← All articles
Build log

Automating SEO with AI: what we actually automated (and what we deliberately didn't)

Nick van der Blom5 min read

Automating SEO with AI: what we actually automated (and what we deliberately didn't)

Search for "automate SEO" and you get two kinds of answers. Tools that promise you'll publish a hundred blog posts with one click. And agencies that tell you AI is "an assistant" and show nothing concrete beyond that.

This article is the third kind. We've been running AI systems on real client accounts since 2024. This is what works, what we deliberately never automated, and where you start yourself.

Search Consolefetch queriesClusterby search intentSERP checkrank per clusterReport + briefsdecisionsqueries ↑clusters ↑rankings ↑draft ↑Client gives feedbackquality round, alwaysLiveautomaticAUTOMATIC · runs on scheduleCLIENTAUTOMATICWhat you see: one content decision, from raw search data to publication. Nothing goes live before the client has looked.Feedback? It is worked into the article; after that the pipeline still publishes automatically.
The SEO pipeline: four automatic steps, one feedback round with the client, then live automatically.

What SEO automation is and isn't

SEO automation means software carries out recurring SEO work without someone having to be there every time. That's something different from "AI writes my copy". The copy is the smallest part of the work. The biggest part is analyzing, checking and deciding. That's where the time goes, so that's where the gains are.

Our starting point: an AI agent may read everything, may propose, and may only execute within boundaries that were set in advance. Autonomy is earned per task, not granted up front.

Three things that run fully automated for us

1. The technical site audit is a single command

Before: start a crawl, open the exports, manually lay the index data next to the crawl, a list a day later. Now: one command. The agent crawls the site, compares what it finds with what Google has actually indexed, and flags what's broken: noindex errors, broken canonicals, orphaned pages, redirect chains.

What this delivered in concrete terms: at one client, the audit found a site-wide noindex that was keeping an entire country site out of Google. Nobody had noticed, because the pages were simply there. They were just invisible.

2. Search Console data is automatically turned into content decisions

This is the pipeline we're proudest of, because it replaces the work that costs an SEO specialist the most time:

  1. Fetch: all queries from Search Console for the past period.
  2. Cluster by intent: the agent groups thousands of queries into topics. Not by word overlap, but by what the searcher wants.
  3. SERP check: per cluster, the agent looks at where the client actually stands in the search results right now, and who does rank.
  4. Report: a client report with the gaps, the opportunities and a proposal per cluster.

What used to be a day of spreadsheet work is now a pipeline that runs while we do something else. The result isn't a "list of keywords" but a list of decisions: we're missing this topic, here we're on page two, here a competitor wins with a category page.

3. Content briefs in the client's house style

The clusters produce briefs. The agent writes one brief per topic, with the search intent, the subtopics the SERP expects, the internal links and the client's tone of voice. A second system then renders the draft page in the client's house style, so the preview looks like a page on their own site. The client doesn't review a Word document, but a page that looks like what will go live. After the client's feedback, the page goes to the site automatically.

What we deliberately did not automate

This is the part the tool vendors skip.

Going live before the client has looked. Publishing itself is automated: the article goes from the pipeline straight to the site, including an image from the image library. What we don't do is push it live before the client has seen the draft page and given feedback on it. Purely for quality reasons: the client knows their products and their customers better than any model, and we work that feedback into the article before it goes live. We've seen what happens without that step: an agent wrote a product claim that wasn't true. The feedback round caught it. Since then, that round isn't an option but a rule.

Strategy. Which topics matter to the business, what margin a product group carries, where the company is headed: the agent doesn't know that and shouldn't guess it either. It lives in the business context we give it.

Link building. Anything that communicates externally on the client's behalf is done by people.

Why this works: three ingredients

An AI system that takes over SEO work needs three things, and most failed attempts are missing one.

Ingredient What it is What goes wrong without it
Operator expertise The SEO knowledge lives in the instructions and checklists the agent follows The agent does "something" with data, but not the right thing
Cadence The work runs on a schedule (weekly, monthly), not whenever someone remembers The system becomes a demo that ran once
Business context Product range, margins, USPs, what the client never wants to claim Generic output the client doesn't recognize

Where to start yourself

You don't have to start with a pipeline. Start with one task you do by hand every week and that only requires reading.

  1. Pick a read-only task. For example: "summarize this week's Search Console data and name the three biggest shifts."
  2. Write down what good looks like. What you want to see, what the agent must never conclude without evidence.
  3. Let it run for four weeks and compare with what you would have spotted yourself.
  4. Only then add a task that makes proposals. Execution comes after that, with boundaries.

Anyone who reverses this order and lets an agent publish right away without someone watching learns the same lesson we learned, just at a higher price.

Frequently asked questions

Can AI take over SEO completely? No. AI takes over the analyzing, checking, preparing and even the publishing. Deciding, the client feedback round and anything that communicates externally on the client's behalf remain human work.

Which tools do you use? A combination of Claude (via Claude Code), the Search Console API, a crawler and our own scripts. The tool matters less than the instructions and the boundaries you give it.

Does this also work for a small online store? Yes, especially. The pipeline scales down. For a small site, the site audit and the Search Console analysis cost less than an hour of compute per month.

What does it cost to have this built? That depends on what's already in place. We always start with a free audit of your current setup, so you know where the gains are before you build anything.


Want to know which SEO tasks at your online store are the first to automate? Request a free audit. We're honest about whether it makes sense, and fast when it does.

Nick van der Blom

Nick van der Blom

Founder of OnestoMedia. Builds AI systems that run the marketing work of online stores: Google Ads agents, SEO pipelines and the command center that brings them together.

Next articleAutomating Google Ads with an AI agent: how an online store's account runs for a month without a manual change