PublicRisk.ai

STORM Document Generator

Complete guide to using STORM (Synthesis of Topic Outline and Research Methodology) for AI-powered document creation with research, citations, and professional formatting

Overview

The STORM Document Generator is PublicRisk.ai's advanced AI-powered document creation system. Based on Stanford's research methodology, STORM mimics how expert researchers create comprehensive documents through iterative research, outlining, and synthesis.

Production Ready: STORM is fully operational and used daily for creating risk assessments, policy documents, compliance reports, and more.


What is STORM?

STORM (Synthesis of Topic Outline and Research Methodology) is an AI system that generates comprehensive, research-backed documents automatically.

How STORM Differs from Simple AI Writing

FeatureSimple AI WritingSTORM
ResearchNo pre-researchSearches entire document library first
StructureGeneric formatCreates custom outline based on topic
CitationsNo sourcesEvery statement cited with source documents
QualityOne-shot generationIterative refinement with human approval
ConsistencyMay varyQuality review ensures coherence
ControlTake it or leave itApprove/modify outline before writing

The STORM Process


When to Use STORM

Ideal Use Cases

Risk Assessment Reports

  • Property risk assessments with hazard analysis
  • Vendor risk evaluations
  • Cybersecurity risk reports
  • Portfolio risk summaries

Policy Documents

  • Insurance policy summaries
  • Internal policy manuals
  • Procedure documentation
  • Compliance handbooks

Compliance Reports

  • Regulatory compliance audits
  • SOX compliance documentation
  • GDPR compliance reports
  • Industry-specific compliance reviews

Grant Applications

  • FEMA hazard mitigation grant proposals
  • Federal disaster recovery applications
  • Risk mitigation justifications
  • Loss estimation documentation

Legal Documents

  • Contract analysis summaries
  • Insurance clause reviews
  • Indemnification language drafts
  • Certificate of Insurance requirement documents

Training Materials

  • Employee safety training guides
  • Risk management training modules
  • Compliance training documentation
  • Procedure manuals

When NOT to Use STORM

Simple, short-form content (use Query Explorer instead) ❌ Content requiring no research (generic templates better) ❌ Highly creative content (marketing copy, creative writing) ❌ Content requiring real-time data (use live dashboards) ❌ Single-paragraph answers (Query Explorer more efficient)


Getting Started

Prerequisites

Before using STORM, ensure you have:

Upload Source Documents

STORM generates content from your document library. Upload relevant documents first:

  • Navigate to RAG Management
  • Upload policies, regulations, reports, or reference materials
  • Tag documents appropriately
  • Wait for processing to complete (documents must be indexed)

Define Your Topic

Know what you want to create:

  • What type of document (report, policy, analysis)?
  • What's the main topic or subject matter?
  • Who is the intended audience?
  • What level of detail is needed?

Prepare Key Points

List 3-5 key points that must be covered:

  • Specific risks or hazards to address
  • Regulatory requirements to include
  • Critical findings or recommendations
  • Important background or context

Step-by-Step Guide

1. Navigate to STORM Generator

From the main sidebar:

  • Click "Document Generator" or "STORM Generator"
  • You'll see the STORM configuration interface

2. Choose Document Template

STORM provides pre-built templates optimized for common document types.

Available Templates

Risk Assessment Report Template

Includes:

  • Executive Summary
  • Risk Identification & Analysis
  • Hazard Assessment (flood, earthquake, wildfire)
  • Loss Estimation
  • Mitigation Recommendations
  • Implementation Plan

Best For:

  • Property risk assessments
  • Portfolio risk analysis
  • Vendor risk evaluations
  • Cybersecurity risk reports

Configuration:

  • Property details (address, value, year built)
  • Hazard types to analyze
  • Risk scoring methodology
  • Mitigation budget range

Compliance Audit Report Template

Includes:

  • Compliance Overview
  • Regulatory Requirements
  • Current State Assessment
  • Gap Analysis
  • Risk & Impact Evaluation
  • Remediation Plan
  • Appendices (regulations, evidence)

Best For:

  • SOX compliance audits
  • GDPR compliance reviews
  • Industry regulation compliance
  • Internal policy compliance

