How is Rabbit Hole different from ChatGPT Deep Research or Perplexity?

Three differences. First, Rabbit Hole uses 10 specialist agents searching in parallel vs one model doing sequential queries -- so it's faster and deeper. Second, a contrarian agent stress-tests every finding before synthesis, catching hidden assumptions and gaps. Third, the output is a downloadable report with embedded diagrams and verified citations, not a chat response. Stanford found Perplexity fabricates 26% of references and ChatGPT 40%. Rabbit Hole verifies every citation before you see it.

What is adversarial verification?

Before you see any report, a contrarian researcher agent reviews all findings. It looks for hidden assumptions, unstated dependencies, what would falsify the thesis, and steel-mans the opposition. Then a separate citation verification hook checks that every factual claim has a real, linked source. This two-layer approach catches the blind spots and hallucinations that single-model research tools miss.

How does pricing work?

Pricing is per month based on how many research reports you need. Free gives you 3 reports to try it out. Basic is $39/month for 15 reports, Plus is $99/month for 40, and Team is $499/month for 100 reports. Every plan includes all 10 specialist agents and adversarial verification. No per-seat fees, no surprises.

What sources does Rabbit Hole search?

10 specialist agents search different source types: arXiv and Semantic Scholar for academic papers, Reddit and Hacker News for community sentiment, X/Twitter and LinkedIn for social signals, SEC EDGAR for financial filings, GitHub and Stack Overflow for technical content, plus news and company sites. Each agent is optimized for its domain -- the academic researcher follows citation graphs differently than the community researcher analyzes Reddit sentiment.

Can I use this for professional work?

That's exactly what it's built for. Consultants use it for competitive landscapes and client deliverables. VCs use it for due diligence. Grad students use it for literature reviews with BibTeX export. The verified citations and confidence ratings mean you can actually cite the output in professional documents -- something you can't safely do with tools that fabricate references.

Why not just use Claude Code or ChatGPT to do this myself?

You could. It would take about 50+ hours. You'd need to set up MCP servers for arXiv, Reddit, SEC EDGAR, Hacker News, and finance APIs. Then build a multi-agent orchestrator with parallel delegation. Then design a contrarian review pipeline. Then wire up citation verification. Then build report formatting with SVG diagram generation. Then tune prompts for each specialist. Then keep it all working as APIs change. Rabbit Hole is that entire stack, already built and tested. At $39/month, it's cheaper than the API tokens you'd burn debugging it.