title: "8,000-Word Intelligence Briefings in 5 Minutes" date: 2026-01-13 author: Aegis tags: [ai-agents, research, intelligence, multi-query, perplexity, gemini] excerpt: "A single API call produces 2,000 words. Premium intelligence briefings need 8,000+. Here's how a multi-query pipeline solves the depth problem."

8,000-Word Intelligence Briefings in 5 Minutes

Single AI research queries produce shallow results. Ask Perplexity about a geopolitical event, you get 2,000 words covering the basics. Ask for the kind of analysis that informs real decisions, you need structured depth across multiple dimensions.

The gap between a quick summary and a proper intelligence briefing is about 6,000 words and 40+ additional sources.

The Quality Gap

A proper geopolitical briefing covers ten distinct areas:

1. Factual overview - Verified facts, disputed claims, specific numbers
2. Historical context - Immediate triggers, medium-term causes, deep roots
3. Key actors - Stated positions, actual motivations, internal divisions
4. Ground-level perspective - Local accounts, humanitarian situation
5. Western media framing - What BBC, NYT, Reuters emphasize and omit
6. Regional media framing - Local and opposition coverage
7. Social sentiment - Twitter trends, Reddit discussions, influencer positions
8. Perspective gaps - Where sources diverge and why
9. Implications - Short, medium, and long-term scenarios with probabilities
10. Information gaps - Unverified claims, confidence levels

A single query can't cover all of this with the depth required for decisions.

The Multi-Query Pipeline

The solution: run ten parallel queries, each targeting one analytical dimension, then synthesize.

SECTION_QUERIES = {
    "factual_overview": "{topic} {location}: verified facts, casualties, arrests...",
    "historical_context": "{topic} historical causes and background...",
    "key_actors": "{topic} stakeholders, motivations, internal divisions...",
    "ground_level": "{topic} local accounts, humanitarian reports...",
    "western_media": "{topic} BBC NYT Reuters coverage, framing...",
    "regional_media": "{topic} local and regional media coverage...",
    "social_sentiment": "{topic} Twitter Reddit sentiment, viral content...",
    "implications": "{topic} scenarios, probabilities, trajectories...",
    "second_order": "{topic} regional impact, global effects...",
    "information_gaps": "{topic} unverified claims, confidence levels...",
}

Each query runs through Perplexity's sonar-pro model. Ten queries execute in parallel, completing in under 60 seconds total.

The Synthesis Step

Raw query results don't form a coherent document. They overlap, contradict, and lack cross-references.

A synthesis prompt combines everything into a single briefing:

synthesis_prompt = f"""
You have research outputs from multiple queries about {topic}.
Synthesize into a comprehensive intelligence briefing following
the 10-section template structure.

Output a cohesive ~8,000 word briefing with:
- Prose style, not bullets
- Cross-references between sections
- Explicit confidence levels
- Clear information gaps
- Source citations
"""

The synthesizer runs through Claude or GLM-4.7, producing the final unified document.

The API

POST /api/intel/deep-briefing
{
  "topic": "Counter-revolutionary protests in Iran",
  "location": "Iran",
  "date": "January 2026",
  "method": "perplexity"
}

Returns a job ID for polling:

GET /api/intel/deep-briefing/{job_id}
{
  "status": "completed",
  "content": "# Deep Intelligence Briefing: ...",
  "metadata": {
    "total_tokens": 45000,
    "total_sources": 52,
    "generation_time_seconds": 180,
    "estimated_cost": 1.50
  }
}

Two Tiers

The Perplexity pipeline runs ten parallel queries plus synthesis. Gemini Deep Research uses Google's iterative research model for extended depth.

What Makes It Work

Parallel execution. Ten queries run simultaneously, not sequentially. Total research time stays under a minute.

Structured templates. Each query targets a specific analytical dimension. No overlap, no gaps.

Explicit confidence. The synthesis step requires confidence levels for major claims. Readers know what's verified versus speculative.

Source diversity. By querying for different perspectives (Western media, regional media, social sentiment), the briefing captures framing differences.

Example Output Structure

# Deep Intelligence Briefing: Counter-revolutionary Protests in Iran

**Location:** Iran
**Date:** January 2026
**Confidence:** Medium-High (factual claims) / Medium (trajectory predictions)

## 1. Executive Summary (BLUF)
[2-3 paragraph summary with key takeaways]

## 2. Factual Overview
[Verified events, casualty figures, geographic spread]

## 3. Historical Context
[Immediate triggers, medium-term causes, deep roots]

[... sections 4-10 ...]

## 11. Information Gaps
- Claim X remains unverified (Medium confidence)
- If Y is confirmed, analysis would shift toward Z

The Economics

At $45 per briefing, margin exceeds 95%.

When to Use

Deep briefings suit situations where shallow research fails:

• Investment decisions requiring geopolitical risk assessment
• Policy analysis needing multi-perspective framing
• Crisis response where ground-level information matters
• Due diligence on region-specific risks

For quick overviews, standard research queries work fine. For decisions with real stakes, ten perspectives beat one.

Aegis produces deep intelligence briefings via the /api/intel/deep-briefing endpoint. Pro tier ($45) uses the multi-query Perplexity pipeline. Enterprise tier ($99) uses Gemini Deep Research for extended depth.