On this page
A senior analyst spends 4-6 hours researching a single market question: querying multiple databases, reading 20-30 articles, cross-referencing data points, and synthesizing findings into a coherent brief. Most of that time is spent on mechanical work -- finding sources, extracting relevant paragraphs, checking for contradictions -- not on the actual analysis.
CrawlForge's deep_research tool compresses this entire workflow into a single API call. It automatically expands your query, searches multiple sources, verifies credibility, detects conflicting information, and produces a synthesized report with citations. This guide shows you how to build a production research agent around it.
Table of Contents
- What Is Deep Research
- Architecture Overview
- Step 1: Configure the Research Agent
- Step 2: Run Multi-Source Research
- Step 3: Process and Validate Findings
- Step 4: Generate Research Reports
- Step 5: Build Recurring Research Workflows
- Credit Cost Analysis
- Results and Benefits
- Frequently Asked Questions
What Is Deep Research
What is deep research in AI? Deep research is an automated multi-stage information gathering process where an AI agent systematically queries multiple sources, extracts relevant findings, cross-references data for accuracy, detects contradictions, and synthesizes results into a structured report with source citations and credibility scores.
Unlike a simple web search that returns 10 links, deep_research operates as a complete research workflow:
| Stage | What Happens | Why It Matters |
|---|---|---|
| Query expansion | Generates synonym and related queries | Catches results a single query would miss |
| Multi-source search | Queries 10-50 sources in parallel | Breadth of coverage |
| Content extraction | Pulls relevant passages from each source | Depth of information |
| Source verification | Scores each source for credibility | Quality assurance |
| Conflict detection | Flags contradictory information | Accuracy |
| Synthesis | Produces a coherent report with citations | Actionable output |
How does deep research work? The tool takes a research topic, automatically generates expanded search queries, crawls and analyzes pages from multiple sources, scores source credibility, detects conflicting claims across sources, and synthesizes everything into a structured report. The entire process runs in 30-120 seconds depending on scope.
Architecture Overview
A research agent combines deep_research with supporting tools:
| Component | Tool | Credits | Purpose |
|---|---|---|---|
| Core research | deep_research | 10 | Multi-source investigation |
| Follow-up extraction | extract_content | 2 | Deep-dive on specific sources |
| Document analysis | process_document | 3 | Parse cited PDFs and reports |
| Summary generation | summarize_content | 2 | Executive summary creation |
| Supplemental search | search_web | 5 | Targeted follow-up queries |
Step 1: Configure the Research Agent
Set up a research agent with configurable parameters for different research types.
Step 2: Run Multi-Source Research
Execute the research with deep_research and handle the structured output.
Step 3: Process and Validate Findings
For critical research, drill deeper into key sources and validate specific claims.
Step 4: Generate Research Reports
Transform raw research output into polished, shareable reports.
Step 5: Build Recurring Research Workflows
For ongoing research needs, set up scheduled workflows that track how topics evolve over time.
Credit Cost Analysis
deep_research at 10 credits per call is the most expensive single CrawlForge tool, but it replaces what would otherwise require 15-30 individual tool calls.
| Research Type | Tools Used | Total Credits | Manual Equivalent |
|---|---|---|---|
| Quick topic research | deep_research | 10 | 2-3 hours |
| With source validation | + extract_content x5 | 20 | 4-5 hours |
| With PDF analysis | + process_document x2 | 26 | 5-6 hours |
| Full report generation | + summarize_content | 28 | 6-8 hours |
Monthly costs for a research team:
| Usage | Credits/Month | Recommended Plan |
|---|---|---|
| 10 reports/month | 280 | Free tier (1,000 credits) |
| 50 reports/month | 1,400 | Hobby ($19/mo, 3,000 credits) |
| 200 reports/month | 5,600 | Professional ($99/mo, 15,000 credits) |
Results and Benefits
A CrawlForge research agent delivers:
- Speed: Complete a research brief in 2-5 minutes instead of 4-6 hours
- Breadth: Analyze 20-50 sources per query vs. 5-10 manually
- Accuracy: Built-in conflict detection catches contradictions humans miss
- Consistency: Same methodology every time, no researcher bias
- Auditability: Every source cited with credibility scores
The tool is particularly powerful for recurring research -- tracking how a market evolves weekly, monitoring regulatory changes, or keeping a competitive landscape document current.
Frequently Asked Questions
How does deep_research compare to Perplexity or ChatGPT Search?
deep_research analyzes significantly more sources (up to 50 per query vs. 5-10 for chat-based search), includes credibility scoring, detects conflicts between sources, and returns structured data you can process programmatically. Chat-based tools are better for quick, conversational answers. CrawlForge is better for systematic, repeatable research workflows.
Can I use deep_research for academic research?
Yes. Set researchApproach: 'academic' and sourceTypes: ['academic', 'government'] to prioritize scholarly sources. The credibility threshold filters out low-quality sources automatically. Note that deep_research works with publicly accessible web sources -- it cannot access papers behind paywalls.
Is 10 credits per call worth it?
Consider the alternative: manually, you would use search_web (5 credits) + multiple extract_content calls (2 credits each) + analyze_content (3 credits each). Doing this for 20 sources would cost 100+ credits. deep_research packages all of this into a single 10-credit call with built-in synthesis.
Try deep research right now. Start free with 1,000 credits -- enough for 100 research reports. No credit card required.
Related resources:
