SEO audit automation

A marketing agency's website audit went from 4 to 7 hours of hand work per site to under an hour of review. The workflow I delivered runs nine tools and builds the slide deck, with AI drafting the findings.

Marketing

Project snapshot

Industry
Marketing
Input modalities
Website and SEO tool outputs
The first steps of the audit workflow, each marked with a tick: merge the inputs, normalize the input, create a browser session, create a folder, create an Airtable row and update the run result link.
How a run starts, from a test run in October 2025. The workflow tidies the input, opens a browser session, creates a folder for the screenshots, adds a row to the table and records the link to the results.

The situation

Each audit meant running nine separate tools by hand, taking screenshots, writing up the findings and building the deck. That took 4 to 7 hours per site.

What I recommended and set up

The agency's earlier automation had run into problems. I rebuilt it so that entering a site starts everything. The tools run, the screenshots land in the agency's slide template, and AI drafts the findings for a person to review.

What changed after delivery

  • Before: 4 to 7 hours per site to collect results, take screenshots, write the analysis and build the deck.
  • After: under an hour of review per site.
  • A list of sites can be queued and run as one batch.
  • Each deck is a draft for a person to review before it goes out.

The result

  • 70 to 85% less time to produce an audit.

How the audit was done by hand

A marketing agency sends website audits to prospective clients. Building one meant running nine separate tools by hand: page speed on mobile and desktop, accessibility, privacy, security, search visibility against competitors, and the technology the site is built on. Someone took a screenshot of each result, wrote up what it meant, and assembled the deck.

That took 4 to 7 hours per site.

What had gone wrong before

The agency had already tried to automate it, and that attempt had run into problems. When I started:

  • Screenshots broke whenever a tool changed its interface.
  • One tool was blocked by a CAPTCHA, and another was blocked for a reason nobody had found.
  • New entries in the agency's table did not start the workflow.
  • The agency had been advised that screenshot output might not be possible in a first version, and was considering HTML output instead.

What I set up

  • One place to start. The team adds a site to a table, or uploads a whole list. Each entry starts a run.
  • The tools run themselves. For each tool, the workflow opens a browser session, signs in where the tool needs it, runs the check on the site and takes the screenshot. For one tool I switched to its official API, which removed the CAPTCHA problem.
  • Everything is filed. Screenshots go to the agency's Google Drive, with links recorded against the site.
  • The deck builds itself. The workflow copies the agency's own Google Slides template and places each screenshot, resized and cropped to fit.
  • A first draft of the findings. For each audit screen, AI drafts the findings, following the agency's own prompt rules.
  • A queue for batches. A list of sites runs one after another, so a long list stays within the platform's concurrency limits.
  • Logs. Every run is logged and monitored.

Each deck is a draft for a person to review before it goes to a prospect. The workflow does the collecting and the first draft. The agency keeps the judgment.

How it went

  • September 2025. Work started on the earlier workflow.
  • Mid-December. Screenshot capture was complete. The findings had not been started.
  • Week ending 19 December. Findings were done for 4 of the 11 audit screens, and bulk runs had been redesigned around the platform's concurrency limits.
  • Week ending 26 December. A complete run with no manual steps: site in, deck out.
  • January 2026. The engagement closed.

Before and after

  • Before: 4 to 7 hours per site.
  • After: under an hour of review per site.
  • 70 to 85% less time to produce an audit, and three to five times the audit throughput.

The agency rated the work 5 out of 5 on Upwork (opens in a new tab) and endorsed it as "Committed to Quality", "Solution Oriented" and "Accountable for Outcomes".

More screens from this project

An automation workflow laid out as rows of connected steps, one row for each check, all joining at the end.
The audit workflow in n8n, zoomed out, from a working copy in October 2025. Each row runs one check.
A completed workflow run that searches for pending sites, loops over them, starts an audit for each, waits, and marks each one processed.
The queue that works through a list of sites one after another. I covered the web address under one step.