Back
AI systems builder · SEO

Most of my SEO work now starts in Claude.

This is the AI system I run my SEO practice on - designed, built, and refined over the last year. It turns recurring problems into workflows: figuring out why traffic moved, comparing pages that rank, mapping search demand to pages, turning recommendations into implementation, and checking what happened afterward.

SEO is the domain I know cold - so it's where I proved the approach. The approach travels.

Explore the workflows ↓
Why I started doing this

The handoff between tasks was always the interesting part.

I've always found the handoff between SEO tasks more interesting than the tasks themselves.

An audit ends. Someone else picks up the keyword research. A strategist writes the brief. A writer gets the brief. A developer gets a ticket. A month later, someone pulls a report.

Before
AUDIT
KEYWORD RESEARCH
BRIEF
TICKET
REPORT
A lot can disappear along the way.
After - the reasoning stays closer together
Context
Measure
Understand
Research
Analysis
Decide
Build
Verify
back to Measure
How a piece of work moves

Five stages. The useful part is the handoff.

Measure, understand, decide, build, verify - then start again. The five stages are simple; what passes between them is the point. Scroll to walk through them, and flip the switch to see the methodology resolve into the actual Claude skills.

Methodology Claude
One problem, start to finish

Watch a problem move through the stages.

Pick something a client might say. Nothing here is an answer - it's how the work moves: check what happened, narrow down where, work toward a cause, then follow the decision into implementation and back to measurement.

A few workflows

A few of them, in more detail.

Root cause analysis
A traffic change is a starting point, not an explanation.

I first check that the change is real, then narrow down where it happened, build possible explanations, and try to disprove them. Only then do I settle on a cause and decide what to do about it.

View workflow →
Ranking analysis
Two pages can target the same query and perform very differently.

I built a workflow for comparing them systematically: the pages themselves, the SERP, intent, content, links, entities, and the other factors that might explain the difference.

View workflow →
Keyword mapping
A keyword list sitting in a spreadsheet isn't the useful part.

The useful output is a set of page-level decisions: what belongs together, what deserves its own page, what an existing page should own, and where the gaps actually are.

View workflow →
Technical SEO audit
A crawl can produce thousands of findings. That doesn't mean thousands of things worth doing.

This workflow takes crawler output and works toward a smaller set of issues that can be investigated, prioritized, and tested.

View workflow →
AI visibility
A separate set of workflows around how brands and entities appear in AI-generated answers.

It's still developing. I'm interested in what can actually be measured, what makes a useful comparison, and where the limits of the available data are.

View workflow →
Where the work connects

Take keyword research.

On its own, it produces information about searches. Mapped to pages, it becomes a set of decisions. Those decisions can feed on-page work, headings, metadata, internal links, schema, and content. Once the work is shipped, verification gives you something new to measure. tap to replay

"running shoes"
keyword-research
keyword-map
↓ feeds
on-page-seo
h-tags
metadata
schema
internal-links
content
implementation
verification

That's why I care about the connections between workflows.

A useful output shouldn't have to be recreated by the next person who touches the work.

A couple of rules I keep coming back to
Don't make up the number.

If the source doesn't contain it, say so.

Prove the symptom first.

A drop in organic traffic doesn't automatically mean an SEO problem.

Look at the SERP.

The company the client considers a competitor may have very little to do with the pages actually competing for the query.

Know what a page is for.

Before changing a page, I want to know what it should own.

Built in Claude

Where the workflows come together.

Claude is where I've been putting these workflows together. Each one has a defined job, the information it needs, rules around evidence, and a useful output. Some feed directly into other workflows.

66Workflows
8Areas
5Stages
1Feedback loop
Actively evolving Last updated · August 2026

There are 66 of them at the moment, covering 8 areas of SEO work. That number will change. Search below, or open the full library.

Browse the full library →
Still building

I keep adding to it.

I add to the library when I run into a problem often enough that it seems worth working out properly. Some workflows are finished enough to use. Others are experiments. A few have already changed considerably from their first version.

That's what I'm documenting here.