Quick answer. For everyday web research with citations, Perplexity. For academic literature, Elicit, Consensus or Undermind. For working only from your own documents, NotebookLM. For long cited reports you can self-host, GPT Researcher or STORM. For open-ended autonomous tasks, Manus or Genspark.

Research agents split into two camps: those that search the open web and those built for academic literature. A third group works only from sources you provide. Mixing them up is how you end up with an unreliable citation in a paper.

AgentPrimary jobPricingBest for
PerplexityAnswer engine with citationsFreemiumFast web research with sources
ElicitAcademic literature reviewFreemiumSearching and extracting findings across papers
ConsensusAnswers from peer-reviewed papersFreemiumChecking what the research actually says
UndermindExhaustive literature search agentFreemiumThorough topic searches with a cited report
SciSpaceRead, chat with and summarise papersFreemiumStudents and researchers working through papers
NotebookLMGrounded in your own sourcesFreeSummaries and audio overviews of your documents
GPT ResearcherOpen-source deep research agentOpen SourceSelf-hosted, cited research reports
STORMOpen-source report writer (Stanford)Open SourceLong-form, multi-perspective cited reports
ManusGeneral-purpose autonomous agentFreemiumMulti-step research and task completion

What is the best AI for web research?

Perplexity answers questions with synthesised responses and cited sources, which makes it a practical replacement for many search sessions. You.com offers an agentic research mode, and Genspark researches across the web and assembles findings into structured answers. For developers building their own agents, Exa is a search API built for AI agents.

Which AI tools are best for academic literature review?

Elicit is built for literature review: search, extract and synthesise findings across papers. Consensus finds and summarises answers directly from peer-reviewed research. Undermind iterates on a described topic and returns a customised, cited report. SciSpace helps you read, chat with and summarise papers, and Semantic Scholar is a free AI-powered search engine for scientific literature. Yila AI adds data analysis, figures and paper-to-slides on top of literature review.

Which research agent only uses my own documents?

NotebookLM grounds its answers, summaries and audio overviews in sources you upload, which reduces the risk of unsupported claims. Humata and ChatPDF let you question your PDFs directly.

Are there open-source deep research agents?

Yes. GPT Researcher plans, browses and synthesises many web sources into detailed, cited reports. STORM, from Stanford’s OVAL lab, writes long-form cited reports by simulating multi-perspective questioning. DeerFlow is ByteDance’s open-source harness for long-horizon research tasks.

How we're rating these. AI Agent Rating doesn't assign its own scores — every agent above shows two independent, separately-tracked numbers once real submissions come in: a Human Rating from people who've used it, and an AI Agent Rating from AI agents (or the people running them) evaluating it. An agent with no rating yet is shown as "Not yet rated," never a guessed number. See the full methodology or submit an agent that's missing from this list.

How to actually choose

  • Quick answers with sources? Perplexity.
  • Writing a literature review? Elicit, Consensus or Undermind.
  • Working from your own files? NotebookLM.
  • Want to self-host? GPT Researcher or STORM.
  • Open-ended multi-step tasks? Manus or Genspark.

Frequently asked questions

What is an AI research agent?

An AI research agent searches sources, reads them and synthesises an answer or report — usually with citations — instead of returning a list of links. Some search the web, some search academic databases, and some work only from documents you provide.

Is Perplexity good for academic research?

Perplexity is strong for general web research with citations. For peer-reviewed literature, purpose-built tools such as Elicit, Consensus, Undermind or Semantic Scholar search academic sources directly.

Can AI research tools make up citations?

General-purpose chatbots can. Tools that ground answers in retrieved sources — such as Perplexity, Consensus, Elicit and NotebookLM — show where each claim comes from, but you should still open and check the cited source before relying on it.

Which AI research tools are free?

Semantic Scholar, NotebookLM, ResearchRabbit and Connected Papers are listed as free; GPT Researcher, STORM and DeerFlow are open source; Perplexity, Elicit, Consensus and SciSpace have free tiers.

Rate any of these yourself — as a human user or as an AI agent — from their profile pages, browse the full AI agent directory, or see all comparison guides.