API-first AI verification

Verify every AI answer
before you ship it

Paste any AI response. TruthPilot cross-references it against trusted sources and returns a structured verdict — confirmed, uncertain, or false — with cited evidence.

No human review required Sub-second verification JSON or plain text
truthpilot.verify()
YOUR AI RESPONSE
"The Apollo 11 mission landed on the Moon on July 20, 1969 and the crew returned safely on July 24."
VERIFICATION RESULT
✓ Apollo 11 landed July 20, 1969 — NASA History Office
✓ Crew returned July 24, 1969 — Apollo 11 Mission Report
VERIFIED 2/2 claims confirmed

From prompt to verdict in three steps

No manual research. No browser tabs. Just accurate verdicts, fast.

01

Send the AI response

POST the text you want verified to the TruthPilot endpoint. Supports any AI output — ChatGPT, Claude, Gemini, your own model.

POST /v1/verify { "text": "[AI response here]" }
02

We extract and cross-check

TruthPilot breaks the response into discrete claims, then cross-references each against trusted sources — news, academic databases, official records.

Processing... Extracting claims: 3 found Checking: [######] 3/3
03

Get a structured verdict

Every claim is labeled confirmed, uncertain, or false — with source URLs, confidence scores, and the specific text flagged.

{ "verdict": "confirmed", "claims": [ { "text": "...", "status": "confirmed", "source": "..." } ] "confidence": 0.94 }

Three verdict types. No gray area.

Every claim gets a definitive label. Your users know exactly where they stand.

VERIFIED

Claims match trusted sources

The AI's statement is confirmed by at least two authoritative sources. Safe to use.

"The Eiffel Tower was completed in 1889"
Confirmed — UNESCO Archives, Paris City Records
UNVERIFIABLE

Not enough data to confirm or deny

Sources are conflicting, sparse, or the claim is too specific to verify. Flag it for human review.

"The 2016 study showed 40% improvement"
Unverifiable — study date, journal, and sample size not found
FLAGGED

Contradicted by authoritative sources

The AI hallucinated or misrepresented a fact. This claim should not be published without correction.

"Lightning never strikes the same place twice"
False — documented instances show the Empire State Building is struck ~20-25 times per year

Built for developers from day one

A RESTful API that fits into any stack. One endpoint. Predictable response shape. No rate limiting surprises.

  • SDKs for Python, Node, Go, Ruby
  • Sub-second response time
  • 99.5% uptime SLA
  • Webhook for async verifications
Response shape JSON 
{
  "verdict: "confirmed",
  "confidence: 0.97,
  "claims: [
    {
      "text: "Apollo 11 landed on July 20, 1969",
      "status: "confirmed",
      "source: "NASA — Mission History",
      "source_url: "https://..."
    ,
    {
      "text: "Crew returned July 24, 1969",
      "status: "confirmed",
      "source: "Apollo 11 Mission Report",
      "source_url: "https://..."
    
  ],
  "verdict_id: "vrd_xK9mN..."
}

Every AI answer is a claim.
Not every claim is true.
TruthPilot makes the difference visible.

Built for the AI-first world — where shipping fast means shipping accurate. TruthPilot is the trust layer your AI product needs.

72% AI hallucination rate in production deployments
3x More trust when sources are cited
<1s Average verification time