AI for Research Design: Build a Plan You Can Defend

Use AI for research design by giving it a specific question, the evidence you already have, and your practical limits, then asking it to challenge the proposed method. Check its suggestions against sources and expert review before adopting them. fenbs can keep the methodological decisions, source links, and unanswered questions together as the plan develops.

The useful output is a research plan you can explain: what you want to learn, how the observations will answer it, and what they cannot establish. A polished proposal is less valuable than a short plan whose assumptions are visible. The same approach works for a student study or a design team investigating a product problem.

What should you give AI before asking for a research plan?

1. State the decision or knowledge gap. Explain who needs the answer and what would change because of it.
2. Describe the population, setting, and access you actually have. Separate confirmed access from hoped-for recruitment.
3. Supply checked background sources with short notes on what each supports. Mark conflicting findings and missing evidence.
4. List constraints: time, resources, approved tools, privacy boundaries, and the person who will review the method.

For example, a hypothetical college library might want to understand why students abandon its room-booking process. Give the assistant the actual booking steps and approved, anonymous observations. Do not start with “prove that the form is confusing,” which turns a possible explanation into a conclusion before the study begins.

How can AI help sharpen the research question?

Ask for a critique before asking for a rewrite. Have the assistant identify vague terms, hidden causal claims, competing explanations, and information the proposed study cannot collect. Then request two narrower versions, explaining what each version gives up. You choose the question after checking those trade-offs.

For the library, “Why do students abandon booking?” could become “At which steps do first-time users hesitate or stop during observed booking attempts?” That question supports observation of the process. It does not establish how common each problem is across the entire student population or prove that a redesign will fix it.

Keeping those boundaries is part of rigorous design. NIH describes scientific rigor↗ (opens in a new tab) in terms of controlling bias across design, methods, analysis, interpretation, and reporting. Save the chosen question and its limits as a human decision in fenbs, so later drafting sessions do not silently broaden the claim.

How do you compare methods without letting AI choose for you?

Request a comparison tied to your question, rather than a list of fashionable techniques. Every candidate method should say what evidence it produces, what access it needs, and which conclusions remain unavailable. Reject options that depend on resources or participants you cannot obtain.

When using AI for design research, keep observed behavior separate from interpretations of it. An assistant can suggest another explanation for a pause; it cannot turn that suggestion into something a participant said. Ask an experienced researcher to check whether your chosen approach can answer the narrowed question.

What weaknesses should you ask AI to look for?

Give the assistant the proposed recruitment, procedure, measures, and analysis approach. Ask it to name the strongest objection to each, identify the assumption behind that objection, and suggest a feasible check. Require it to distinguish a documented problem from a possibility that needs investigation.

• Selection: who is likely to participate, and whose experience may be missing?
• Measurement: does the measure capture the thing you care about, or only an easy substitute?
• Comparison: what else could explain a difference between groups or conditions?
• Analysis: how will missing observations, exclusions, and unexpected results be handled?

The NIH study-design guidance↗ (opens in a new tab) offers concrete checks, including justified sample sizes, clear inclusion criteria, bias controls, and alternative interpretations. Its context is biomedical research; adapt relevant questions rather than treating every item as a rule for every field. Have a qualified reviewer check statistical calculations and their assumptions.

What should a pilot check before the main study?

Use a pilot to test the procedure: whether instructions are understandable, recruitment criteria work, recording captures the needed evidence, and the session fits the available time. Ask AI to inspect the materials for leading questions or missing steps. Then try the procedure with appropriate people; an imagined participant response is not pilot evidence.

For US research involving people, take questions about review, consent, and data handling to your institution or responsible research lead. OHRP guidance↗ (opens in a new tab) covers human research protections, including consent and protocol review. Use approved, minimized material in AI tools; a design critique rarely needs names, contact details, or raw sensitive records.

How do you preserve the reasons behind the final plan?

For every important choice, record the option selected, the reason, the supporting source, who approved it, and any remaining limitation. Keep rejected options when their reasons will matter later. This is the practical value of an AI brain: the next assistant can read the reasoning instead of reconstructing it from a final document.

Separate planned analyses from ideas that appear after seeing results. The Center for Open Science explains preregistration↗ (opens in a new tab) as recording a research plan in a registry before the relevant work, helping distinguish planned and unplanned analyses. A decision record supports the work but does not replace a formal preregistration when you need one.

Keep the registry reference and any later change notes with the research decisions in fenbs, so the next assistant can check the same record.

How to do this with fenbs

1. Create a free account and add a project for the study, with the question, scope, and reviewer.
2. Record approved design choices in Decisions and rules. Add checked references to Links and listings, with notes explaining what they support.
3. Create tasks for the pilot, unresolved methods questions, and source checks. Put the proposed steps and review criteria in each task plan.
4. Connect a supported assistant through Integrations. Ask it to read the project context before critiquing the plan and save progress before stopping.

fenbs is free for one person, no credit card. It does not run studies, calculate statistical power, or replace a research repository. Use it to maintain the decisions and next steps; keep participant records and formal protocols in the approved systems for your study.

When is the design ready for a human review?

Hand over the question, proposed method, recruitment plan, analysis approach, source list, pilot findings, and a short list of unresolved issues. Label the limitations explicitly. A reviewer should be able to trace each major design choice to evidence or an acknowledged practical constraint, and see which conclusions the study will leave open.

For finding material or organizing references, use the separate guides to AI research search and research bookmarks. For the broader review habit, see using AI at work and the AI at work hub. Keep design review focused on whether the evidence can support the answer you intend to give.