Can Generative AI Evaluate Whether a Research Question Is Feasible?

Can AI Tell You Whether You Can Actually Answer Your Research Question?

A research question may be novel and scientifically important yet still be a terrible choice for your particular project.

Perhaps the required participants are inaccessible. The necessary dataset is proprietary. The outcome requires five years of observation but your thesis must be completed in twelve months. The instrument has not been validated for your population. The analysis requires expertise your team does not have. The intervention would be ethically difficult to justify.

Generative AI can identify many constraints like these. But feasibility is unusually context-dependent. A study that is entirely realistic for one research team may be impossible for another.

AI Can Stress-Test Feasibility, but Only With Enough Real Project Information

Research Feasibility Is More Than Asking Whether a Study Sounds Possible

A feasible question needs a feasible route to evidence

A research question is not feasible merely because someone somewhere could investigate it.

The relevant question is whether the proposed research can be conducted rigorously enough, under the actual conditions available to the researchers, to produce evidence capable of addressing the question.

This distinction appears in formal research evaluation. Under the current NIH Simplified Review Framework, reviewers separately assess the importance of proposed research and its rigor and feasibility. Feasibility includes whether the proposed studies can be done well within the stated timeframe, while expertise and resources are also evaluated in relation to the work proposed.

AI needs your constraints before it can evaluate them

Ask a model, “Is this study feasible?” and it may make assumptions about resources you do not have.

It may casually recommend recruiting 500 participants, conducting longitudinal follow-up, obtaining institutional data, using eye-tracking equipment, performing advanced computational analysis, or running a randomized experiment.

Those suggestions may be methodologically imaginable while being completely unrealistic for your situation.

A more useful feasibility evaluation provides explicit constraints: project duration, population access, expected sample pool, available software and equipment, budget, research-team expertise, ethics requirements, data availability, institutional permissions, and any methodological boundaries you already know.

Feasibility has several dimensions

Researchers sometimes reduce feasibility to “Can I recruit enough participants?” Recruitment matters, but it is only one part of the assessment.

AI can help build this inventory. It cannot verify every item without access to the relevant real-world information.

Feasibility begins with alignment between the question and the design

One of the most consequential feasibility problems occurs when the available method cannot answer the question.

Suppose your question asks whether generative AI use causes a decline in students' independent problem-solving ability. You have access only to a one-time self-report survey asking students how often they use AI and how confident they feel about problem solving.

The survey may be easy to administer. The causal question remains infeasible with that design.

You could change the design, narrow the inference, or reformulate the question. What you should not do is call the study feasible merely because data can be collected.

AI can expose hidden dependencies

Research plans often depend on conditions researchers have not yet made explicit.

A longitudinal study depends on participant retention. A study using institutional learning analytics depends on data access and appropriate permissions. A multilingual instrument may depend on valid adaptation. A classroom intervention may depend on instructor cooperation and timetable stability. A study of proprietary AI systems may depend on continued platform availability, pricing, features, and access.

You can ask AI to identify every condition that must remain true for the proposed study to work. This “dependency audit” can reveal fragile parts of the project before they become expensive problems.

AI can help estimate workflow complexity

A proposed project can sound manageable at the level of its research question while becoming intimidating when decomposed into actual tasks.

Ask AI to convert the proposed methodology into a chronological workflow: instrument development, validation or adaptation, ethics review, permissions, recruitment, pilot testing, data collection, cleaning, transcription where relevant, coding or analysis, robustness checks, interpretation, and reporting.

Then attach realistic time estimates yourself or obtain them from experienced researchers and institutional processes.

The model can help reveal the workflow. It should not invent certainty about how long your ethics committee, school system, laboratory, archive, or recruitment process will take.

Feasibility depends on expertise, not just software availability

Having access to R, Python, NVivo, structural equation modeling software, machine-learning libraries, or an AI coding assistant does not mean you have the methodological expertise required to use them appropriately.

