Verify AI output before anything acts on it.
Your product writes documents, summaries, answers — and increasingly acts on them. Lenz fact-checks that output claim by claim, against real independent sources, before a wrong summary becomes a wrong recommendation becomes a wrong action.
A hallucination in an AI-generated output isn't a UX flaw. It's a compliance flag, a liability event, or a customer lost. Validate LLM output the way you validate everything else you ship.
A document in. A verdict per claim out.
The real flow: /extract splits the text into atomic claims, then each is verified independently.
“Looking at the data, Einstein had a remarkable career. He won the 1921 Nobel Prize in Physics for his theory of general relativity. His famous equation E=mc² was first published in 1905, the same year he completed his PhD at the University of Zurich.”
Each verdict carries cited sources and the full reasoning trail.
Independent verification, with receipts.
An outside check — not your own model grading its own work. Lenz checks the claims with multiple frontier models from rival vendors, grounded in real curated sources: an adversarial debate surfaces the disagreement, and three independent reviewers adjudicate on the evidence.
Every verification ships the full trace: framing, sources, citations, debate, and the panel's reasoning. When a customer, auditor, or regulator asks how the content was checked, you have an independent, third-party record.
Where teams verify AI output.
Customer-facing documents & memos
AI-drafted briefs, reports, and client deliverables get fact-checked before customers act on them. /extract + /verify on the final draft; the audit trail is your due-care record.
AI-generated content
Articles, summaries, and marketing copy written by models, checked claim by claim before publication. Wire it into the pipeline with automated fact-checking.
Chat & agent products
Gate outbound answers with /assess (~5-10s fits a chat-completion timeout); escalate low-confidence claims to /verify in the background.
Verify a document in 10 lines
Python pip install lenz-io
# AI-generated document in — claim-by-claim verdicts out from lenz_io import Lenz client = Lenz(api_key="lenz_...") claims = client.extract(text=ai_document).identified_claims for claim in claims: v = client.verify_and_wait(claim=claim) print(v.verdict, v.lenz_score, claim) # False 1 Albert Einstein won the 1921 Nobel Prize ... # True 9 Albert Einstein first published E=mc² in 1905.
No code? The Workbench runs the same pipeline from the browser.
Pricing
Start free, no card required.
Free
$0
/extract · 1,000/day
/assess · 100/mo
/verify · 10/mo
/ask · 20/mo
Prototype today, zero spend. No card required.
Developer
$99/mo
/extract · 1,000/day
/assess · 5,000/mo
/verify · 500/mo
/ask · 1,000/mo
Self-serve. $999/yr annual.
Scale
$399/mo
/extract · 1,000/day
/assess · 20,000/mo
/verify · 2,000/mo
/ask · 4,000/mo
$3,990/yr annual. For production integrations.
Enterprise
Volume beyond Scale, SLAs, white-label, custom integration support.
Frequently asked questions
/extract pulls the verifiable claims out of the output, /assess gives a fast 3-model panel verdict per claim, and /verify runs the full sourced verification with citations. Python and TypeScript SDKs wrap all three.
Yes — that's the canonical flow. /extract splits a document, report, or memo into atomic verifiable claims (resolving pronouns and compound sentences), then each claim is verified independently. You get a claim-by-claim verdict list for the whole document.
Yes. The pipeline verifies factual claims regardless of who wrote them — AI output is simply where the volume and the risk concentrate.
A verdict (True, Mostly True, Mixed, Mostly False, or False), a 1-10 Lenz Score, cited sources graded for authority, and the full reasoning trail — research, adversarial debate, and three independent reviewer assessments.
Every verification ships a complete audit trail — framing, sources, citations, debate, and panel reasoning. When a customer, auditor, or regulator asks how a claim was checked, you have the record.
By use case
Stop shipping unverified AI output.
Self-serve from day one. Free tier, no card required.