Research With ChatGPT: Analyzing Hypothesis Results in Discussion

A significant relationship in a study isn’t enough for a strong discussion section—researchers also need a credible argument for why results were significant or why they turned out insignificant. The core problem addressed here is the common early-career snag: after finding an insignificant effect (even when a hypothesis predicted significance), it becomes difficult to explain the mismatch in a way that fits the study’s theory, context, and measures.

The discussion section is framed as a structured exercise. First, researchers briefly restate the study’s objectives or rationale. Next, each hypothesis is paired with the result—supported or not supported. Then comes the critical work: linking each hypothesis to prior research, and explaining why the observed outcome aligns or diverges from what earlier studies found. That explanation must also be grounded in the study’s context—such as higher education, manufacturing, or services—and tied back to the theory used in the literature. The emphasis is on generating “reasons” that can be placed into the discussion, not just listing outcomes.

To help generate those reasons, the session argues for using AI tools—while warning that AI should not be used blindly. The workflow starts with knowing how the discussion section should be written, so the researcher knows what to search for and where the output will go. With that structure in mind, AI can be used to brainstorm plausible explanations for an insignificant relationship.

A concrete example is given: transformational leadership shows an insignificant impact on leader–member exchange (LMX). AI is prompted to produce possible reasons, and one line of explanation that emerges is cultural context. The argument is that transformational leadership—especially its emphasis on individualized consideration—may not translate into stronger leader–member exchange in cultures where expectations and interpersonal dynamics differ.

To make the cultural argument specific rather than generic, the session uses Hofstede’s cultural typology for Pakistan. Pakistan is described as high on power distance, achievement motivation, uncertainty avoidance, and collectivism. Those cultural traits are then used to justify why individualized leadership behaviors might resonate differently, potentially weakening the expected relationship with LMX.

But the session stresses that arguments alone aren’t enough; they need references. The workflow therefore shifts from AI-generated reasoning to evidence gathering. AI tools like Elicit and Perplexity are used to convert the argument into searchable questions and to surface studies that connect transformational leadership effects to collectivism or cultural moderation. The output is then checked for fit: if an AI suggests a reason that contradicts the study’s actual sample size or measurement design (for example, claiming “small sample size” when the study has 800 participants), that reason is discarded. The researcher is also advised to ensure the argument matches the measurement structure—such as the number of items used to assess LMX or transformational leadership.

Finally, the session recommends using Google Scholar with targeted keywords (e.g., transformational leadership and collectivism/collective goals) to locate and read the relevant papers, then adapt their arguments to the study’s discussion. The takeaway is a repeatable method: use AI to generate candidate explanations, validate them against study details, and anchor them in scholarly references that match the theoretical and cultural context.