AI-Powered Literature Reviews: How Data2Paper Generates Research Reports from a Topic

If you have ever stared at a blank document titled "Chapter 2: Literature Review," you know the feeling. You have a research topic, maybe a handful of papers you have already read, and a vague sense of which themes matter. But between that starting point and a finished literature review sit dozens of hours of searching, reading, filtering, organizing, and writing.

Data2Paper's Research Report feature compresses that process into a single pipeline. You enter a topic, pick a language, and wait about 30 minutes. What comes back is not a summary paragraph — it is a complete, citation-backed literature review with a bibliography, source metadata, and downloadable files in PDF, Word, and LaTeX.

This post walks through what happens behind the scenes, what you actually receive, and how to get the most out of it.

What you put in

The input is a research topic or question, up to 2,000 characters. That is it — no data files, no uploaded papers, no pre-compiled reading lists.

The more specific you are, the better the output. Compare these two inputs:

• Vague: "AI in education"
• Specific: "The impact of large language models on academic writing pedagogy in undergraduate humanities courses, 2022-2025"

The second version gives the pipeline more to work with: a clear scope, a time range, a disciplinary focus, and a specific technology thread. The system will refine your topic further in its first stage, but starting specific saves it from having to guess your intent.

You also choose an output language from seven options: Chinese, English, Japanese, Korean, French, German, or Spanish. The entire report — section headings, synthesis prose, citation formatting — will be generated in that language. Internal processing uses English for tool compatibility, but the final deliverable is fully localized.

What happens behind the scenes: four stages

Once you submit, the pipeline runs four sequential stages. You will see the job status update in your dashboard as it progresses through each one.

Stage 1: Research question refinement

The system takes your raw topic and turns it into a structured research plan. This includes:

• Refining the topic into specific, searchable research questions
• Expanding 5 to 12 keyword terms with synonyms and related concepts
• Defining a search strategy with boolean queries, year ranges, and inclusion/exclusion filters
• Identifying 3 to 6 expected subtopics for the final report structure

This stage is pure planning. No searches happen yet — the system is building a map before it starts exploring. The output is a research_plan.json file that guides everything downstream.

Stage 2: Literature search and bibliography

Using the search strategy from Stage 1, the system retrieves relevant academic literature. It aims for 15 to 30 sources, each with structured metadata:

• Title, authors, year, venue, DOI, and URL
• An abstract snippet explaining the source's relevance
• Which subtopic from the research plan it maps to
• A verification status flag

Every bibliography entry is cross-validated between the BibTeX file and the source metadata to prevent orphaned citations or phantom references. If a source cannot be verified through live search, it is flagged — not silently included.

The outputs are references.bib (a standard BibTeX file you can import into Zotero, Mendeley, or any citation manager) and sources.json (structured metadata for each source).

Stage 3: Thematic synthesis

This is the analytical core. The system reads all retrieved sources and organizes them by theme rather than listing them one by one. The synthesis identifies:

• Major themes across the literature
• Points of agreement and disagreement between sources
• Research gaps — what the literature has not addressed yet
• Cross-cutting observations that span multiple themes

Every claim in the synthesis is backed by a citation. The output is synthesis.md, a markdown file that serves as the analytical backbone of the final report.

This thematic approach matters because a good literature review is not a list of paper summaries. It is an argument about what the field knows, where it agrees, where it disagrees, and what remains open.

Stage 4: Report compilation

The synthesis, bibliography, and research plan are compiled into a formatted LaTeX document. The report follows academic conventions:

• Title page
• Abstract (150 to 250 words)
• Introduction with research context
• Three to five thematic sections (derived from the synthesis)
• Discussion of research gaps and implications
• Conclusion
• References in APA 7th edition format

The body text targets 1,500 to 4,000 words. Every non-trivial claim includes an inline citation — the system is explicitly instructed not to invent sources or make unsupported assertions.

The LaTeX file is then compiled to PDF and converted to Word (DOCX), giving you three format options for the same content.

What you receive: five deliverables

When the pipeline finishes, you will see the job marked as "Completed" in your dashboard. Before unlocking the full download, you can preview the PDF to check that the content matches your expectations.

