See llms.txt for all machine-readable content.

Back to Templates

Send QA-checked spoken site audits with Apify, DeepSeek, ElevenLabs, Gmail and Sheets

Created by

Created by: Khairul Muhtadin || khmuhtadin
Khairul Muhtadin

Last update

Last update 2 days ago

Categories

A prospect asks for a site review and usually waits days for a PDF. This workflow takes one URL from a form, crawls the site, measures what the pages actually contain, writes a short audit with the fixes that matter and emails back a spoken version next to the written list. It also skips repeat requests, catches crawls that fail or return too little text, and holds weak AI drafts for a human before anything goes out.

Last updated: November 2026.

Quick Overview

A hosted form collects the site URL, the requester email and the audit focus. A Code node validates the URL and the address, and bad requests are logged and answered. Before any Apify credit is spent, the audit_log tab is checked so the same site is not crawled twice in a week. The Apify website content crawler reads up to six pages with three retries, and a failed crawl is logged and reported. A coverage gate stops sites with almost no readable text. DeepSeek writes the audit as JSON, a Code node checks it, and drafts with too few issues or a bad narration go to an audit_review tab instead of the client. Passing audits are voiced with ElevenLabs, emailed with the mp3, and logged with the Gmail message id. If the voice step fails, the written audit still goes out. Every Monday a digest email counts what happened.

How it works

  1. Audit Request Form, Normalize Request and Request Usable? collect and validate the request. Invalid ones go to Log Bad Audit Request and Email Fix Request.
  2. Read Recent Audits and Check Recent Audit look for a finished audit of the same domain in the last 7 days. Audited This Week? answers repeats with Email Recent Audit Notice and the last top issue.
  3. Crawl Site Pages runs the Apify crawler (6 pages, depth 1, cheerio) with retries. Its error output feeds Log Crawl Failure and Alert Crawl Failure. Get Site Pages pulls the dataset.
  4. Shape Audit Input measures words per page, thin pages, missing titles and missing meta descriptions. Enough Pages Read? needs at least one page and 150 words, otherwise Log Thin Crawl and Email Thin Crawl Notice explain why.
  5. Audit The Site returns a headline, summary, four to eight issues and a 90 to 140 word narration. Parse Audit builds the email.
  6. Check Audit Quality needs three or more issues, a top issue and a 60 to 200 word narration. Failures go to Queue Audit For Review and Email Audit Reviewer.
  7. Speak Site Audit makes the mp3 and Email Spoken Audit sends it. If the voice step errors, Email Text Only Audit sends the written version. Log Site Audit records status, delivery type and message id.
  8. Weekly Audit Report reads audit_log, Summarize Audit Week counts sent, rejected, failed and thin audits plus the most common top issues, and Email Audit Digest sends the summary.

Setup

  1. Connect Apify in Crawl Site Pages and Get Site Pages.
  2. Connect your DeepSeek credential (OpenAI compatible) on Audit Model.
  3. Connect ElevenLabs in Speak Site Audit and pick a voice. The default is Laura.
  4. Connect Gmail in all email nodes and replace [email protected] with your own address.
  5. Connect Google Sheets. Create audit_log with created_at, site_url, requester_email, pages_crawled, issue_count, top_issue, audio_file, status, delivery and message_id, plus audit_review with created_at, site_url, requester_email, issue_count, narration_words, qa_problems, draft_body and status.
  6. Send one test request against a site you own, then adjust the 7 day repeat window or the QA thresholds in the Code nodes if needed.

Quick Answers

What if someone submits the same site twice?
Check Recent Audit finds a finished audit of that domain from the last 7 days, and the requester gets the date and top issue instead of a second crawl.

What happens when the AI writes a weak audit?
Check Audit Quality blocks drafts with fewer than three issues, no top issue or a narration outside 60 to 200 words. They land in audit_review and a reviewer gets an email.

Does the audit check page speed or mobile rendering?
No. It only sees page text, titles, meta descriptions and headings, and the agent is told not to claim checks it did not run.

What if the voice step fails?
Speak Site Audit retries three times, then its error output sends the written audit without audio and the log marks it text_only.

Additional info

Built with n8n. Need an assessment on your business? Feel free to reach out at https://khmuhtadin.com/consultation/