Configuration:

  • Regulation or standard (SOX, GDPR, HIPAA, etc.)
  • Scope of audit
  • Compliance period
  • Risk tolerance level

Policy & Procedure Template

Includes:

  • Policy Purpose & Scope
  • Definitions
  • Policy Statement
  • Procedures & Implementation
  • Roles & Responsibilities
  • Compliance & Enforcement
  • Review & Updates

Best For:

  • Internal policy manuals
  • Safety procedures
  • IT security policies
  • HR policies

Configuration:

  • Policy category (HR, IT, Safety, etc.)
  • Approval authority
  • Effective date
  • Review frequency

Custom Document Template

Includes:

  • User-defined sections
  • Flexible structure
  • Custom formatting

Best For:

  • Unique document types
  • Non-standard formats
  • Highly specialized content

Configuration:

  • Define section structure manually
  • Specify research focus areas
  • Set formatting preferences

3. Configure Parameters

Fine-tune STORM's behavior with these settings:

Research Depth

Controls how extensively STORM searches your document library.

// Research Depth Options
{
  shallow: {
    documents: 5-10,
    timeSeconds: 15-30,
    detail: "Quick overview, limited depth"
  },
  moderate: {
    documents: 15-25,
    timeSeconds: 45-60,
    detail: "Balanced coverage, good for most reports"
  },
  deep: {
    documents: 30-50,
    timeSeconds: 90-120,
    detail: "Comprehensive, exhaustive research"
  }
}

When to use each:

  • Shallow: Quick summaries, draft documents, time-sensitive needs
  • Moderate: Most reports, standard documentation (recommended default)
  • Deep: Critical reports, comprehensive analyses, grant applications

Document Length

Target length for the final document.

SettingPagesSectionsWord CountGeneration Time
Brief3-53-51,500-2,5003-5 minutes
Standard8-125-84,000-6,0007-10 minutes
Comprehensive15-258-127,500-12,50012-20 minutes
Extensive30-5012-2015,000-25,00025-40 minutes

Length is a Target: STORM prioritizes quality over hitting exact page counts. Documents may be slightly shorter or longer based on available research material.


Citation Style

Choose how STORM formats references.

APA (American Psychological Association):

In-text: (Smith, 2024, p. 15)
Reference: Smith, J. (2024). Risk Assessment Methodologies. 
           Insurance Journal, 45(3), 123-145.

MLA (Modern Language Association):

In-text: (Smith 15)
Reference: Smith, John. "Risk Assessment Methodologies." 
           Insurance Journal, vol. 45, no. 3, 2024, pp. 123-145.

Chicago:

In-text: Smith 2024, 15
Reference: Smith, John. 2024. "Risk Assessment Methodologies." 
           Insurance Journal 45 (3): 123-145.

IEEE:

In-text: [1]
Reference: [1] J. Smith, "Risk Assessment Methodologies," 
           Insurance Journal, vol. 45, no. 3, pp. 123-145, 2024.

AI Model Selection

Choose the AI model based on quality vs. speed trade-offs.

// Available Models
const models = {
  "gpt-4-turbo": {
    quality: "Highest",
    speed: "Slower (15-20 min)",
    cost: "$$$",
    bestFor: "Critical reports, grant applications, legal documents"
  },
  "claude-3-sonnet": {
    quality: "Very High",
    speed: "Moderate (10-15 min)",
    cost: "$$",
    bestFor: "Most reports (recommended), policy documents"
  },
  "gpt-3.5-turbo": {
    quality: "Good",
    speed: "Fast (5-8 min)",
    cost: "$",
    bestFor: "Draft documents, quick summaries, internal memos"
  }
};

Model Comparison:

ModelQuality ScoreSpeedCostUse Case
GPT-4 Turbo95/100SlowHighProduction documents
Claude 3 Sonnet92/100MediumMediumDaily use (recommended)
GPT-3.5 Turbo82/100FastLowDrafts and quick tasks

4. Define Topic & Context

This is the most critical step—providing clear, detailed instructions ensures high-quality output.

