How to Use ChatGPT for Thesis Writing

ChatGPT can be useful during a thesis, but asking it to "write my literature review" is one of the least useful ways to use it. A thesis is supposed to demonstrate your judgment: how you define a question, evaluate evidence, choose methods, and defend a conclusion. Handing those decisions to an AI tool weakens the work and may violate your institution's academic integrity policy.

A better approach is to use ChatGPT as a research interface. With the Scite plugin for ChatGPT, ChatGPT can search Scite's literature index, retrieve verifiable papers, and inspect how later research has cited them. That makes it useful for discovery, source evaluation, reference checking, and finding evidence you may have missed—while the argument and writing remain yours.

The best way to use ChatGPT for a thesis

Before getting into specific prompts, keep four principles in mind:

  1. Use AI for research tasks, not authorship. Let it help you search, organize, compare, and check. Write the thesis yourself.
  2. Ask for evidence, not just answers. Request titles, authors, publication years, DOIs, and links that you can inspect.
  3. Work in small, reviewable steps. A focused search or reference check is easier to verify than a generated chapter.
  4. Keep a source trail. Record why you included each paper, what claim it informs, and what you verified in the original text.

Your university, department, journal, or funder may have its own rules for disclosing AI use. Check those rules before uploading unpublished writing or using AI-generated material in your thesis.

How Scite works inside ChatGPT

The Scite plugin includes a connector backed by the Model Context Protocol, or MCP. MCP is an open standard that lets an AI assistant call an external tool during a conversation. In this case, ChatGPT can send a literature query to Scite and use the returned papers and citation data in its response.

Scite combines scholarly metadata with citation statements extracted from full text. Its Smart Citations show the context around a citation and classify whether the citing statement supports, contrasts with, or mentions the cited work. That distinction matters because a citation count tells you that a paper received attention, but not how later researchers discussed it.

The connection does not make every ChatGPT response automatically correct. Scite returns records from its index, but ChatGPT still interprets and summarizes the results. Open the sources, confirm the metadata, and read the relevant passages before relying on them in your thesis.

How to add Scite to ChatGPT

  1. Open the Scite plugin in ChatGPT.
  2. Select the option to add or install Scite.
  3. Start a new ChatGPT conversation and type @Scite to use the plugin in your prompt.

If you use ChatGPT through a university or organization, an administrator may control which plugins are available. Ask your workspace administrator if you cannot install or invoke Scite.

You can try up to 10 Scite MCP requests without creating a Scite account or starting a trial. Continued use requires signing in with an eligible Scite plan. See current Scite plans and MCP allowances for details.

ChatGPT can occasionally disconnect tools between sessions. If Scite stops responding, open the menu next to Scite and select Reconnect, then include @Scite in your next prompt.

Turn a broad topic into a search strategy

Early thesis searches often fail because the topic is still too broad. Instead of asking ChatGPT for "papers about social media and mental health," use Scite to test the boundaries of the question and identify useful search terms.

Help me turn this thesis topic into a literature search plan: [topic]. Identify the main concepts, synonyms, related terms, likely exclusions, and a set of Boolean searches. Then use @Scite to test the searches and explain which versions return the most relevant peer-reviewed literature.

This gives you a starting strategy that can be documented and refined. For a systematic or scoping review, follow the reporting standard required by your field and run the final strategy in the appropriate subject databases as well.

Find papers that match a specific claim

When you need evidence for a precise statement, describe the population, exposure or intervention, outcome, study design, and date range where relevant. Ask ChatGPT to explain why each result matches instead of returning a long, unranked bibliography.

Use @Scite to find peer-reviewed studies on [specific claim]. Prioritize research involving [population or context] published since [year]. For each result, provide the title, authors, year, DOI or stable link, study design, and one sentence explaining its relevance. Flag any result you could not verify.

Scite searches across indexed full text and publication metadata, which can surface relevant methods, findings, and claims that do not appear in a paper's title.

Build an evidence map before writing

An evidence map helps prevent a literature review from becoming a sequence of disconnected paper summaries. Ask ChatGPT to organize sources around questions you will later evaluate yourself: Where do studies agree? Where do they differ? Which populations and methods are underrepresented?

Using these papers and @Scite, create an evidence map for my thesis question. Group the studies by theme or proposed mechanism. For each study, record the population, method, main finding, limitations, and relevant supporting or contrasting citation context. Separate information retrieved from sources from your interpretation.

Export the useful fields into your own notes, reference manager, or evidence table. The goal is not to generate prose for your chapter; it is to make the structure of the evidence easier to inspect.

Check references and claims in a draft

References accumulate errors during months of drafting: a wrong year, a misspelled author, a DOI copied from another paper, or a citation that does not support the sentence attached to it. You can paste a short section and ask ChatGPT to use Scite as a first-pass reference check.

