#6 ChatGPT Limitations in Academic Research—What You Need to Know

ChatGPT can produce plausible-sounding answers that are wrong, fabricated, or out of date—making it risky to treat outputs as research-ready without verification. A key example involves a request for a “completely new” research topic linking social media and mental health. ChatGPT suggested a specific angle—social media driven peer pressure and depressive symptoms among teenagers—while implying it had never been studied. That claim doesn’t hold up: the topic has already generated “hundreds” of papers, including a thesis by a PhD student referenced by the instructor. The takeaway is blunt: anything ChatGPT provides—especially novelty claims—needs independent checking before it shapes a research plan or argument.

Numbers and statistics pose a second, high-stakes problem. ChatGPT may supply figures with citations, but those citations can still be incorrect or misleading. The transcript emphasizes that researchers should not copy statistics directly into papers. Instead, they should require credible sourcing and then verify each link themselves. In the example workflow, ChatGPT returns statistics paired with sources, and the responsibility shifts to the user to open each link and confirm the numbers match the underlying evidence.

A third limitation targets literature reviews and reference lists. ChatGPT can generate references that look legitimate but don’t exist. When asked to write a short literature review on exercise and blood pressure with five references, it produced a set of citations; one link, when checked, failed to resolve as a real paper. Searching in Google Scholar turned up nothing, indicating the reference was fabricated. The practical guidance: if ChatGPT is used for any part of a literature review, every citation must be opened and validated as a genuine publication.

Out-of-date knowledge is another recurring concern. The transcript notes that the free version of ChatGPT may lag behind recent developments because it only “knows things up to a certain point in time.” The instructor demonstrates this by asking how current ChatGPT is; it claims knowledge up to November 2024, while the current month is December—meaning it misses information published in the last few weeks. For research that depends on the latest studies, trends, or rapidly evolving events, relying on ChatGPT alone can quietly introduce staleness.

Overall, the message is not to avoid AI entirely, but to use it as a starting point rather than an authority. Novelty claims, statistics, and citations require verification; and currency should be checked against other sources. Tools like Ref-n-write are positioned as support for the verification and writing workflow, including referencing, plagiarism checking, proofreading, paraphrasing, and an academic phrase bank.