Topic Field

What to include:

// Example: Risk Assessment Report
Topic: "Comprehensive property risk assessment for 123 Main Street, 
San Francisco, CA. Property value: $2.5M, built 1965, wood frame 
construction. Analyze flood, earthquake, and wildfire risks using 
FEMA HAZUS methodology."

// Example: Compliance Report
Topic: "SOX 404 compliance audit for Q4 2024. Focus on internal 
controls over financial reporting for publicly traded manufacturing 
company. Identify gaps in current controls and provide remediation 
plan with timeline."

// Example: Policy Document
Topic: "IT Security Policy for municipal government. Cover data 
classification, access controls, incident response, acceptable use, 
and BYOD guidelines. Align with NIST Cybersecurity Framework and 
state data protection regulations."

Best Practices:

Be Specific:

  • Include addresses, values, dates
  • Name specific regulations or standards
  • Specify scope and boundaries
  • Define audience (internal, external, regulator)

Provide Context:

  • Why the document is needed
  • What decision it supports
  • Who will review/approve it
  • Any constraints or requirements

Set Expectations:

  • Desired tone (formal, technical, accessible)
  • Critical sections that must be included
  • Information that should be emphasized

Avoid:

  • Vague descriptions ("write about insurance")
  • Missing critical details (address, dates, scope)
  • Conflicting instructions
  • Unrealistic expectations given document library

Key Points

List 3-8 key points that must appear in the document.

Example: Risk Assessment Report

Key Points:
1. Earthquake risk is primary concern due to proximity to San Andreas Fault
2. Property is in FEMA Flood Zone X (minimal flood risk)
3. VHFHSZ wildfire zone designation requires mitigation analysis
4. 1965 construction predates modern seismic codes
5. Estimated annual loss must include all three hazards
6. Mitigation recommendations should prioritize seismic retrofit

Example: Compliance Report

Key Points:
1. Current SOX 404 controls documented but not tested in 18 months
2. Three material weaknesses identified in previous audit
3. Management representation letters lack required sign-offs
4. IT general controls need strengthening (access provisioning)
5. Remediation must be complete before Q1 2025 audit

Tips:

  • Order by priority (most important first)
  • Be specific (not "discuss risk" but "earthquake risk from San Andreas Fault")
  • Include required conclusions or findings
  • Note any mandatory sections (e.g., "include mitigation costs")

5. Start Generation

Once configured, click "Generate Document".

Processing Time: Expect 5-20 minutes depending on document length and research depth. STORM displays real-time progress updates.


Generation Stages

STORM generates documents in transparent, iterative stages:

Stage 1: Research Phase (30-90 seconds)

What Happens:

  • Semantic search across your entire document library
  • Vector similarity matching to find relevant content
  • Source ranking by relevance score
  • Extraction of key passages and data

What You'll See:

🔍 Researching: "earthquake risk assessment methodologies"
📄 Found 47 relevant documents
📊 Analyzing: FEMA HAZUS documentation (relevance: 0.95)
📊 Analyzing: California seismic codes (relevance: 0.89)
✅ Research complete: 47 sources identified

Documents Searched:

  • Your uploaded RAG documents
  • Pre-loaded PublicRisk.ai knowledge base
  • Federal regulations and standards (if enabled)

Search Strategy:

# STORM's semantic search approach
def research_phase(topic, key_points, depth):
    # 1. Extract search queries from topic
    queries = generate_search_queries(topic, key_points)
    
    # 2. Semantic search with vector embeddings
    results = []
    for query in queries:
        embeddings = encode_query(query)
        matches = vector_search(embeddings, top_k=50)
        results.extend(matches)
    
    # 3. Deduplicate and rank by relevance
    sources = deduplicate(results)
    ranked = rank_by_relevance(sources, topic)
    
    # 4. Return top N based on depth setting
    return ranked[:depth_to_count[depth]]

Stage 2: Outline Creation (15-45 seconds)

What Happens:

  • AI analyzes research results
  • Identifies major themes and topics
  • Creates hierarchical document structure
  • Assigns content to sections
  • Balances section depth

Example Outline Output:

