Your legal AI is flying partially blind.

New research tested RAG systems on complex documents.

Every legal professional using AI should be aware of these results.

When humans marked relevant evidence on pages, AI models only found 16-17% of it.

83% of this supporting content could be invisible to your AI.

Tables, charts, and infographics are the hardest content for AI to process.

Legal documents are full of these.

Flowcharts in patents, timelines in litigation briefs, or tables in financial disclosures

What they call "multi-hop queries" perform worst.

These are questions requiring synthesis across multiple documents.

Many legal AI tools rely on text extraction.

They strip out visual context that makes documents comprehensible.

The gap between what AI retrieves and what it should is 10+ percentage points.

That's not a rounding error.....it's missing evidence.

Before trusting AI summaries of complex legal documents, ask:

Can it actually see what you're seeing?