Review the references in this section with @Scite. For each citation, verify that the paper exists and confirm the title, authors, year, journal, and DOI. Compare the claim in my sentence with the evidence you can retrieve. Flag missing metadata, retractions or editorial notices, and claims that need manual review. Do not rewrite the section.

This is a screening step. Always open the original paper and confirm the cited passage. For a full manuscript-level audit, use Scite Reference Check, which is designed to evaluate an entire reference list for retractions, editorial notices, and contrasting citation evidence.

Follow the citation trail around a key paper

A foundational source is a starting point, not the end of the search. Scite can help you look forward from a paper to later work that cited it and identify citation statements that support, contrast with, or simply mention its findings.

Use @Scite to trace the citation history of this paper: [title or DOI]. Find important later studies that cite it, with separate sections for supporting, contrasting, and mentioning citation statements. Summarize the context of each citation and link to the citing paper.

Smart Citation labels apply to individual citation statements, not to an entire paper. Read the surrounding text before concluding that a study as a whole "proves" or "disproves" another.

Look deliberately for counterevidence

Confirmation bias is easy to miss when you already have a working hypothesis. Add an explicit counterevidence step before finalizing a literature review or discussion chapter.

Challenge this thesis claim using @Scite: [claim]. Find credible studies with null, mixed, or contrasting findings. Compare their populations, methods, and limitations with the studies that support the claim. Tell me what evidence would make the claim more precise or appropriately qualified.

This can reveal boundary conditions, differences between populations, measurement problems, or an unresolved debate that deserves space in your thesis.

Find methodological precedents

ChatGPT can also help locate papers that used a particular instrument, dataset, experimental design, statistical method, or qualitative framework. That is often more useful than asking it to design a method from scratch.

Use @Scite to find recent papers that studied [research question] using [method, instrument, dataset, or framework]. For each paper, identify the sample, procedure, analysis approach, and any limitations reported by the authors. Include a DOI or stable link so I can review the methods section.

Use these papers as precedents to evaluate your own design with your supervisor. Do not treat a generated methods summary as a substitute for reading the paper or understanding the assumptions behind the method.

Organize promising sources as you go

Long research chats become difficult to audit. When a search produces useful papers, save them to a Scite Collection or your reference manager instead of relying on the conversation history. Scite Collections can keep related papers together and help you monitor new citations, editorial notices, and supporting or contrasting evidence.

You can also ask ChatGPT to produce a short search log after each session: the queries used, filters applied, papers retained, papers excluded, and questions that remain open. That record is useful when you return to a chapter weeks later or need to explain your search process to a supervisor.

Prepare for supervisor meetings and your defense

Once you have verified the underlying sources, ChatGPT can help you pressure-test your reasoning. Give it your claim and an evidence table, then ask for questions rather than prose.

Based only on this verified evidence table, list the strongest questions a thesis supervisor or examiner might ask about my interpretation. Focus on conflicting findings, methodological weaknesses, generalizability, and missing literature. Do not draft answers for me.

Answering those questions yourself can reveal where the argument needs more evidence or a narrower claim.

What ChatGPT should not do for your thesis

Do not use ChatGPT to fabricate citations, paraphrase papers you have not read, hide uncertainty, or produce a chapter that you submit as your own work. Do not assume a polished reference is real because it contains plausible authors, a journal title, and a DOI-shaped string.

The strongest workflow keeps responsibility in the right place: ChatGPT helps you ask and organize questions, Scite retrieves literature and citation context, and you read the sources, make the decisions, and write the thesis.

Frequently asked questions

Can ChatGPT write a thesis?

ChatGPT can generate text, but submitting AI-written work may violate academic integrity rules and does not demonstrate the judgment a thesis is meant to assess. Use it for bounded, reviewable research tasks and follow your institution's disclosure policy.

Can ChatGPT find academic sources?

Yes, but citations generated from model knowledge or general web results still need verification. The Scite plugin lets ChatGPT search Scite's indexed scholarly literature and return records with metadata and citation context that you can inspect.

Can ChatGPT verify references?

With Scite connected, ChatGPT can help confirm whether a paper exists, check bibliographic metadata, and surface relevant citation context or editorial notices. It remains a first-pass check; verify important references in the original publication and use Scite Reference Check for a complete manuscript audit.

Do I need a Scite account to try the ChatGPT plugin?

No. You can try up to 10 Scite MCP requests without creating an account, entering a credit card, or starting a trial. Continued use requires authentication and an eligible Scite plan.

Get started

Add Scite to ChatGPT and use @Scite in a new conversation. Start with one focused research question, verify every source you plan to cite, and keep the final analysis and writing your own.