1. Executive Summary
   - Overall risk assessment
   - Key findings
   - Critical recommendations

2. Property Overview
   2.1. Location & Description
   2.2. Construction Details
   2.3. Occupancy & Use

3. Hazard Analysis
   3.1. Earthquake Risk
       - Fault proximity analysis
       - PGA assessment
       - Ground motion estimates
   3.2. Flood Risk
       - FEMA zone designation
       - Base flood elevation
       - Flood depth analysis
   3.3. Wildfire Risk
       - VHFHSZ designation
       - Ignition probability
       - Fuel load assessment

4. Loss Estimation
   4.1. HAZUS Damage Functions
   4.2. Expected Annual Loss
   4.3. Probable Maximum Loss

5. Mitigation Recommendations
   5.1. Immediate Actions (0-30 days)
   5.2. Short-term Initiatives (1-6 months)
   5.3. Long-term Improvements (6-12 months)

6. Implementation Plan
   6.1. Cost Estimates
   6.2. Timeline & Milestones
   6.3. Resource Requirements

7. Conclusion

8. Appendices
   8.1. HAZUS Methodology
   8.2. Source Documents
   8.3. Glossary

Outlining Algorithm:

// STORM's outline generation logic
function generateOutline(sources, key_points, template) {
  // 1. Extract topics from sources
  const topics = extractTopics(sources);
  
  // 2. Cluster related topics
  const clusters = clusterTopics(topics);
  
  // 3. Map clusters to template sections
  const sections = mapToTemplate(clusters, template);
  
  // 4. Ensure all key points are covered
  sections = ensureKeyPoints(sections, key_points);
  
  // 5. Order logically
  const ordered = orderSections(sections);
  
  // 6. Add subsections for depth
  const detailed = addSubsections(ordered);
  
  return formatOutline(detailed);
}

Stage 3: Outline Review & Approval

Your Turn! This is where you guide the document's direction.

Review Interface:

📋 Generated Outline (8 sections, estimated 15 pages)

✓ All 6 key points covered
⚠ Section 3.2 (Flood Risk) may be brief due to limited sources
✓ Balance: Even distribution across sections

[View Detailed Outline]
[Edit Outline Manually]
[Regenerate Outline]
[Approve & Continue]

Editing Options:

Add Section:

Click: [+ Add Section]
Title: "Cost-Benefit Analysis"
Position: After "Mitigation Recommendations"
Depth: Moderate

Remove Section:

Click: [✗] next to "Appendices"
Confirm: "Remove this section? Content won't be generated."

Reorder Sections:

Drag and drop sections to reorder
Or use ↑↓ arrows

Adjust Depth:

Section: "Hazard Analysis"
Current: Comprehensive (5 subsections, 6 pages)
Change to: Brief (2 subsections, 2 pages)

Important: Take time during outline review. Changes after content generation require full regeneration of affected sections.


Stage 4: Content Generation (5-20 minutes)

What Happens:

  • STORM writes each section individually
  • Synthesizes information from multiple sources
  • Maintains consistent tone and style
  • Adds in-text citations
  • Preserves technical accuracy

Real-time Progress:

📝 Writing Section 1/8: Executive Summary (15% complete)
📝 Writing Section 2/8: Property Overview (27% complete)
📝 Writing Section 3/8: Hazard Analysis (41% complete)
   └─ 3.1 Earthquake Risk ✓
   └─ 3.2 Flood Risk ✓
   └─ 3.3 Wildfire Risk (in progress...)
📝 Writing Section 4/8: Loss Estimation (55% complete)

Estimated time remaining: 8 minutes
[Pause] [Cancel]

Content Generation Process:

# STORM's content writing approach
def write_section(section, sources, context):
    # 1. Filter sources relevant to this section
    relevant_sources = filter_by_relevance(sources, section.topic)
    
    # 2. Extract key information
    key_info = extract_information(relevant_sources)
    
    # 3. Generate section outline
    section_outline = create_section_outline(key_info)
    
    # 4. Write introduction
    intro = write_introduction(section, context)
    
    # 5. Write each subsection
    content = []
    for subsection in section_outline:
        text = generate_text(subsection, relevant_sources)
        citations = add_citations(text, relevant_sources)
        content.append(text + citations)
    
