Academic Writing with ChatGPT (Part 1 - Beginner)

Academic writing with ChatGPT is less about copying finished text and more about steering the model through careful prompting, structured workflows, and controlled “memory” so outputs match a specific thesis or literature review plan. The core message: use ChatGPT to generate ideas, outlines, and drafts as starting material, then edit heavily, verify facts, and format the work to look like the student’s own voice—because wholesale copy-paste triggers plagiarism/AI-detection risk and produces generic, hard-to-defend writing.

The session begins with practical guidance on which ChatGPT versions matter for academic work. ChatGPT 3.5 is positioned as a beginner-friendly option with a knowledge cutoff around September 2021, while ChatGPT 4 extends coverage to roughly March 2023, and “GPT-4o” (referred to as “4 40”) reaches about October 2023. Multimodal capability is highlighted as a major upgrade: text plus image, audio, and video-related features (with “Sora AI” mentioned as upcoming). The instructor also warns that free tiers may be slower during peak usage and that message limits can change over time.

A second pillar is prompt engineering—turning a vague request into a detailed task. The guidance emphasizes giving ChatGPT enough context (goal, audience, role/persona, tone, format, and examples) so it can produce more accurate and usable academic output. Role-play is encouraged (e.g., “act as a professor” or “act as a reviewer”), along with specifying output structure such as tables, bullet points, or email formats. The session also recommends batching tasks efficiently: instead of one request per message, combine multiple sub-tasks in a single prompt (e.g., analyze, summarize, categorize, then draft), while noting that splitting into multiple prompts can help if results degrade.

To keep outputs consistent across a long writing process, the session introduces “memory” as a way to store reusable preferences—like a thesis structure template. A demonstration shows saving a thesis outline into memory (e.g., title page, abstract, acknowledgements, table of contents, introduction, literature review, methodology, results, discussion, conclusion) and then reusing it later so ChatGPT can generate section-by-section drafts aligned to that plan. The instructor cautions that saving incorrect or overly broad instructions can “corrupt” future outputs, so memory should be set deliberately and updated when needed.

For literature reviews, the session outlines an 11-step approach built around keywords and synthesis rather than copying: brainstorming keywords, forming a research problem statement, creating an outline, optionally defining key terms, then writing a critical review that is coherent, well-structured, and properly referenced. “Coherence” is treated as a skill: paragraphs must link logically, not just summarize separate papers. Critical review is framed as identifying both supporting and contradicting findings and using citations to prove claims. The session repeatedly stresses verification: rely on multiple sources/tools, use citation tools or plugins for accurate referencing, and never trust AI-generated citations blindly.

Finally, the session provides a hands-on workflow for building a research proposal and literature review around an example topic (virtual reality and mathematics education for elementary learners). It demonstrates generating keywords, proposing research gaps, drafting research objectives and research questions, and sketching hypotheses and study design elements (e.g., experimental vs traditional groups, pre-test/post-test, and statistical comparisons). The overarching takeaway is that ChatGPT becomes useful when treated like a research assistant that needs direction, constraints, and human editing—not a copy-and-paste author.