ChatGPT

What it does

ChatGPT is a general-purpose AI assistant from OpenAI that accepts text, files, images, and data as input and generates text, code, and structured output in response. It’s built on the GPT-4o family of models (free and Plus tiers) and the o1/o3 reasoning models (Pro tier).

In a research context, ChatGPT is most commonly used for four tasks:

1. Drafting and revising prose — writing first drafts of introductions, methods, and abstracts; improving clarity and flow in existing text; adapting language register for different audiences
2. Explaining and summarizing — breaking down a paper, concept, or dataset in plain language; generating structured summaries of documents you paste in
3. Research code — writing, debugging, and explaining Python, R, or MATLAB code for data processing and analysis
4. Ideation and structuring — outlining arguments, generating hypotheses, or stress-testing a research design

Best for

Researchers who spend significant time writing or who need occasional coding assistance. Most useful in the later stages of a project (writing, revision, reporting) rather than in search or discovery. If your bottleneck is getting words on the page or explaining your methods clearly, ChatGPT offers a low-friction way to move faster.

Pricing

Freemium. Three main tiers:

• Free — GPT-4o mini with limited GPT-4o access. Sufficient for occasional use and short documents
• Plus ($20/month) — Full GPT-4o access, file uploads (PDF, CSV, images), web search, higher usage limits
• Pro ($200/month) — Unlimited o1 and o3 reasoning models; better for tasks requiring extended multi-step reasoning (complex math, detailed code review, long document synthesis)

For most research use cases, Plus is the appropriate tier. Pro is worth considering only if your work regularly involves reasoning-heavy tasks like proof-checking, advanced statistical methodology, or extended code generation.

Strengths

• The most widely supported AI writing assistant — integrations with Word, Google Docs, and VS Code exist
• File upload (Plus+) handles PDFs and datasets directly, so you don’t need to paste content manually
• Strong at prose improvement: tightening overly long sentences, adjusting formality, restructuring paragraphs — tasks where a human editor would also add value
• Python and R code generation is reliable for standard analyses; the generated code is inspectable and editable
• Web search capability (Plus+) retrieves current information, unlike answers from training data alone

Limitations

Do not ask ChatGPT to generate citations. This is the most important caveat for researchers. When asked to cite sources, ChatGPT frequently fabricates plausible-sounding paper titles, author names, and DOIs that do not exist. This has caused significant problems for researchers who included AI-generated citations without verifying them. Use Semantic Scholar, Scopus, or PubMed to find actual papers; use ChatGPT only to help write about sources you’ve already verified.

Additional limitations:

• Training data cutoff — the model’s knowledge has a cutoff date (typically several months before the current date). For recent developments, enable web search or use a retrieval-based tool like Perplexity instead
• No memory of prior sessions by default — each conversation starts fresh; use Projects (Plus+) to maintain persistent context across sessions
• Overconfident tone — ChatGPT rarely hedges appropriately on uncertain claims. Any factual claim it makes should be verified against a primary source before you rely on it
• Privacy — by default, conversations may be used to improve OpenAI’s models. For sensitive research, turn off “Improve the model for everyone” in settings, or use the API with a data processing agreement

How it compares

Related content

• Tool: Claude — alternative general-purpose LLM with stronger long-document handling
• Tool: NotebookLM — for grounded document synthesis without hallucination risk on citations
• Tool: Perplexity — for web-grounded answers with inline source links