    # 6. Write conclusion/transition
    conclusion = write_conclusion(section, context)
    
    return combine(intro, content, conclusion)

What Makes STORM's Writing Special:

Multi-Source Synthesis: Combines information from multiple documents into coherent narratives ✅ Citation Tracking: Every factual claim linked to source document ✅ Consistency: Maintains terminology and style across all sections ✅ Context Awareness: Later sections reference earlier findings ✅ Professional Tone: Formal, technical language appropriate for business documents


Stage 5: Quality Review (30-60 seconds)

What Happens:

  • Citation validation (all sources accessible?)
  • Consistency checks (terminology, formatting, abbreviations)
  • Completeness verification (all outline sections written?)
  • Coherence analysis (sections flow logically?)
  • Tone consistency (formal language throughout?)

Quality Checks Performed:

// STORM's quality review checklist
const qualityChecks = {
  citations: {
    allValid: boolean,        // All citations link to real sources
    properlyFormatted: boolean, // APA/MLA/Chicago format correct
    noOrphans: boolean        // No citations without sources
  },
  consistency: {
    terminology: boolean,     // "risk assessment" not "risk eval"
    abbreviations: boolean,   // Define before using
    formatting: boolean       // Heading styles uniform
  },
  completeness: {
    allSectionsWritten: boolean,
    keyPointsCovered: boolean,
    minimumLength: boolean
  },
  coherence: {
    logicalFlow: boolean,
    noContradictions: boolean,
    transitionsSmooth: boolean
  }
};

Example Quality Issue & Fix:

⚠ Issue: Section 3.1 uses "PGA" without defining
✓ Fixed: Added definition "(Peak Ground Acceleration, PGA)" at first use

⚠ Issue: Section 4.2 references "Table 3" but table not included
✓ Fixed: Added Table 3 with loss estimates

⚠ Issue: Inconsistent citation format in Section 5
✓ Fixed: Standardized all citations to APA format

Stage 6: Final Document Delivery

What You Receive:

📄 Complete Document with:

  • Cover page (title, author, date, organization logo)
  • Table of contents (auto-generated with page numbers)
  • Executive summary (if requested)
  • All content sections with headings
  • In-text citations throughout
  • References/bibliography page
  • Appendices (if configured)

Example Document Structure:

Risk Assessment Report.pdf (23 pages)

├─ Cover Page
├─ Table of Contents
├─ Executive Summary (2 pages)
├─ 1. Introduction (1 page)
├─ 2. Property Overview (2 pages)
├─ 3. Hazard Analysis (8 pages)
│  ├─ 3.1 Earthquake Risk
│  ├─ 3.2 Flood Risk
│  └─ 3.3 Wildfire Risk
├─ 4. Loss Estimation (3 pages)
├─ 5. Mitigation Recommendations (4 pages)
├─ 6. Implementation Plan (2 pages)
├─ 7. Conclusion (1 page)
├─ References (47 sources)
└─ Appendices
   ├─ A. HAZUS Methodology
   └─ B. Source Documents

Editing & Refinement

Built-in Editor

STORM includes a rich text editor for post-generation refinement.

Editing Capabilities:

Text Formatting:

  • Bold, italic, underline
  • Heading levels (H1-H6)
  • Bullet and numbered lists
  • Block quotes
  • Code blocks

Structure:

  • Add/remove sections
  • Reorder sections
  • Insert tables
  • Add images

Content:

  • Edit any text inline
  • Add custom paragraphs
  • Insert data from external sources
  • Modify citations

Example Edit Session:

// Edit specific section
editor.selectSection("3.1 Earthquake Risk");
editor.insertParagraph(
  "Recent USGS updates indicate increased seismic activity in the region.",
  position: "after:paragraph:2"
);

// Add table
editor.insertTable({
  section: "4.2 Expected Annual Loss",
  rows: 4,
  columns: 3,
  headers: ["Hazard", "Loss Estimate", "Probability"]
});

// Modify citation
editor.updateCitation({
  oldSource: "FEMA 2023",
  newSource: "FEMA 2024 (Updated)"
});

Regenerate Sections

If a section doesn't meet your needs, regenerate it without affecting the rest.

