Research paper summarizer that cites the section for every claim and separates findings from interpretation
A Claude Project for analysts, students and product teams reading academic papers: upload a PDF and get a structured summary where each point cites the section or page it came from, with methods, limitations and what the paper does not show.
Agent in Research by AISkillHub Library
What it does
Paper summaries often blur what the authors measured, what they concluded, and what the reader would like to be true. This project produces a structured summary where every statement is traceable to a section of the paper, and limitations are given the same weight as findings.
**What it does**
- Reads an uploaded paper (PDF or text) in project knowledge or in the chat. - Produces a summary in a fixed structure: citation, question, method, data, main findings, effect sizes or key numbers as reported, authors' interpretation, limitations stated by the authors, limitations the reader should consider, and what the paper does not show. - Cites a section and, when the PDF has page numbers, a page for each point, for example `(§4.2, p.7)`. - Distinguishes **reported results** from **authors' claims** from **reader notes**, using labels. - Answers follow-up questions with section citations, and says "the paper does not say" when it does not. - Offers a plain-language version for a non-specialist audience and a one-paragraph version for a newsletter or literature review.
**When to use it**
- Deciding whether a paper is worth a full read - Literature reviews, where you need comparable notes across many papers - Briefing stakeholders on a study someone cited in a meeting
**What it will not do**
- It does not judge whether a paper is correct or replicate analyses. It notes design features that limit conclusions (sample size, lack of a control group, self-reported measures) when they are visible in the paper. - It does not add results from other papers or general knowledge into the summary. Background it adds is labelled as such. - It does not fabricate citations or page numbers. If the text has no page markers, it cites sections only.
**Example**
"Main finding: In the randomized trial (n = 412), the intervention group completed onboarding more often than control (64% vs 51%) [Reported, §5.1, Table 2]." followed by "The authors suggest the effect is driven by reduced form length [Authors' claim, §6], which the study design does not isolate [Reader note]."
How to use it
**Set up as a Claude Project** (Projects are available on claude.ai; limits depend on your plan).
1. In Claude, open **Projects** and create a new project named `Paper Summaries`. 2. Open the project's instructions setting (the panel for custom instructions that apply to every chat in the project) and paste the full contents of `assistants/research-paper-summarizer/instructions.md`. 3. Add reference material to the project's knowledge area (upload files or paste text): (optional) papers you want to compare. You can also attach a paper directly in a chat. For a literature review, add a short text file stating your review question so summaries highlight what is relevant to it. 4. For scanned PDFs, check that the text is selectable; image-only scans may not be read accurately. 5. Start a new chat inside the project. Every chat there uses these instructions and can read the project knowledge.
**Using it:** attach a paper and type `summarize`. Then try `plain language`, `one paragraph`, `compare with <other paper>`, or ask specific questions such as `What was the control condition?`.
Files
Save each file at the path shown, relative to your project root.