AI for Research Methodology: Plan and Check Your Approach

Use AI for research methodology to compare approaches to a defined question and challenge the assumptions behind each one. Ask for alternatives, limitations, and the evidence each approach could produce; have a supervisor or qualified researcher review the resulting plan. A fluent methods section is not proof that a study can answer its question.

A tool like fenbs keeps approved methodological decisions and their rationale available across assistant sessions. That helps stop a later draft from quietly changing the study design because a different method is easier to describe.

How should you begin the planning session?

1. State the research question and the kind of answer you need: an estimate, an explanation, an interpretation, or another defined contribution.
2. List the participants, records, materials, or observations you can realistically access. Distinguish confirmed access from hoped-for access.
3. Provide the project constraints, including time, skills, available equipment, institutional requirements, and supervisor instructions.
4. Ask AI to compare a small number of plausible approaches and explain what each could and could not establish.
5. Take the comparison to a supervisor or methods specialist before treating a generated plan as approved.

Include relevant disciplinary guidance or examples you have actually read. An assistant given only a topic may fill in your missing constraints with a conventional design that does not fit your purpose. Ask it to list unanswered planning questions before recommending an approach.

The broader guide to using AI at work covers keeping responsibilities clear. Here, the specific responsibility is defending why your method can support the conclusion you intend to draw.

What is the difference between methods and methodology?

For planning purposes, treat methods as the practical procedures you use and methodology as the reasoning that connects those procedures to your question. Conducting interviews is a method. Explaining why interviews suit an inquiry into how participants describe an experience is part of the methodological argument.

Terminology varies between disciplines, so follow your program’s conventions. Avoid asking AI to decorate a methods list with abstract labels. Each important term should explain a real choice, such as what counts as evidence or how you will interpret it.

A useful check is to ask: “If I changed this choice, what kind of answer would become weaker or impossible?” If the answer never connects back to the question, you may have a description of procedures without a defensible rationale.

How can AI compare approaches without choosing for you?

Give each candidate the same comparison criteria. A table can make tradeoffs visible, but avoid a single numerical score that hides an impossible requirement. No amount of convenience makes unavailable data available.

This hypothetical example does not imply that these approaches are interchangeable. A record of appointments cannot explain every reason a student stayed away. An interview account cannot by itself estimate how common that experience is across a whole campus.

Save the human choice and the rejected alternatives in fenbs Decisions and rules, so later drafting keeps the methodological boundary intact.

Which assumptions should you ask AI to challenge?

Ask what must be true for the planned evidence to support your interpretation. Consider how participants or records enter the study, whether a measure reflects the idea you care about, and what other explanation could produce the same pattern. Make the assistant name the consequence of each weakness, not merely call it a limitation.

• Selection: who or what would be excluded, and could that change the answer?
• Measurement: what does each observation represent, and what might it miss?
• Comparison: are you comparing relevant groups, periods, texts, or conditions?
• Analysis: what decisions must be specified before you see the results?
• Interpretation: which conclusions would go beyond the evidence collected?

In its biomedical context, NIH guidance on rigorous study design↗ (opens in a new tab) emphasizes justified approaches, appropriate comparisons, bias controls, and transparent reporting. Use guidance suited to your discipline; do not transplant an experimental checklist unchanged into every qualitative or humanities project.

Review this proposed design against the research question and constraints. List assumptions, likely sources of bias, unanswered feasibility questions, and conclusions the design would not support. Separate established problems from possible concerns. Suggest questions for my supervisor, not approvals. Do not invent citations or sample-size justifications.

What needs human review before work starts?

Review access, consent, privacy, recruitment, and analysis with the appropriate people. For a study involving people or identifiable private information, ask your institution which review process applies before collecting data. The HHS Office for Human Research Protections decision charts↗ (opens in a new tab) provide a US starting reference; an AI answer cannot make an institutional determination.

Use hypothetical or appropriately de-identified examples during planning, subject to your institution’s rules. Do not paste restricted records into an assistant to make a comparison easier. NIST’s generative AI risk profile↗ (opens in a new tab) includes data privacy among the risks that organizations need to manage.

Check every external source used to justify the design. Ask a qualified specialist about sample size or analytical choices when your expertise is insufficient. If the supervisor changes a central assumption, revise the connected recruitment, measurement, and analysis choices as well.

Keep approval status and open questions in the fenbs project record, with a person’s decision clearly distinguished from an assistant’s suggestion.

How to do this with fenbs

1. Create a free account and add your study as a project with the question, constraints, and current planning status.
2. Add tasks for comparing approaches, checking feasibility, and obtaining the reviews required by your institution. Give each task a concrete next action.
3. Record the supervisor’s methodological choices in Decisions and rules, including the rationale and unresolved limits. Keep proposed changes open until a person decides.
4. Connect your assistant through Integrations. Ask it to read the project context before drafting and save completed checks and remaining questions in Where we left off.

fenbs does not approve research ethics, perform statistical analysis, or replace a protocol repository. Keep the protocol and research data in the systems your institution requires. Use fenbs when the surrounding decisions and follow-up work need a shared home.

How do you prevent the plan from drifting?

Before each revision, compare the question, design, evidence, and intended conclusion. A change in one can affect the others. If recruitment becomes narrower, for example, revisit the claims you expect to make rather than leaving the original ambition in the introduction.

Keep a short record of each approved change and its reason. At handoff, state what remains unapproved and which version people should review. The AI brain guide explains how preserving this work context helps another assistant continue without guessing.

Literature discovery belongs in the research search guide. For related practical workflows, visit the AI at work hub. The method still needs to earn its place by answering your actual research question.