How to Regenerate:

Select Section

Click the section heading or use the section menu.

Choose Regeneration Mode

[Full Regeneration]  - Completely rewrite section
[Expand]             - Add more detail
[Condense]           - Make more concise
[Different Angle]    - Same content, different approach

Add Guidance (Optional)

Guidance: "Focus more on structural damage estimates. 
Include comparison to 1989 Loma Prieta earthquake."

Generate

STORM rewrites the section with your guidance. Other sections remain unchanged.

When to Regenerate:

  • Section is too brief or too detailed
  • Missing critical information
  • Wrong tone or emphasis
  • Better sources available now

Export Options

PDF Export

Standard PDF:

  • Professional formatting
  • Embedded fonts
  • Table of contents with links
  • Proper page numbers
  • Footer with document title

Settings:

{
  pageSize: "Letter" | "A4",
  margins: {
    top: "1in",
    bottom: "1in",
    left: "1in",
    right: "1in"
  },
  fontSize: 11 | 12,
  font: "Times New Roman" | "Arial" | "Calibri",
  colorMode: "color" | "grayscale"
}

Word Export (.docx)

Features:

  • Fully editable in Microsoft Word
  • Styles preserved (headings, lists, tables)
  • Track changes enabled
  • Comments preserved
  • Compatible with Word 2016+

Best For:

  • Documents requiring extensive editing
  • Collaborative review with track changes
  • Integration with existing Word templates
  • Final formatting in corporate style

Markdown Export

Use Cases:

  • Version control (Git-friendly)
  • Web publishing
  • Integration with documentation tools
  • Plain text workflows

Format:

# Risk Assessment Report

## Executive Summary

The property at 123 Main Street faces significant seismic risk...

## 1. Introduction

### 1.1 Purpose and Scope

This comprehensive risk assessment evaluates...

[Source: FEMA HAZUS 5.1 Documentation, 2024]

HTML Export

Features:

  • Responsive design (mobile-friendly)
  • Interactive table of contents
  • Syntax highlighting for code
  • Print stylesheet included

Best For:

  • Publishing to internal portals
  • Sharing via web link
  • Embedding in websites
  • Mobile viewing

Advanced Features

RAG Integration

STORM integrates with your Retrieval-Augmented Generation (RAG) system.

How RAG Enhances STORM:

RAG Benefits:

  • ✅ Search across thousands of documents in seconds
  • ✅ Find conceptually similar content (not just keyword matches)
  • ✅ Rank by relevance to your specific topic
  • ✅ Extract precise passages for citation
  • ✅ Update as you add new documents (no retraining)

RAG Settings:

{
  vectorStore: "ChromaDB",
  embeddingModel: "nomic-embed-text",
  dimensions: 768,
  topK: 50,                    // Return top 50 matches
  similarityThreshold: 0.7,    // Minimum relevance score
  chunkSize: 1000,            // Characters per document chunk
  chunkOverlap: 200           // Overlap between chunks
}

Multi-Model Support

STORM supports multiple AI models with different strengths.

Model Comparison:

ModelProviderStrengthUse Case
GPT-4 TurboOpenAIOverall qualityProduction documents
Claude 3 SonnetAnthropicLong context, analysisComplex reports
GPT-3.5 TurboOpenAISpeed, costDrafts, quick docs
Llama 3.1 70BOpenRouterOpen-source, privacySensitive content

Switching Models Mid-Generation:

// If generation is slow, switch to faster model
if (generationTime > 15 * 60) {
  storm.switchModel("gpt-3.5-turbo");
  storm.resume();
}

Template Customization

Create custom document templates for recurring document types.

