Deep Research System

Last Updated: March 2026

Overview

The Deep Research System is Scrapalot's advanced AI research feature that transforms your questions into comprehensive, academically rigorous research reports. It automatically plans research strategies, searches multiple sources, validates information, and synthesizes findings into well-structured reports with proper citations.

What is Deep Research?

Deep Research goes beyond simple question-answering by conducting thorough, multi-step research similar to how a human researcher would approach a complex topic. Instead of just retrieving existing documents, it actively searches the web, evaluates sources, cross-validates information, and produces original research reports.

Key Features

• 🎯 Intelligent Planning: Automatically creates structured research plans with clear objectives
• 🔍 Multi-Source Search: Searches across multiple providers to find the most relevant and credible sources
• Quality Validation: Evaluates source credibility and cross-checks facts across multiple sources
• 🩺 Self-healing sources: If a search source is down or out of quota, research doesn't quietly give up — it automatically continues with the next working source and shows you a small notice ("this source was unavailable, continued with that one") so a partial run is never presented as if it were complete
• 🎭 Research Council (optional): Turn on a panel of AI "council members" that argue the question from different angles and surface convergences, tensions, and blind spots. Each member can run on a different model, and you can save reusable panels ("rosters") and load one into any run. If a member's model fails, it's shown as a greyed-out seat with the reason — a partial panel is never passed off as complete
• 🛰️ Run in the background: Turn on "Run in background" in the research setup and the investigation runs on the server instead of your browser — you can close the tab and come back later. Progress shows in the header, and you'll get a notification the moment it finishes
• ✅ See what it proved: When you reopen a finished report, a hypothesis ledger shows each claim the research set out to test and whether it was supported, refuted, or left inconclusive, with a confidence bar and how much evidence backed the verdict — so you can judge the conclusion, not just read it
• 📋 Custom research templates: Besides the built-in templates (Scientific Paper, Literature Review, PRISMA, …), you can create your own in Settings → Prompts → Research Templates — pick the writing tone, report type, citation style, and research depth/breadth once, name it, and it appears as a one-click choice in the research setup in chat. Cloning a built-in template gives you a pre-filled starting point
• 📊 Comprehensive Synthesis: Combines findings into coherent, well-structured reports
• 📚 Proper Citations: Includes citations and bibliography in academic formats (APA, MLA, Chicago, IEEE)
• ⚡ Parallel Processing: Executes research tasks simultaneously for faster results

Success Metrics

Our Deep Research system delivers:

• High Accuracy: >85% factual accuracy in synthesized reports
• Comprehensive Coverage: Average 15+ sources per complex research topic
• Fast Results: <5 minutes for comprehensive research on complex topics
• Quality Consistency: 95% of reports meet academic quality standards

When to Use Deep Research

How Deep Research Works

The system follows a five-phase approach to ensure thorough and reliable research:

Phase 1: Research Planning

The system analyzes your question and creates a structured research plan:

• Identifies key research objectives
• Breaks down the topic into focused sections
• Selects the appropriate research methodology
• Estimates research scope and complexity

Phase 2: Task Breakdown

Your research is organized into specific, manageable tasks:

• Each task has clear objectives and success criteria
• Tasks are organized to run in parallel when possible
• Dependencies between tasks are identified
• Quality checkpoints are established

Phase 3: Coordinated Research

Multiple specialized research agents work together:

• Geographic Research: Regional data and location-specific analysis
• Temporal Analysis: Historical trends and timeline construction
• Statistical Data: Quantitative analysis and data validation
• Policy Analysis: Regulatory frameworks and policy impacts
• Content Synthesis: Multi-source integration and narrative building

Phase 4: Multi-Source Search

The system searches across multiple providers to ensure comprehensive coverage:

• Primary sources: Academic databases and authoritative publications
• General web search: Current information and diverse perspectives
• Specialized search: Domain-specific resources

Source Evaluation:

• Credibility scoring (0.0-1.0) based on authority, accuracy, and reliability
• Bias detection to ensure balanced perspectives
• Quality assessment of evidence and scholarly rigor

Phase 5: Synthesis & Quality Assurance

Your research is synthesized into a comprehensive report:

Synthesis Features:

• Combines findings from all sources
• Resolves contradictions through cross-validation
• Organizes content logically (analytical, comparative, or narrative style)
• Adds executive summary and key insights

