Skip to content

How to fact-check claims

The /factcheck command verifies factual claims using web search and LLM analysis. It extracts claims from text, searches for evidence, and provides verdicts with explanations and sources.

What is fact-checking?

When you use /factcheck:

  1. Extracts claims from your text (or uses selected text)
  2. Searches the web for evidence supporting or refuting each claim
  3. Analyzes results using an LLM to determine a verdict
  4. Displays results in an interactive factcheck node with verdicts, explanations, and source links

Using factcheck

Method 1: Fact-check selected text

  1. Select a node containing claims you want to verify
  2. Type /factcheck in the chat input
  3. Optionally add text after the command (e.g., /factcheck verify these)
  4. Press Enter

The factcheck feature extracts claims from the selected node's content and verifies them.

Method 2: Fact-check typed claims

  1. Type /factcheck followed by your claims
  2. Press Enter

Example:

/factcheck The Eiffel Tower is 330 meters tall. Paris is the capital of France.

Method 3: Fact-check a numbered list

If you have a note node with numbered claims:

  1. Select the note node containing the numbered list
  2. Type /factcheck (or /factcheck verify these)
  3. Press Enter

The feature automatically extracts each numbered item as a separate claim.

Claim selection modal

If more than 5 claims are found, a modal appears for you to select which claims to verify:

  • Select All / Deselect All - Toggle all claims at once
  • Individual checkboxes - Select specific claims
  • Warning - If you select more than 5 claims, a warning appears (informational only)
  • Execute - Starts verification for selected claims

Verdict types

Each claim receives one of these verdicts:

Verdict Badge Meaning
VERIFIED ✅ Claim is accurate and supported by reliable sources
PARTIALLY_TRUE ⚠️ Claim is mostly correct but contains inaccuracies or missing context
MISLEADING 🔶 Claim is technically true but presented in a misleading way
FALSE ❌ Claim is factually incorrect
UNVERIFIABLE ❓ Cannot determine truth due to lack of reliable sources
ERROR ⚠️ Verification failed due to an error

Reading factcheck results

The factcheck node displays:

  • Claim text - The original claim being verified
  • Status badge - Visual indicator of the verdict
  • Explanation - Brief explanation of why this verdict was reached (1-2 sentences)
  • Sources - Links to supporting evidence (up to 3 sources per claim)

Expanding claims

Click on any claim header to expand and see the full explanation and sources. Click again to collapse.

Tips

Vague references

If you use vague references like "verify this" or "fact check these", the feature will:

  1. Use the selected node's content as context
  2. Resolve the vague reference into specific claims
  3. If resolution fails, use the context directly

Best practices

  • Be specific - Clear, factual statements work best
  • Numbered lists - Automatically extracted as separate claims
  • Select nodes - Provides context for vague references
  • Multiple claims - Can verify up to 10 claims at once (5 recommended for performance)

Performance

  • 5 or fewer claims - Verified immediately in parallel
  • More than 5 claims - Modal appears for selection (you can still select all)
  • Parallel processing - All selected claims are verified simultaneously

Examples

Example 1: Simple claim

/factcheck The speed of light is 299,792,458 meters per second

Result: ✅ VERIFIED with explanation and sources.

Example 2: Multiple claims

/factcheck Water boils at 100°C at sea level. The Earth is flat. Python is a programming language.

Result: Three claims with different verdicts (✅, ❌, ✅).

Example 3: Numbered list in note

  1. Select a note node containing:
1. The Eiffel Tower was built in 1889
2. Paris is the capital of France
3. The moon is made of cheese
  1. Type /factcheck verify these

Result: Three claims extracted and verified (✅, ✅, ❌).

Limitations

  • Search quality - Verdicts depend on search result quality
  • LLM analysis - Verdicts are AI-generated and should be verified
  • Source limits - Up to 3 sources per claim
  • Search queries - Up to 3 search queries per claim
  • Claim limit - Maximum 10 claims per factcheck