
No single AI tool covers research from the first search to the final draft, so most researchers combine a paper finder, a citation or mapping tool, a reader for their own sources and a check against the original paper. This guide groups the tools by job and links each one to its own site, not to a review of it.
Georgetown University Library’s guide to AI tools for research lists 13 tools: Connected Papers, Consensus, Elicit, Keenious, Research Rabbit, scite, Scholarcy, Semantic Scholar, Undermind, ChatGPT, Claude, Gemini and Perplexity. This guide keeps all 13 and adds tools found in other library guides, in the Ithaka S+R Generative AI Product Tracker and in 2026 reviews.
BSD did not test these tools. Each description says what the tool is for, as library guides and reviews describe it. Prices and plans change often, so this guide gives none; the tool’s own site has the current ones.
Names change too: Google renamed NotebookLM to Gemini Notebook on July 16, 2026, and older guides still use the old name.
Six jobs. The tools fall into literature search, mapping and citation checking, reading your own sources, deep research agents, data analysis and writing, and reference management. Most projects need two or three of them.
Own-site links. Every tool name links to the vendor’s or publisher’s own site. Where a fact comes from a library guide or a news page, that page is linked next to the fact.
Scoping a topic: Perplexity, Gemini or ChatGPT Deep Research for a first overview, then Connected Papers for a map.
Finding papers: Semantic Scholar, Consensus, Elicit, Undermind or Ai2 Asta, because they search scholarly records.
Following citations: Research Rabbit, Litmaps and scite.
Key point: The links go to each tool’s own site, but the descriptions are BSD’s summary of what library guides and reviews say. BSD’s reading, not a claim the vendors make: a tool that suits one researcher’s field may suit another’s poorly.
Search and answer. They find papers, answer a research question with cited sources or pull findings from many papers into one view.
Library licences. Two vendor tools run on publisher databases that your library may already license, so check there before you pay for anything.
Elicit screens papers and extracts findings into a table, which suits review-style work.
Consensus answers narrow questions with evidence from papers and shows how far studies agree.
Undermind runs an iterative search, which helps when keyword search returns little.
SciSpace is a workspace for reading papers one at a time, chatting with PDFs and comparing results.
Ai2 Asta finds papers and writes cited literature summaries. Ai2’s overview is at allenai.org/asta. The ZBW library says it is built on Semantic Scholar’s database.
Semantic Scholar is a free academic search engine from Ai2 with short summaries and citation graphs.
Paperguide combines paper search, PDF chat, writing and citation management in one place.
Keenious recommends papers based on the text or document you give it.
R Discovery, AnswerThis, Sourcely, Anara and Iris.ai appear in other lists of literature tools.
Scopus AI runs on the Scopus database.
Web of Science Research Assistant runs on the Web of Science database.
The HKUST Library compares Scopus AI with Scite, Elicit and Consensus in its tool comparison.
Syracuse University Libraries note that Web of Science and Semantic Scholar have built-in AI tools in their research guide.
Maps. A mapping tool takes one seed paper and shows what cites it and what it cites. Maps work best in fields with dense citation networks; in a young field the graph has little to show.
Checks. A citation tool shows how later papers treated a claim, which is the step most searches skip.
Research Rabbit builds citation-based maps that surface similar papers and related authors.
Connected Papers draws a graph of a field from one seed paper.
Litmaps keeps literature maps up to date and sends alerts when new work cites your seed papers.
scite classifies citations as supporting, contrasting or only mentioning a claim, which makes it a tool for checking claims.
Grounded answers. When a tool answers only from papers you supply, you can check each answer against a known set of documents instead of the open web.
Old names. Google announced the rename of NotebookLM to Gemini Notebook in its announcement post and calls it the same standalone product.
Gemini Notebook, formerly NotebookLM, lets you chat with sources you upload.
Scholarcy turns papers you upload or link into summary cards. It reads; it does not search.
ChatPDF answers questions about a single PDF.
STORM, from Stanford, drafts cited, Wikipedia-style overviews of a topic. Check the draft like any generated text.
Open web first. These assistants run a multi-step search and write a report. They draw on the open web more than on scholarly databases, so the quality of their sources differs from report to report.
Use for orientation. BSD’s reading, not a claim the vendors make: a report from one of these tools is a first pass to verify, not a literature review.
ChatGPT has a Deep Research mode for long reports.
Gemini has Deep Research. One 2026 review reports that Google announced Deep Research agents for developers on April 21, 2026.
Perplexity gives answers with inline citations and has a Deep Research mode.
Claude reads long papers and drafts in one pass. It is not a scholarly database.
Gemini for Science is Google’s set of science tools and experiments. Google describes a Science Skills bundle that links more than 30 life-science databases.
Firecrawl offers retrieval infrastructure for developers building research agents. It is not a tool for an individual researcher.
Not discovery tools. None of these finds papers. They analyse data, polish language or store citations.
First pass only. Treat analysis output from any of them as a first pass; formal statistical reporting still needs a statistician’s checks.
Julius AI analyses spreadsheets and statistical files from plain-language prompts.
NVivo from Lumivero is a qualitative-analysis package that has added AI features.
Paperpal edits academic language and checks a manuscript before submission.
Jenni AI, Writefull and Yomu help with drafting and academic writing.
Granted AI is built for grant applications.
Zotero manages citation libraries. AI chat over saved papers comes from plugins, not from Zotero itself.
Logically, formerly Afforai, combines a reference manager with an AI research assistant and writer.
Features, names and plans were not tested on the vendors’ sites for this guide. Open each tool’s own page before you pay or upload anything.
Wrong details. One reviewer on DEV Community reports that every tool in this group can state a detail wrongly, such as a study’s sample size.
Paywalls. Queen’s University Belfast warns in its library guide that generative AI tools typically cannot reach subscription-only sources, so use them next to your library’s databases.
Use at least two tools for any search that matters, because one tool misses work that another finds. Georgetown’s list names several for that reason.
Check the copyright limits on any content you upload. Georgetown’s guide advises awareness of them.
Many “best tool” lists come from vendors that rank their own product highly. Use them to find tools, then judge the tools yourself.
The Ithaka S+R Generative AI Product Tracker lists products marketed to college and university faculty and students, or already used by them.
Library guides with their own lists: Buffalo (which credits Georgetown’s guide as its model), York, Illinois, Virginia, HKUST, Yale, Syracuse and Binghamton.
Bottom line: Use these tools to find and organise sources, and read the paper before you repeat a claim from any of them.