Search & Research

Learning Objectives

• Distinguish between AI search tools, deep research agents, source-grounded tools, and academic research platforms
• Compare Perplexity, ChatGPT Deep Research, NotebookLM, Elicit, and Consensus for different research use cases
• Select the appropriate research tool based on source requirements, research type, and citation needs

Research Tools Are Not All the Same

"AI for research" covers a surprisingly wide range of capabilities. Understanding the differences prevents the common mistake of using a general-purpose chatbot for academic research, or a citation-focused tool for exploratory brainstorming.

Four distinct categories:

1. AI search — searches the web in real time, synthesizes results with citations (Perplexity)
2. Deep research agents — autonomously searches dozens of sources over hours and produces comprehensive reports (ChatGPT Deep Research, Gemini Deep Research)
3. Source-grounded tools — answers only from documents you provide (NotebookLM, Claude with uploads)
4. Academic research — searches scientific literature specifically (Elicit, Consensus, Semantic Scholar)

Perplexity — AI-Powered Search

Perplexity is the most polished AI search product: ask a question, receive a synthesized answer where every factual claim links to its source.

The key design principle differentiating Perplexity from ChatGPT or Claude: every response is grounded in real-time web retrieval, and every claim cites its source. You can verify anything it says, trace it back to the original page, and explore further.

Practical strengths:

• Current information: No knowledge cutoff — searches the web at query time
• Source citation: Every claim links to its source; no guessing where the information came from
• Follow-up questions: Ask follow-up questions that refine or deepen the original search
• Perplexity Pro: Access to GPT-5.5, Claude Opus 4.7, and Gemini 3.1 Pro on the same interface — useful for comparing how different models answer the same question
• Spaces: Organize research by topic; share collections of related searches

The right mental model: use Perplexity when you want reliable, cited information about current topics. It's the replacement for searching Google and reading through multiple result pages.

ChatGPT Deep Research — Autonomous Research Agent

ChatGPT Deep Research (available on ChatGPT Plus and Pro) takes a fundamentally different approach: rather than answering one question with web search, it conducts multi-hour autonomous research and produces a comprehensive report.

The workflow: you provide a research question and optional constraints ("focus on peer-reviewed sources," "include recent developments from the last 6 months"). Deep Research runs in the background, searching dozens to hundreds of sources, following relevant links, extracting key information, and synthesizing it into a structured research report.

The output quality at its best: comparable to a research brief that a human analyst would produce in a day, delivered in 15-60 minutes. Comprehensive citations, structured argument, identification of key debates and open questions.

Limitations: Deep Research can miss nuances, occasionally hallucinates citations for complex or obscure topics, and doesn't have the domain expertise of a specialist researcher. Treat its outputs as a strong first draft requiring expert review for high-stakes decisions.

Best use cases: competitive intelligence research, understanding a new technical domain, policy landscape analysis, literature surveys in unfamiliar fields, market research briefs.

Gemini Deep Research

Gemini Deep Research is Google's equivalent capability, with one meaningful differentiator: Gemini's research is grounded in Google Search — the most comprehensive web index available.

For research topics where recency and comprehensive web coverage matter most, Gemini's Google Search grounding provides an advantage. The output format is similar to ChatGPT Deep Research: a structured report with sources.

Deep Research is available to Gemini Advanced subscribers (included with Google One AI Premium).

Claude for Document Research

Claude (particularly with the Projects feature) serves a different research pattern: deep analysis of documents you provide.

Where Perplexity searches the web and Deep Research tools roam the internet autonomously, Claude with uploaded documents is for research within a bounded set of sources:

• Upload 10 research papers and ask Claude to synthesize the findings
• Upload your company's policy documents and ask Claude to identify gaps or inconsistencies
• Upload several competitor websites and ask for a comparative analysis

The 1 million token context window means Claude can hold an enormous amount of document content in a single analysis — multiple long reports simultaneously, the equivalent of several full-length books at once. Citations in Claude's responses reference specific sections of the uploaded documents.

Projects (in Claude.ai) enables persistent workspaces where you can upload sources once and return to them across multiple sessions — avoiding re-uploading the same documents every time.

NotebookLM — Source-Only Research

NotebookLM shares Claude's source-grounded design but makes it more explicit and structured:

• Every answer cites specific passages from specific uploaded documents
• The AI refuses to answer from general knowledge — only from sources you provided
• The Audio Overview feature converts your uploaded sources into a 15-20 minute podcast-style conversation between two AI hosts

NotebookLM is the right tool when you need to work intensively with a specific set of documents and want guaranteed source grounding rather than general AI knowledge. Research synthesis, document review, technical documentation Q&A.

Elicit — Academic Literature

Elicit is purpose-built for academic research, searching across 200+ million scientific papers:

• Enter a research question; Elicit searches across academic databases
• Papers are ranked by relevance and summarized with key findings, methodology, and limitations
• Extract specific information types from papers ("what was the sample size?" "what were the main findings?") across many papers simultaneously
• Identify consensus and disagreement across the literature

Elicit was built by Ought, an AI safety research organization — the design reflects a commitment to reliable, verifiable academic research rather than general-purpose web browsing.

Consensus — Scientific Consensus Detection

Consensus focuses on a specific research question: "what does the scientific literature say about X?"

The distinctive capability: Consensus aggregates findings across papers and tells you where the scientific literature agrees or disagrees. For factual questions with scientific backing ("does exercise improve sleep quality?"), Consensus provides a literature-based answer with papers cited.

For professionals who need to know the scientific consensus on a topic — not just opinions or news coverage, but what the peer-reviewed literature actually says — Consensus provides direct access without requiring JSTOR access or deep academic search skills.

Tavily — Research API for AI Agents

Tavily is a web search API designed specifically for integration into AI applications and agents:

from tavily import TavilyClient
client = TavilyClient(api_key="...")
results = client.search("latest developments in LLM reasoning", max_results=5)

Where most search APIs return HTML requiring parsing, Tavily returns clean, LLM-ready text extracted from relevant pages. It's optimized for the use case where a coding agent or research agent needs to search the web as part of a multi-step workflow.

Choosing the Right Research Tool

Key Takeaways

• AI research tools differ fundamentally by source: Perplexity and Deep Research search the open web; NotebookLM and Claude answer only from your documents; Elicit and Consensus search scientific literature specifically
• Perplexity is the daily-use AI search tool; ChatGPT/Gemini Deep Research produces comprehensive autonomous research reports; NotebookLM provides guaranteed source-grounded answers
• For academic and scientific research, Elicit (literature review) and Consensus (scientific consensus detection) are purpose-built tools that general AI assistants don't replace
• Tavily is the developer tool: a clean web search API optimized for LLM agent integration