Quality Assurance:

• Completeness (25%): Coverage of all research objectives
• Accuracy (25%): Factual correctness and data integrity
• Coherence (20%): Logical flow and narrative consistency
• Citation Quality (15%): Proper attribution and bibliography
• Balance (15%): Source diversity and bias mitigation

Quality Levels:

• Exceptional (0.95-1.0): Publishable academic quality
• Excellent (0.90-0.94): High-quality, publication-ready
• Very Good (0.85-0.89): Solid research, minor improvements possible
• Good (0.80-0.84): Acceptable quality, moderate refinements suggested

Using Deep Research

Enabling Deep Research

1. In Chat Settings: Toggle "Deep Research" mode when asking complex questions
2. Research Depth: Choose depth level (1-5) for how thorough you want the research
3. Research Breadth: Select breadth level (1-5) for how wide-ranging the search should be

Best Use Cases

Ideal for Deep Research:

• Complex topics requiring multiple perspectives
• Questions needing current web information
• Research requiring fact-checking and validation
• Topics where source credibility matters
• Academic or professional research needs

Better for Standard Chat:

• ❌ Simple factual questions
• ❌ Queries about your uploaded documents only
• ❌ Quick lookups or definitions
• ❌ Time-sensitive questions (when speed is critical)

Example Research Topics

Business Analysis:

"Analyze the current state and future trends of renewable energy adoption in the European Union, including policy impacts and economic factors."

Technical Research:

"Compare the security features and performance characteristics of modern container orchestration platforms for enterprise deployments."

Academic Research:

"Examine the relationship between remote work adoption and urban housing market trends in major metropolitan areas post-2020."

Understanding Research Reports

Your final research report includes:

Report Structure

1. Title: Clear, descriptive title for your research
2. Executive Summary: High-level overview of key findings
3. Main Content: Organized sections with detailed analysis
4. Key Insights: Bullet-point summary of main discoveries
5. Recommendations: Actionable takeaways (when applicable)
6. Citations: Inline citations for all factual claims
7. Bibliography: Complete reference list in your chosen format

Citation Formats

Choose your preferred citation style:

• APA: American Psychological Association (default)
• MLA: Modern Language Association
• Chicago: Chicago Manual of Style
• IEEE: Institute of Electrical and Electronics Engineers

Quality Indicators

Each report includes quality metrics:

• Overall quality score (0.0-1.0)
• Individual dimension scores
• Academic readiness indicator
• Improvement suggestions (when applicable)

Research Progress Tracking

During research, you'll see real-time updates:

Planning Phase:

• Analyzing your question
• Creating research plan
• Identifying research sections

Execution Phase:

• Number of tasks completed
• Parallel task execution status
• Source discovery progress

Synthesis Phase:

• Content synthesis progress
• Validation checks running
• Citation processing status

Configuration Options

Research Depth (1-5)

• Level 1-2: Quick research, 5-10 sources, <2 minutes
• Level 3: Standard research, 10-15 sources, 2-4 minutes
• Level 4-5: Comprehensive research, 15+ sources, 4-6 minutes

Research Breadth (1-5)

• Level 1-2: Focused on main topics
• Level 3: Balanced coverage
• Level 4-5: Explores related topics and diverse perspectives

Quality Standards

• General: Good quality for everyday use
• Professional: Higher standards for business use
• Academic: Strict standards for academic/research purposes

Tips for Best Results

1. Be Specific: Detailed questions get better research plans
2. Indicate Scope: Mention time periods, regions, or specific aspects you care about
3. Set Appropriate Depth: Use higher depth for complex topics
4. Review Progress: Watch the research progress to understand what's being investigated
5. Check Quality Score: Higher scores indicate more reliable reports

Limitations & Considerations

Current Limitations:

• Research time: 2-6 minutes for comprehensive topics
• Source access: Limited to publicly accessible web content
• Language: Primarily optimized for English research
• Real-time data: Information current as of search time

Privacy & Security:

• Research queries are processed to generate search strategies
• Sources are fetched from public web content
• No proprietary or confidential data is accessed
• All research is associated with your user account

Related Features

• RAG Strategy: Learn about document-based question answering
• Document Processing: Upload your own documents for research
• Model Management: Choose AI models for research synthesis

---

Need Help? Check our User Guide or FAQ for more information about using Deep Research effectively.