Fact-check the draft before it goes out.

Fact-check AI-generated newsletters, briefs, posts and reports before publishing.

1. 1

   The draft is ready

   An issue, a brief, a post, a report — whether a model wrote the first pass or you did.

2. 2

   Lenz finds the checkable statements

   You hand over the whole piece rather than a list. Lenz picks out the factual statements it can check against independent public sources and leaves the opinions and the recommendations alone.

3. 3

   Each one comes back with a result

   An assessment returns a verdict and a confidence level per statement in about 10 seconds; that is how you sort a whole issue. Send the ones that matter to a full verification, where the verdict comes from sources Lenz went and found.

4. ✓

   You decide what to publish

   Read the results before the issue goes out: cut a line, fix a figure, or run a full verification on the one you are least sure of. The editorial call is yours.

5. 5

   Publish, with the checks behind it

   Each full verification keeps a record you can reopen: the statement as framed, the verdict and score, every source with its date, the reasoning. A figure questioned later has a check behind it.

Try it on your own draft →

Market briefs, news digests and articles

Paste the whole piece. Lenz returns a result per statement, not one verdict for the document, so a wrong figure is named with its own sources.

Blog posts, press releases and social media posts: the statistics in them

A statistic is checked like any other statement: Lenz looks for where the number was published and reports what it found, including when that is nothing.

Check ChatGPT-drafted content and X or LinkedIn posts inside Claude or Cursor.

Connect Lenz to Claude with one sign-in, no key; Cursor and Claude Code take one command with your key. Paste the post or the draft and ask for a check; the results come back in the same window.

Connect Claude →

Automate checks on recurring LLM output: briefs, digests, documents.

For a daily brief or a batch of model-drafted documents, a short Python script: extract the statements, assess them together, route False, Mixed and low-confidence results to your reviewer. The routing is your policy.

# a recurring draft: extract the statements, assess them together, route the flagged ones
from lenz_io import Lenz

client = Lenz(api_key="lenz_...")
extracted = client.extract(text=draft)
# identified_claims is empty when the text yields a single statement
claims = extracted.identified_claims or [extracted.claim]
r = client.assess(claims=claims)   # one row per statement, same order

# your routing policy. "Error" means the item was not checked — never read it as clean
flagged = [c for c in r.claims
           if c.verdict in ("False", "Mostly False", "Mixed", "Error")
           or c.confidence == "low"]

send_to_reviewer(draft, flagged)   # your own function: an email, a ticket, a Slack message

A scheduled digest or a batch of documents runs unattended: Lenz posts each result back when it lands, so nothing polls, and a retried request never runs the same check twice.

What a full verification looks like.

A full verification is the investigation, not a longer assessment. Lenz searches independent public sources for the statement, has models argue both sides on what it found and a panel review the arguments, then returns the verdict with those sources and the reasoning. About 90 seconds per statement.

Frequently asked questions

Check the next issue, statement by statement.