The kind of study you run determines the kind of conclusions you can draw. Picking the wrong framing — or pretending you have a controlled experiment when you actually have a correlational one — will make your discussion section much harder to write honestly.

• A confirmatory finding can be reported as evidence for or against a specific hypothesis
• An exploratory finding is a suggestion for a future study to test, not a conclusion in itself
• The difference matters because anything will look interesting if you go fishing for it. Reporting a fishing-trip finding as if it were a tested prediction is one of the most common mistakes in published science

• Confounds: something other than your manipulation differs between conditions (e.g., one condition was always run in the morning, the other always at night)
• Demand characteristics: participants guess what you expect and shift their behavior accordingly
• Missing baseline: without a comparison, you can't tell whether your "effect" is anything at all
• Too many variables at once: one clear comparison beats five murky ones
• Outcome is unmeasurable: "creativity," "happiness," "engagement" can be hard to operationalize; pick something concrete you can actually score
• The "ceiling" or "floor" trap: if everyone scores 100% (or 0%), you can't see your effect even if it's real

GenAI-built experiments often look correct but mis-record data, skip conditions, or misorder trials. Pilot test with yourself first — and check the saved data file before recruiting any participants.

• Ask for feedback on your experiment before you run it on participants
• Ethics approval looks for potential harms to participants. It does not provide feedback on whether your experiment is well-designed or will actually answer your question. You need to ask for that separately!