Gains:
- Ability to use artificial intelligence as a thought partner in sharpening the research question, generating hypotheses and variable definition, and method options
- Ability to obtain quantitative, qualitative and mixed design, sample and validity-reliability options from artificial intelligence and evaluate them according to their field and ethical rules.
- Ability to understand that methodological decisions and ethics committee (IRB) responsibility belongs to the researcher and that the artificial intelligence proposal is only a draft.
After getting to know the literature, it is time to design your own study. Research design is the process of starting with a question and turning it into a measurable, answerable study: sharpening the question, establishing a hypothesis, defining variables, choosing the appropriate design and method, determining the sample, considering validity and reliability. This is the most mentally demanding part of research, and AI can be a powerful thought partner here — generating options, evoking blind spots, comparing alternatives. But the core rule of this unit is clear: Methodological decisions and ethical responsibility belong to the researcher; AI only provides options to evaluate.
From question to hypothesis
A good research question is focused, answerable and original. A common framework is PICO in experimental/clinical studies: Population, Intervention, Comparison, Outcome. In social sciences, the question is generally posed as a relationship between variables. AI helps you narrow down a broad question and suggests several alternative formulations.
A hypothesis is a transformation of the problem into a testable prediction; usually involves a direction (“X increases Y”). Variables are the things you measure: independent variable (which you influence), dependent variable (the outcome you measure), control variables (which you hold constant). AI is useful in sharpening the hypothesis, distilling the variables into operational definitions (measurable definitions), and recalling possible confounders (the third factor that secretly influences the outcome).
Hint: Tell the AI “also give arguments to disprove this hypothesis.” It is valuable to see the weaknesses of your own design early, so that you can learn now rather than during the referee process.
Design, sample and validity
The research design is how data will be collected and structured to answer your question. There are three broad families: quantitative (numerical data, statistical testing—experimental, quasi-experimental, correlational, survey), qualitative (text/observation, meaning and experience—interview, case study, phenomenology, grounded theory), and mixed method (combining the two). AI can compare each pattern with its pros and cons and discuss which one might suit your question better; but the ultimate choice is yours, based on your sources, field tradition, and the nature of your question.
The sample is the subset from your universe that you examine. In quantitative, sample size is calculated by power analysis—the smallest sample required to capture a statistically significant effect; AI reminds the formula and assumptions, but you should confirm the number with a software (like GPower). Validity (do you really measure what you measure) and reliability* (can the measurement be repeated) are quality assurances of design; AI lists threats, you take precautions.
The most critical point is ethics. Any study involving human or animal participants requires ethics committee (IRB) approval: informed consent, confidentiality, nonmaleficence. AI can help you draft an ethics committee application, but it is never a substitute for approval and ethical responsibility. AI cannot say "this design is ethical"; This is decided by your institution's ethics committee.
Mixed methods rationale and preregistration
An important question when choosing a mixed method is why and how to combine the two data types. Simply saying “I conducted both a survey and an interview” is not mixed method; The design must have a logic. Common designs include exploratory sequential (first quantitative, then qualitative to deepen findings), exploratory sequential (first qualitative, then scale development and quantitative testing), and concurrent nested designs. AI can compare these patterns according to your question and discuss which one fits the integration logic; but the choice depends on what your question really requires.
A powerful tool for methodological integrity is preregistration: registering your hypothesis, variables, and analysis plan in an open repository (such as OSF) before data collection. This prevents hindsight hunting (HARKing — making up hypotheses after seeing the result) and increases the credibility of your study. AI can help you structure a draft of the pre-registration form and spot any ambiguities in your plan; However, every decision you record is yours and it is your responsibility to stick to it after recording.
three mini cases
Case 1 — Narrowing the question. A graduate student's question was "How does social media affect young people?" It was so vast and immeasurable. He put AI into a PICO-like framework; The question is "Is there a relationship between daily Instagram usage time and sleep quality in 14-17 year old high school students?" has become. A measurable, limited and answerable question; AI helped narrow it down, the student made the decision.
