How to Use AI for a Literature Review (Real Citations)

Your supervisor wants Chapter Two by the end of the month. You have a topic, a vague idea of what's been written about it, and no idea where to start. So you ask ChatGPT for "10 sources on mobile money adoption in Nigeria" and get a neat list of papers, authors and years.

Then you try to find them. Half don't exist.

This is the most common way students get into trouble with AI in research. General chatbots can invent citations that look completely real. Used the right way, though, AI can make your literature review faster and better. This guide shows you which tools find real papers, a step-by-step workflow from topic to written chapter, and how to make sure every citation is genuine.

Why ChatGPT Alone Is the Wrong Tool

General chatbots like ChatGPT, Gemini and Claude generate text that sounds right. When you ask for sources without giving them any, they can produce references with real-sounding authors, titles and journals that don't exist. Researchers call these hallucinated citations, and they've even been found in papers submitted to major conferences.

The fix is simple: use research tools that search real paper databases to find sources, then use general chatbots only to work with papers you've already found and checked.

The Tools That Find Real Papers

These tools search indexed databases of published research, so the papers they return exist. You still need to read and check them.

Semantic Scholar. A free academic search engine covering over 200 million papers, with short AI summaries for many of them. A strong free starting point for any topic.

ResearchRabbit. Free. Add a few papers you trust and it maps related work, showing who cited whom. Great for finding the key papers in your area.

Consensus. Answers research questions with evidence from peer-reviewed papers, and shows whether studies tend to agree. Useful for yes-or-no questions such as "Does social media use affect student grades?" Has a free tier.

Elicit. Finds papers and pulls the same details from each one into a table, such as method, sample size and findings. Good for comparing many studies. Has a free tier.

Scite. Shows whether later papers support or contrast a study's claims. Useful for checking a source is still accepted.

Gemini Notebook. Once you've collected your PDFs, it lets you ask questions across all of them, with citations to the exact passages. Our guide to studying with Gemini Notebook shows how to set it up.

You don't need all of them. Semantic Scholar plus ResearchRabbit plus Gemini Notebook is a free stack that covers most final year projects.

A Step-by-Step Literature Review Workflow

Step 1: Sharpen your research question

A vague topic gives you a vague review. Use a chatbot to narrow it, but write the final question yourself.

My final year project topic is [topic]. Suggest five narrower research questions a
undergraduate could answer in one semester, and for each, the key concepts and search
terms I should use. Do not suggest specific papers.

Step 2: Find your core papers

Search Semantic Scholar and Consensus with your key terms. Filter by year so you include recent work, and look for review articles first. They summarise a field and point to the important studies.

Aim for five to ten strong papers to start. Download the PDFs you can access, including through your university library.

Step 3: Map the research

Add your core papers to ResearchRabbit to find the papers they cite and the newer papers that cite them. This is how you find the key studies you'd otherwise miss.

Step 4: Read and extract

For each paper, record the same details: author and year, aim, method, sample, key findings and limitations. Elicit can do this in a table, or you can keep your own spreadsheet. Read at least the abstract, method and conclusion of every paper yourself.

Step 5: Find themes and gaps

Upload your PDFs to Gemini Notebook and ask questions across them:

Using only my sources, group these studies into three to five themes. For each theme,
list which studies support it, where they disagree, and what none of them has studied.
Cite every claim.

The gaps are your project's justification: the thing your study adds.

Step 6: Write it yourself

Write your chapter theme by theme, not paper by paper. Compare and contrast studies; don't just summarise each one in turn. Use AI as an editor, not an author:

Here is a paragraph from my literature review. Point out where my argument is unclear,
where I summarise instead of compare, and any claim that needs a citation.
Do not rewrite it.

For more research systems like this, see the Research Workflows.

How to Verify Every Citation

Before any source goes into your chapter, run this check:

Find the paper yourself. Search the exact title in Semantic Scholar or Google Scholar. If it doesn't appear, don't use it.
Check the details. Authors, year, journal and DOI should match exactly.
Read the part you're citing. Make sure the paper actually says what you claim. AI summaries can misstate findings.
Use a reference manager. Tools like Zotero or Mendeley store your sources and format citations in APA, Harvard or your department's style, which avoids typing errors.

A single fake reference can raise questions about your whole project, so this step isn't optional.

Using AI Within Your Institution's Rules

Rules on AI in final year projects vary by university and even by department. Many allow AI for finding and summarising sources but not for writing your chapters, and some ask you to declare any AI use.

Check your department's policy and ask your supervisor if you're unsure. Keep notes of how you used AI, in case you're asked. Our guide on using AI to study without getting flagged for cheating explains the general principles.

Common Mistakes to Avoid

Asking a chatbot for sources. This is how fake citations get into projects. Use research search tools instead.

Citing papers you haven't read. An AI summary is not the paper. Read what you cite.

Summarising instead of synthesising. A good review compares studies and identifies themes and gaps, not a list of summaries.

Ignoring recent work. Filter for the last five years as well as the classic studies, so your review is current.

Leaving references to the end. Add each source to your reference manager as you find it.

Worked Example: From Topic to Themes

Here's how the workflow plays out for a real-sounding project: "Mobile money adoption among small traders in Nigeria".

Sharpen the question. A chatbot suggests narrower angles. You choose: "What factors influence mobile money adoption among market traders in south-west Nigeria?"

Find core papers. On Semantic Scholar, you search "mobile money adoption small business Nigeria" and "financial inclusion mobile payments Sub-Saharan Africa", filter to the last ten years, and save eight papers, including two review articles.

