Fact-Check Draft Articles with Perplexity Before You Publish
An n8n workflow that takes a draft article, extracts its factual claims, asks Perplexity to verify each one against live sources, and logs a pass/fail report with citations to Google Sheets — so writers catch wrong stats and outdated facts before they go live, not after a reader does.
- 1
Receive the draft
Add a
Webhooknode (POST) that accepts a JSON body with the drafttitleandtext. Send drafts to it from a Google Docs add-on, a form, or your CMS's 'ready for review' hook. - 2
Extract the claims
An
HTTP Requestto Claude asks for strict JSON: an array of discrete factual claims (stats, dates, attributions, superlatives), ignoring opinion. This turns prose into a checkable list. - 3
Verify each claim with Perplexity
An
Item Listsnode splits the claims, then anHTTP Requesttosonar-prochecks each: 'Is this claim accurate as of today? Verdict: supported / contradicted / unverifiable. Cite sources.' The live search is what catches outdated numbers. - 4
Shape the report rows
A
Codenode builds one row per claim: the claim text, the verdict, a short note, and the citation URLs. Colour-coding by verdict in the sheet later makes the failures jump out instantly. - 5
Log to Google Sheets
A
Google Sheetsnode appends the rows under a section for that article. The writer opens the sheet, filters tocontradictedandunverifiable, fixes those lines in the draft, and publishes with confidence.
Frequently asked questions
How does it decide what a 'claim' is?
A first Claude (or Perplexity) call extracts discrete, checkable factual statements — numbers, dates, attributions, superlatives — and ignores opinion or fluff. Each extracted claim is then verified individually so the report is granular. You review only the flagged ones, not the whole article.
How is a claim marked pass or fail?
For each claim, Perplexity returns a verdict (supported / contradicted / unverifiable) plus the sources it found. The workflow writes all three states to the sheet. 'Contradicted' and 'unverifiable' are the rows a writer must look at; 'supported' with a good source can be trusted.
How do I feed a draft into it?
Any way you like — the trigger is flexible. The example uses a Webhook you can POST a draft to (from a Google Docs add-on, a form, or your CMS), but you could also watch a Google Drive folder or a 'Ready for check' status in a Notion/Sheets content calendar. The draft text is all the workflow needs.
Auto-Turn Every New YouTube Video into Instagram Reels Clips with AI
An n8n workflow that watches your YouTube channel, sends each new upload to an AI clipping service, and drops three vertical, caption-ready Reels clips into Google Drive — so one long video becomes a week of Instagram content without any manual editing.
Free recipe
Promote Every New YouTube Upload Across Email, Twitter and Discord
The moment you publish a video, automatically announce it to your email list, tweet a thread, and post in Discord — one upload, full promotion.
Free recipeDraft AI Replies to New YouTube Comments for One-Click Approval
Never leave a YouTube comment unanswered — AI drafts a friendly, on-brand reply to each new comment and sends it to you in Slack for approval.