This AI Tool Wrote My Literature Review - and It’s a Game-Changer

Thesis AI is positioned as an academic writing assistant that can draft full scientific documents—especially literature reviews and paper introductions—by ingesting a set of reference PDFs and producing a structured, editable draft in minutes (with longer generation time). The practical pitch is straightforward: upload up to 100 papers, set citation and formatting preferences, write a short prompt describing the topic, and receive a ready-to-edit document output (including PDF export). For researchers facing the time sink of synthesizing sources into a coherent narrative, the workflow aims to compress days of drafting into a first pass that already looks like a thesis section.

In a live walkthrough, the process starts with configuration. Users choose a citation style (the example uses IEEE), set the document language, and adjust a “temperature” parameter—kept at zero to prioritize strict, non-fanciful output. After payment, the creator generates documents with up to 50 pages output and can preview up to 10 pages on the free tier. The core input step is uploading PDF references; the example uploads 21 papers, then leverages Thesis AI’s ability to work from those sources rather than relying on a generic web search.

A key differentiator is how Thesis AI handles literature-review sourcing. The workflow demonstrates using a separate paper-finding tool (Elicit) to identify relevant papers, filter them by criteria like “has PDF” and publication window (2014 to present), and then bulk-upload the resulting PDFs into Thesis AI. Once uploaded, a simple prompt requests “the latest technology, latest methods, latest materials” for indoor and outdoor applications. Generation takes up to about 30 minutes, during which the user can do other tasks.

The output is presented as a structured document with headings and subheadings that resemble a real academic draft. In the example, the system produces a 17-page document from 21 uploaded references, while not necessarily using every uploaded paper; it later reports using 16 references in the final bibliography. The draft includes an abstract, body sections, conclusions, and a references list, with citations mapped to the uploaded sources. The creator notes that the structure can feel “manufactured” at first glance (for example, consistent heading depth), but also emphasizes that the layout is usable as a starting point.

A major usability claim is export and editing flexibility. Thesis AI outputs LaTeX (LaTeX/LATEX integration is highlighted), enabling direct import into Overleaf for full editing rather than being locked into an online editor. The workflow is further strengthened by Overleaf’s integration with Right for proofreading and rewriting, allowing users to refine academic tone and clarity.

Finally, the transcript includes an AI-detection test. Thesis AI is marketed with low detectability claims (“less than 25% AI detectability with zero GPT”), and the example compares detection results across tools. One detector reports a low probability of AI generation (about 10.06% in the cited test), while another flags the text as highly likely AI-generated when run through a different service. The takeaway is less about a guaranteed pass and more about the need to review, edit, and align the writing with the researcher’s own voice and references.

Overall, Thesis AI’s value proposition hinges on turning a curated PDF set into a structured, citation-backed literature review draft that can be exported to LaTeX/Overleaf for customization—then optionally polished with writing assistance—while acknowledging that AI-detection outcomes vary by tool and that human editing remains essential.