Map the field. In ResearchRabbit, your eight papers lead you to studies from Kenya and Ghana that are cited again and again. You add four of them as comparison studies.

Extract. In a spreadsheet, you record each study's country, method, sample and key findings. A pattern appears: most studies use surveys, and few focus on market traders specifically.

Find themes. In Gemini Notebook, you ask for themes across your sources, with citations. You get four: trust and security, cost and fees, network reliability, and digital literacy.

Spot the gap. Most studies are about urban businesses or other countries. Few look at traders in open markets in south-west Nigeria. That gap is your project's justification.

Notice that AI helped at every stage, but every paper was real, found in a research database and read by you.

How to Structure Your Literature Review Chapter

Departments vary, so follow your handbook first. If you have freedom, this structure works for most undergraduate projects.

Introduction. What the chapter covers, how you searched for sources and how the chapter is organised.
Key concepts. Short definitions of the main terms, such as "mobile money" and "financial inclusion", with sources.
Theoretical framework. The theory or model your study uses, if your department expects one, and why it fits.
Themes. One section per theme. In each, compare what studies found, where they agree and where they disagree.
Gap in the literature. What hasn't been studied, and how your project addresses it.
Summary. A short paragraph linking the review to your research questions.

Gemini Notebook can help you check the balance of each section. Ask: "Using my sources, which theme has the least evidence, and which studies support it?"

Phrases That Show You're Comparing, Not Listing

The difference between a weak and a strong review is often the language that connects studies. Instead of "Author A said… Author B said…", use phrases that show relationships:

"Similarly, …" or "This finding is consistent with …"
"In contrast, …" or "However, … found the opposite among …"
"While most studies focus on …, few examine …"
"These differences may reflect … such as sample size or location."
"Taken together, these studies suggest …"

These phrases force you to think about how studies relate, which is exactly what supervisors are marking.

A Two-Week Literature Review Plan

If your deadline is close, this plan gets you from topic to a first draft in two weeks of focused work.

Day - 1 to 2

Task: Sharpen your research question and agree it with your supervisor

Day - 3 to 4

Task: Search for core papers and review articles; save 8 to 10

Day - 5 to 6

Task: Map the field in ResearchRabbit; add 10 to 20 more papers

Day - 7 to 9

Task: Read and extract details from every paper into a table

Day - 10

Task: Find themes and gaps using Gemini Notebook; plan your sections

Day - 11 to 13

Task: Write the chapter yourself, theme by theme

Day - 14

Task: Verify every citation, format references and get AI feedback on clarity

For a fuller set of research systems, the Research Workflows break each stage into repeatable steps.

Getting Useful Feedback From Your Supervisor

AI can speed up your literature review, but your supervisor's feedback is what shapes it into a strong chapter. Make the most of their time.

Share a plan before a full draft. Send your research question, list of themes and key sources first. It's much easier to fix direction early than to rewrite a finished chapter.

Ask specific questions. Instead of "Is this okay?", ask "Are these four themes the right ones?" or "Is my gap convincing?"

Be open about your AI use. Tell your supervisor which tools you used and how, for example "I used Semantic Scholar to find sources and Gemini Notebook to identify themes, and I wrote the chapter myself." It builds trust and avoids misunderstandings.

Keep a feedback log. Note each comment and how you addressed it. When you send the next draft, include a short summary of changes.

Bring evidence. If your supervisor questions a claim, be ready to show the paper and the exact passage. Your extraction table makes this easy.

Final Thoughts

AI can cut weeks off your literature review, but only if you use the right tools for each job. Research search tools find real papers, Gemini Notebook helps you see themes across them, and general chatbots help you edit what you've written.

The one non-negotiable is verification. Find, check and read every source yourself, and you'll get the speed of AI without the risk.

If you're aiming for a postgraduate place after your project, Nexframe Opportunities lists checked scholarships and fellowships, and our guide to writing a scholarship essay with AI helps you apply.

Frequently Asked Questions

Can ChatGPT find sources for my literature review?

Not reliably. When you ask a general chatbot for sources, it can invent citations that look real. Use research tools that search real databases, like Semantic Scholar, Consensus or Elicit, then verify every paper.

What is the best free AI for a literature review?

Semantic Scholar for finding papers, ResearchRabbit for mapping related research and Gemini Notebook for asking questions across your PDFs. All three have free options.

How do I know if a citation is fake?

Search the exact title in Semantic Scholar or Google Scholar. If you can't find it, or the authors, year or journal don't match, treat it as fake and don't use it.

Can I use AI to write my literature review chapter?

Most universities expect you to write it yourself, and many require you to declare AI use. Use AI to find, organise and edit, and check your department's policy.

How many sources do I need for a final year project literature review?

It depends on your department and topic. Many undergraduate projects use roughly 20 to 40 sources, but ask your supervisor for their expectation.

What's the difference between Elicit and Consensus?

Elicit is best for extracting the same details from many papers into a table. Consensus is best for quick evidence-based answers to a specific research question.

Can I use Gemini Notebook for my literature review?

Yes. Once you've collected real papers, upload them to Gemini Notebook to ask questions across your sources, find themes and spot gaps, with citations to the exact passages.

How recent should my sources be?

Aim for mostly recent work, often from the last five to ten years, plus any classic studies that defined the field. Your supervisor can confirm what your department expects.

What if I can't access a paper?

Check your university library first. Many papers also have free versions on the author's page or in open-access repositories, which Semantic Scholar often links to.