Case 2 — Confounding variable warning. One researcher hypothesized “online course → low achievement.” He asked the AI about its weaknesses; AI reminded us that socioeconomic status can be a confounder that affects both course type and success. The researcher didn't know this, but he skipped adding it as a control variable to his design. AI served as a reminder; he made the methodological decision himself.
Case 3 — Ethical boundary. A student was designing a survey on a sensitive issue (substance use) and asked YZ, "Will this survey pass the ethics committee?" he asked. AI listed general principles (consent, anonymity, right to withdraw), but the student did not substitute this for consent; The real ethics committee applied and the board requested corrections in two articles. AI helped draft, the board made the decision.
Copiable templates
1) Narrowing the question:
Your role: a methodology consultant. My broad area of interest is: "[topic]". Narrow this down to 3 focused, measurable alternative research questions. For each, specify the sample, key variables, and possible design. These are the options; I will decide.
2) Hypothesis and variable disambiguation:
My research question: "[question]". Suggest me (a) a testable hypothesis, (b) independent/dependent/control variables, (c) operational (measurable) definition for each variable. Also list possible confounding variables. Discuss the options, the decision is mine.
3) Pattern comparison:
Compare quantitative, qualitative, and mixed methods designs for "[question]" in a table: fit, strength, boundary, resource required. Don't say "choose this" for sure; Give me the pros and cons, I'll make the choice.
4) Validity/reliability threat list:
List the following design's threats to internal validity, external validity and reliability and suggest a possible countermeasure for each.Design: [summarize]. These will be my checklist; I make the decision.
Weak prompt / Strong prompt
Weak prompt:
Give me a research design.
There are no topics, questions, areas and restrictions; AI creates a common, useless template.
Powerful prompt:
Your role: experienced methodologist in educational sciences. My question: "The effect of blended learning on 7th grade science achievement". For a quasi-experimental design: discuss appropriate control group strategy, type of measurement instruments, possible confounders, and internal validity threats as OPTIONS. Say what information is needed for power analysis; I will calculate the number with benG*Power. Tick points that require ethical approval.
Design decision
Role of AI
Whose responsibility
Narrow the question
produces alternatives
The researcher chooses
Hypothesis/variable
Draft + reminder
Researcher defines
Pattern selection
compares
The researcher decides
sample size
Formula reminds
Confirmed by software
Ethical approval
draft assistant
Ethics committee only
Common mistakes
- Making the AI choose the pattern. The methodological decision belongs to the researcher; AI suggests, you choose.
- Bypassing the ethics board. AI cannot say "ethical"; Approval comes from the institution's ethics committee.
- Ignoring mixers. If control variables are not planned, the result cannot be interpreted.
- Getting the exact number of samples from AI. Verify power analysis with appropriate software.
- Not narrowing down the question. An immeasurably broad question means a study that cannot be designed.
Caution: A method suggested by AI may not be standard in your field. Compare each recommendation with methodology resources in your field (textbook, consultant, similar publications); AI speaks generally, domain conventions are specific.
In summary
Research design is a chain that goes from sharpening the question to sampling and ethics. AI is a powerful thought partner that produces options in every link of this chain and reminds of blind spots; clarifies the hypothesis, compares patterns, lists threats. But the responsibility for design selection, sample confirmation and especially ethical approval belongs to the researcher and the ethics committee. The shortest rule of thumb: the choice is from AI, the methodological and ethical decision is from you.
Application task
Get your own research idea. With AI, narrow it down to 3 focused questions and choose one. Generate hypotheses, variables and operational definitions for the question you have chosen and correct them according to your field. Ask for a quantitative/qualitative/mixed design comparison and justify which one you chose and why in a paragraph. Tick points that require ethical approval.
checklist
- [ ] Have I narrowed down my research question measurably?
- [ ] Have I clarified the hypotheses and variables with operational definitions?
- [ ] Have I planned confounding variables?
- [ ] Did I choose the pattern for my own reasons?
- [ ] Have I determined the points that require ethics committee approval?