- Integrations
- Zapier
Check facts and citations in any Zap.
The Lenz app for Zapier adds a verification step to any Zap: before the send, before the publish, before the row is written. A sourced verdict with score and citations comes back as mapped fields you can branch on. No HTTP configuration: you sign in to Lenz and approve the access, with no key to paste.
- AppIn Zapier’s directory
- SetupSign in with Lenz
- StepsOne trigger, six actions
- OutputMapped fields
01How Verify runs in a Zap
Verify a Claim waits for a signed callback.
02What you can build
Any Zap that moves content has a place to verify it.
Check AI-drafted copy before the Zap sends it
Use the verdict to route to publish, revise, or your reviewers.
Screen claims pulled out of documents, forms or pages
Check claims from documents, form submissions, or scraped pages before they reach a downstream system.
Assess a batch of claims cheaply
Then escalate only the uncertain ones to the full pipeline.
Start a Zap when a verification finishes
Fan the result out to Slack, a sheet, or a review queue.
03Trigger and actions
One trigger and six actions, each with its own fields.
| Step | Time | Returns |
|---|---|---|
| New Verification CompletedTrigger · polling | — | Fires for every deep check your account completes, whichever surface ran it: this Zap, the website, the MCP server or another connection. Filter on domain or verdict for a subset. verification_id, claim, verdict, confidence, lenz_score, key_finding, executive_summary, suggested_rewrite, created_at. |
| Verify a Claim | ~90 s | The full pipeline: framing, research, debate, panel review, conclusion. Fields detailed below. |
| Assess (Fast) | ~15 s | status (ok | no_claim) and claims[], one entry per claim found in the text, each with claim, verdict, confidence, passed, verification_url, and a reviewer’s note in rationale. A claim that could not be checked comes back with verdict Error, an error_code saying why and a hint; those rows are free. No score, summary, or sources. |
| Extract Claims | free | Pulls the verifiable claims out of a block of text and checks nothing: claim (the primary one), identified_claims, domain, key_entities, presumed_intent. Text can also be a single public web page URL (http or https, nothing else in the field): Lenz reads the page, a Google Doc, a PDF (.pdf), a YouTube video's transcript, or an image (.png, .jpg, .jpeg, .webp), up to its first 50,000 characters. Pages behind a login can't be read. A page read can take longer than a Zap step's 30 seconds; if the step times out, paste the page's text instead. |
| Ask Follow-Up | — | answer: a reply grounded in the research behind a completed Verify a Claim result. Pick a recent verification from the dropdown, or map a verification_id from an earlier step. |
| Review a Draft | 2–4 min | A whole draft in one step: every claim is quick-checked, and the ones that look wrong or uncertain are deep-checked. outcome (clean | issues_found | incomplete | unchecked), clean, and issues[], each with claim, verdict and a suggested_rewrite that is not verified itself, plus a key_finding on the deep-checked ones. A claim the deep check found not on the public record is not an issue and not clean: the outcome is incomplete, never clean. Maximum Citations to Check also checks the draft’s sources (blank skips them; up to 20). credits_charged says what it cost. Waits on a signed callback, like Verify a Claim. |
| Check Citations | under 2 min | Reads each source a draft links to (links, bare URLs, DOIs, [1]-style references) and checks that it says what the draft attributes to it. citations[] with a finding each, the problem ones in citation_issues[], and clean. Up to 20 citations per run, the first in the draft; citation_limit_reached says when there were more. 1 credit per citation checked; one Lenz could not read is free. Waits on a signed callback. |
Clicking Test in the Zap editor returns example output so you can map fields without spending credits on a claim you don’t care about. A turned-on Zap checks the input you actually give it.
04Verify a Claim output
Branch on status. Then on passed.
- Check
statusfirstOnlycompletedcarries a verdict.needs_inputmeans the text held several claims and arrives with areasonand amessage;failedarrives with anerror. - Use
passedto approve or escalateThe ready-made boolean. Addverdictorlenz_scorefor a finer threshold. When no public record settles the claim,verdictreadsNo Public Recordandpassedandlenz_scorearenull: an answer that is neither a pass nor a fail, and not a False. Route it to your reviewers. - Send
key_findingto peopleOne sentence, the short form for Slack or email.executive_summaryis the longer write-up;sourcesis an array, each with atitleand aurl. - Read
suggested_rewritebefore you use itThe claim with its wrong part corrected, written from the findings for a person to review. It is not verified itself, and it is empty when there is nothing to correct. - Keep
verification_idAsk Follow-Up needs it, and it pulls the full reasoning trace from the API later.
Verify a Claim output · JSON
{
"status": "completed", // completed | needs_input | failed
"passed": true, // true for True and Mostly True; null for No Public Record
"verdict": "True", // five labels, True to False, or No Public Record
"lenz_score": 9, // integer 1–10; null for No Public Record
"confidence": "high", // high | medium | low
"key_finding": "...", // one sentence, for Slack or email
"executive_summary": "...", // the longer write-up
"sources": [...], // each with title and url
"suggested_rewrite": "", // not verified; empty when none
"verification_id": "ab12cd34", // look it up later via the API
"task_id": "2f8b2e2b..." // the pipeline run
}
05Getting started
Three steps to a working verification step.
Add a Lenz step to a Zap
Pick a trigger or an action, then choose Connect a new account.
Sign in and approve
Sign in to Lenz and allow the access Zapier asks for. There is no key to paste and no webhook secret to set up.
The app lives in Zapier’s app directory, so there is no manual install. View the Lenz app on Zapier ↗
One Lenz account works across every surface: Zapier, n8n, the MCP server, the SDKs, and the REST API. Credits come from your account balance, not from each integration.
06Other ways to connect
The same verdicts, wherever you work.
- Claude and ChatGPTAdd Lenz to the assistant you already use. One sign-in, no key to paste.
- Google DocsCheck a draft from the Extensions menu. The findings appear beside your text.
- MCP serverClaude Code, Cursor, VS Code, Codex and other coding tools: one command, then sign in.
- n8nA community node with fourteen operations, from a quick check to a full verification.
- API & SDKsThe CLI and the REST API, with typed Python and TypeScript SDKs.
Verify AI output before your Zap sends it.
Self-serve from day one · free tier, no card