Write the research Discussion chapter with ChatGPT

A practical workflow for drafting a dissertation “Discussion” chapter with ChatGPT centers on one principle: accuracy comes from feeding the model your study’s structure in small, controlled pieces and then validating its output against your literature and your own ideas. The core claim is that ChatGPT can produce a discussion that aligns with the literature and results—without inventing details—if the user supplies the right inputs and follows a repeatable prompt-and-check process.

The discussion chapter, as framed here, has three non-negotiable jobs. It must be grounded in the study’s findings from the Results chapter, it must connect those findings to the literature already laid out in the Literature Review, and it must remain tied to the study’s aims and research questions. The chapter is likened to “cooking with your ingredients”: Results provide the ingredients, and the Discussion explains what to “cook” with them—why the findings matter, how they fit (or don’t fit) prior work, and what the study contributes.

To keep the model from going off-script, the transcript recommends breaking the source material into parts rather than pasting everything at once. In the example workflow, the user copies key sections such as the study background, research information, and focus, then separates the Results into multiple text files (three in the example) to reduce the chance of errors caused by large, messy inputs. The same approach is presented as a general best practice: smaller, cleaner inputs make it easier to control what the model uses.

The workflow also emphasizes human control. Instead of outsourcing the entire Discussion immediately, the user is encouraged to generate rough ideas first—then use ChatGPT to generate a draft for comparison. That comparison acts as a validation step: ideas that match the user’s expectations can be treated as stronger leads, while claims that sound too confident—especially contribution-to-knowledge statements—should be reviewed carefully and supported with additional literature.

In the demonstrated case, the model’s draft is described as “not bad at all,” with coverage that tracks key themes from the real discussion section, including attitudes and beliefs and drivers related to English medium instruction (EMI). The output reportedly includes implications and recommendations, with implications largely considered plausible, though the user would likely add more literature support and adjust sections as needed.

Finally, the transcript flags a downstream issue: AI-written text may be detected by AI content tools. The suggested remedy isn’t to ignore the risk, but to reshape the draft—reframing, rephrasing, and “humanizing” the language—so the final chapter reflects the researcher’s voice and reasoning rather than a generic model output.