participant data are collected and when proposed AI activities require ethics review.

Planning should be incorporated into the broader AI-assisted research workflow, including procedures for managing model changes during long projects.

Where AI use emerges after the original plan, accurate documentation helps distinguish legitimate methodological development from undisclosed changes that could affect interpretation.

Planning should protect scientific accountability without preventing researchers from responding thoughtfully to new evidence or improved methods.

Before Beginning a Research Project That May Use AI

Frequently Asked Questions About Planning AI Use in Research

Plan AI Use Early Enough to Protect Methodological and Ethical Integrity

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.