Template Structure:

{
  "name": "Municipal Risk Assessment",
  "description": "Standard risk assessment for city properties",
  "sections": [
    {
      "title": "Executive Summary",
      "required": true,
      "maxLength": 500,
      "subsections": []
    },
    {
      "title": "Property Information",
      "required": true,
      "subsections": [
        {"title": "Location Details"},
        {"title": "Building Description"},
        {"title": "Occupancy Type"}
      ]
    },
    {
      "title": "Hazard Analysis",
      "required": true,
      "subsections": [
        {"title": "Flood Risk", "sources": ["FEMA NFHL"]},
        {"title": "Earthquake Risk", "sources": ["USGS NSHM"]},
        {"title": "Wildfire Risk", "sources": ["NIFC"]}
      ]
    }
  ],
  "requiredFields": ["propertyAddress", "propertyValue", "yearBuilt"],
  "citationStyle": "APA",
  "defaultLength": "standard"
}

Using Custom Templates:

Create Template

Navigate to Settings → Document Templates → Create New

Define Structure

Add sections, subsections, required fields

Set Defaults

Citation style, length, AI model

Save & Use

Template appears in STORM template dropdown


Best Practices

Writing Effective Topics

Good Topic Examples:

✅ "Comprehensive flood risk assessment for 123 Oak Street, Houston, TX. 
Property is in FEMA Zone AE with BFE of 52 feet. Analyze structure 
vulnerability (wood frame, 1985 construction) and estimate expected 
annual loss using HAZUS depth-damage curves."

✅ "SOX 404 internal controls audit for Q4 2024. Focus on IT general 
controls (access provisioning, change management, backup procedures). 
Identify deficiencies compared to COSO framework and provide remediation 
timeline with resource requirements."

✅ "IT Security Policy for 500-employee manufacturing company. Cover 
data classification, access controls, incident response, acceptable use, 
BYOD, and remote work. Align with NIST Cybersecurity Framework and 
include enforcement procedures."

Poor Topic Examples:

❌ "Write about flood risk" 
   (Too vague - which property? what level of detail?)

❌ "Insurance policy document"
   (No context - what type? for whom? what coverage?)

❌ "glp exc CA 2024"
   (Unclear abbreviations - spell out "general liability policy exclusions")

Optimizing Generation Time

Fast Generation (3-5 minutes):

  • ✅ Use GPT-3.5 Turbo model
  • ✅ Select "Brief" document length
  • ✅ Set "Shallow" research depth
  • ✅ Limit to 3-5 key points
  • ✅ Use simple templates

Quality Generation (10-15 minutes):

  • ✅ Use Claude 3 Sonnet model
  • ✅ Select "Standard" document length
  • ✅ Set "Moderate" research depth
  • ✅ Provide 5-8 key points
  • ✅ Review and approve outline

Comprehensive Generation (20-30 minutes):

  • ✅ Use GPT-4 Turbo model
  • ✅ Select "Comprehensive" document length
  • ✅ Set "Deep" research depth
  • ✅ Provide detailed topic context
  • ✅ Customize outline extensively

Source Document Quality

Ensure High-Quality Output:

Upload Authoritative Sources:

  • Federal regulations (FEMA, USGS, NIFC)
  • Industry standards (ISO, NIST)
  • Professional publications
  • Internal policies and procedures

Tag Documents Properly:

  • Use consistent tag taxonomy
  • Tag by topic, document type, and relevance
  • Update tags as content evolves

Keep Documents Current:

  • Remove outdated policies
  • Update regulations when changed
  • Archive superseded documents

Organize Document Library:

  • Group related documents
  • Use folders/collections
  • Maintain naming conventions

Troubleshooting

Generation Fails or Hangs

Symptom: STORM stops responding during generation

Causes & Solutions:

Insufficient Source Documents:

Error: "Could not find enough sources on this topic"

Solution:
- Upload more relevant documents to your RAG library
- Broaden your topic to match available sources
- Use "Shallow" research depth if sources are limited

Model Timeout:

Error: "Generation timeout after 30 minutes"

Solution:
- Reduce document length setting
- Switch to faster model (GPT-3.5)
- Simplify topic and key points
- Retry with "Resume" button (progress is saved)

API Rate Limits:

Error: "Rate limit exceeded. Please wait."

Solution:
- Wait 1-2 minutes and click "Retry"
- Switch to different model provider
- Contact admin to increase rate limits

Poor Quality Output

Symptom: Generated content is generic or lacks depth

Solutions:

Provide More Context in Topic:

Before: "Risk assessment for property"

After: "Comprehensive seismic risk assessment for 2-story wood-frame 
commercial building at 456 Elm St, San Francisco. Built 1972, no 
seismic retrofits. Analyze expected damage from M7.0 earthquake on 
San Andreas Fault using HAZUS building vulnerability curves for W1 
construction type."