AI can explain procedures and help with code, but the researcher remains responsible for understanding assumptions, diagnosing problems, validating analyses, and interpreting results.

If your design requires expertise the team lacks, the solution may be training, collaboration, consultation, methodological simplification, or a different design.

Sample size is not something AI should simply guess

A model may confidently recommend a convenient round number: 100 respondents, 200 participants, 30 interviewees, 10 participants per group.

Those numbers should not be accepted because they sound conventional.

Sample-size requirements depend on the design, estimand, expected precision, statistical model, anticipated effect, clustering, attrition, saturation logic where relevant, and other methodological considerations. Appropriate justification differs substantially across quantitative, qualitative, mixed-methods, and other research traditions.

AI can help identify what information is needed for a defensible sample-size decision. The calculation or rationale should follow the methodological requirements of the study rather than the model's preferred round number.

Ethical feasibility cannot be delegated to AI

Generative AI can identify potential ethical concerns, but it cannot grant ethical approval or determine institutional requirements authoritatively.

A study may require informed consent, privacy protections, safeguards for vulnerable populations, data-security measures, management of incidental findings, permission for secondary data use, or other protections depending on the jurisdiction and research context.

Consult the relevant ethics committee, institutional review board, research office, legal requirements, professional standards, and applicable institutional policies rather than treating chatbot guidance as authorization.

Feasibility can change during a project

A study that looks feasible during proposal development may cease to be feasible when recruitment is slower than expected, access is withdrawn, a platform changes its API, a collaborator leaves, costs increase, or preliminary data reveal that a measurement strategy is inadequate.

Feasibility evaluation should therefore include contingency planning.

Ask: What are the likely failure points? Which ones would merely delay the project? Which would invalidate the design? What alternative strategy could preserve the research question? At what point should the question itself change?

Generative AI can be particularly useful for generating these “what if?” scenarios.

AI may be better at finding problems than giving a yes-or-no verdict

“Is this feasible?” encourages an oversimplified answer.

A more productive request is: “Identify the assumptions, dependencies, resource requirements, methodological risks, ethical constraints, and failure points that determine whether this study is feasible.”

This reframes AI as a diagnostic assistant.

Empirical work on LLM scientific ideation reinforces the need for this caution. In a large-scale comparison with expert researchers, AI-generated ideas were judged more novel on average but somewhat weaker on feasibility, while the researchers also reported limitations in LLM self-evaluation. The ability to propose a research idea therefore should not be confused with reliable knowledge that the idea can actually be executed.

Feasibility is researcher-specific

The same question may be feasible for a national research consortium and infeasible for a master's student working alone.

A laboratory with specialist equipment, statisticians, established participant panels, multi-year funding, and institutional data agreements operates within a different feasibility envelope from a student with one semester and no budget.

This is not a judgment about the intellectual quality of either researcher. Feasibility is partly a property of the relationship between a question and the environment in which it will be investigated.

That is why formal evaluation frameworks consider not only approach but also whether the investigators and environment have the expertise and resources needed for the proposed work.

When an Excellent Research Question Is Too Large for the Project

The important move was not asking AI whether the study was “possible.” It was identifying what evidence the original question demanded and comparing those demands with the researcher's actual circumstances.

Common Mistakes When Using AI to Judge Research Feasibility

Give AI Your Constraints and Ask It to Find Where the Study Breaks

A useful feasibility conversation begins with specifics rather than optimism.

Tell the model what evidence your question requires and what you actually have: time, population access, likely sample, instruments, expertise, budget, software, equipment, institutional permissions, data, and methodological limitations. Then ask it to identify assumptions and failure points.

The objective is not maximum ambition. It is a question whose intellectual ambition and evidentiary demands fit the study you can conduct rigorously.

Before Deciding That Your Research Question Is Feasible

Frequently Asked Questions About AI and Research Feasibility

A Feasible Question Is One You Can Answer Credibly, Not Merely Study Somehow

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.