After unlocking (30 credits), you get five files:

1. Report PDF

The formatted, ready-to-read document. This is what most people will use directly — print it, share it with your advisor, or attach it to a proposal. The PDF includes proper page numbers, section headings, and a formatted reference list.

2. Report DOCX (Word)

The same content in an editable Word document. This is useful when you need to:

• Revise specific sections before including them in a larger document
• Add your own commentary or additional sources
• Share with collaborators who work in Word

3. Report TEX (LaTeX source)

The raw LaTeX file. If you work in Overleaf or a local LaTeX setup, you can import this directly and continue editing with full control over formatting. The LaTeX uses standard packages and APA 7.0 bibliography style.

4. References BIB

A standard BibTeX file containing all cited sources. You can import this into any reference manager. Each entry uses descriptive keys like smith2023deep rather than opaque identifiers, making it easy to find and modify entries.

5. Sources JSON

Structured metadata for every source: title, authors, year, venue, DOI, URL, abstract snippet, relevance explanation, and verification status. This file is useful if you want to programmatically filter or analyze the source list, or if you want to verify specific references yourself.

A practical example

Suppose you are writing a thesis proposal on the relationship between AI-assisted learning, academic integrity, and student performance in higher education. You enter that topic, select English as the output language, and submit.

About 30 minutes later, you have:

• A 12-page PDF with an abstract, five thematic sections covering AI tutoring tools, plagiarism detection challenges, assessment redesign, student perceptions, and institutional policy responses, plus a discussion of research gaps
• 24 cited sources with BibTeX entries ready for your reference manager
• A synthesis document that maps how sources cluster around your subtopics
• A LaTeX file you can drop into your thesis template and keep editing

You did not have to open Google Scholar, read 50 abstracts, decide which 24 to keep, organize them into themes, write transitions between sections, or format a single citation. The pipeline handled the mechanical work; your job is to read the output, decide what to keep, and refine the argument.

When the output needs revision

The report is a starting point, not a finished chapter. Common things you might want to change:

• Add sources you already know about. The pipeline searches broadly but may miss papers you consider essential. Import the BIB file into your reference manager, add your own entries, and update the text.
• Adjust the thematic structure. The pipeline identifies themes automatically, but you may want to merge two sections or split one into finer categories.
• Strengthen specific arguments. The synthesis covers each theme at a moderate depth. If one theme is central to your research, you will want to expand it with closer reading of the cited sources.
• Update the introduction. The pipeline writes a general introduction based on the topic. You may want to rewrite it to connect more directly to your specific research questions.

The DOCX and TEX formats are designed for exactly this kind of follow-up editing. The report gives you a structured draft with verified citations — you contribute the domain judgment and argumentative focus.

Seven languages, same pipeline

The language selector is not a post-processing translation step. The entire Stage 4 compilation happens in the target language, which means:

• Section headings use the correct academic conventions for that language
• Citation formatting follows language-appropriate norms
• The prose reads naturally rather than as a translated document

This is particularly valuable for researchers who need to publish in their native language or who are preparing reports for regional funding agencies. A Japanese researcher writing a literature review for a JSPS grant application can get the full output in Japanese without manual translation.

How it relates to Data2Paper's other products

Data2Paper offers three distinct products that cover different stages of the research lifecycle:

• Generate Paper starts from data files (CSV, XLSX) and produces a complete research paper with statistical analysis. This is for when you have collected data and need to turn it into a paper.
• Research Report (this product) starts from a topic and produces a literature review. This is for when you need to survey existing work before or during your own research.
• Paper Review starts from a finished paper (PDF) and produces peer review feedback. This is for when you have a draft and want to improve it before submission.

The three products share the same output pipeline (PDF, Word, LaTeX), the same seven-language support, and the same dashboard interface. But they serve different inputs and different stages of the research process.

Getting started

Visit the Research Report page to try it. Enter your research topic, choose a language, and submit. You will receive an email notification when the report is ready, and you can track progress in your dashboard.

The best results come from specific, well-scoped topics. If your topic is broad, consider narrowing it to a specific time period, geographic context, methodology type, or theoretical framework. The pipeline will do the searching and synthesizing — your job is to tell it exactly what to search for.