Upload Better Source Documents:

  • Specific regulations for your jurisdiction
  • Recent risk assessment reports
  • Industry best practices
  • Technical standards

Increase Research Depth:

  • Change from "Shallow" to "Moderate" or "Deep"
  • STORM will search more documents and extract more information

Review and Edit Outline:

  • Add sections that are missing
  • Specify more detail in key sections
  • Remove generic sections

Citations Missing or Incorrect

Symptom: No citations or broken citation links

Causes & Solutions:

Source Documents Not in RAG:

Issue: Citations reference documents not in your library

Solution:
- Upload all referenced documents to RAG Management
- Wait for processing to complete
- Regenerate document

Citation Style Mismatch:

Issue: Citations formatted incorrectly

Solution:
- Select correct citation style in settings (APA, MLA, Chicago)
- Regenerate affected sections

Broken Links:

Issue: Citation links don't work

Solution:
- Ensure documents are still in your library (not deleted)
- Re-upload documents if needed
- Use "Refresh Citations" in document editor

FAQs

Q: How long does STORM take to generate a document?

A: Typical generation times:

  • Brief (3-5 pages): 3-5 minutes
  • Standard (8-12 pages): 7-12 minutes
  • Comprehensive (15-25 pages): 15-25 minutes
  • Extensive (30-50 pages): 30-45 minutes

Time varies based on AI model, research depth, and document complexity.


Q: Can I stop generation and resume later?

A: Yes. Click "Pause" at any time. Your progress is automatically saved as a draft. Resume from the Drafts menu.


Q: What happens if I close my browser during generation?

A: STORM continues generating in the background. When you return, you'll see the completed document or current progress. All progress is automatically saved.


Q: Can I use STORM without uploading documents?

A: STORM requires source documents to generate quality, cited content. However, it can access:

  • Your uploaded RAG documents
  • PublicRisk.ai's pre-loaded knowledge base
  • Federal regulations (FEMA, USGS, NIFC) if enabled

For best results, upload documents relevant to your specific topic.


Q: How do I improve generation quality?

A:

  1. Upload more relevant source documents to your RAG library
  2. Provide detailed topic descriptions with specific context
  3. List clear, specific key points that must be covered
  4. Review and improve the outline before content generation
  5. Use higher-quality AI models (GPT-4, Claude 3 Sonnet)
  6. Increase research depth to "Moderate" or "Deep"

Q: Can multiple users collaborate on a STORM document?

A: Yes. Share the draft with team members who can:

  • View and comment
  • Edit content
  • Regenerate sections
  • Export final versions

Use "Share" button to invite collaborators.


Q: Are STORM documents secure?

A: Yes. All documents are:

  • ✅ Encrypted at rest and in transit (AES-256)
  • ✅ Access-controlled by user permissions
  • ✅ Audited (all access logged)
  • ✅ Compliant with SOC 2 Type II standards
  • ✅ Never used to train AI models
  • ✅ Deleted permanently when removed

Q: Can I customize document templates?

A: Yes. Administrators can create custom templates via Settings → Document Templates. Define sections, required fields, default settings, and citation styles.


Q: How accurate are STORM's citations?

A: STORM citations are highly accurate:

  • ✅ Every citation links to a real source document
  • ✅ Page numbers verified
  • ✅ Quotes are exact (not paraphrased without attribution)
  • ✅ Sources ranked by relevance and authority

However, always verify critical information in source documents before publication.


Q: Can STORM generate documents in other languages?

A: Currently, STORM supports English only. Multi-language support is planned for future releases.


Support & Resources

  • Documentation: This guide + User Guides in PublicRisk.ai docs
  • Video Tutorials: Available in Help Center
  • Support: Contact your administrator or support@publicrisk.ai
  • Community: PublicRisk.ai User Forum
  • API Access: For programmatic document generation, see API documentation

Ready to Generate? Navigate to the STORM Document Generator and create your first AI-powered document in minutes.

On this page