Kacper Rafalski  SEO · GEO · AEO
About

Kacper Rafalski,
SEO consultant.

I’ve worked in SEO for 16 years. I started at an agency in 2010, have run my own consultancy since 2014, and since 2018 I’ve led SEO and automation at a large consulting company. Alongside that, I take on projects for selected clients in Poland and abroad, in English and Polish.

Toolkit

My own tools, not off-the-shelf reports

I write Python and run most of my analysis on tools I built myself. Scripts pull data through APIs – Search Console, GA4, backlink tools, and the CMS – join it with a crawl, and let me repeat the same analysis across thousands of URLs instead of checking them by hand.

I also build MCP servers that let AI assistants work directly on the data from those tools. Automation takes the repetitive part of the work, and my time goes to judgment and decisions.

  • Python
  • Search Console API
  • GA4 API
  • SEO tool APIs
  • CMS APIs
  • MCP servers
  • Command-line crawling

What I do

My day-to-day work

Audits and migrations

I crawl large sites, group findings by template, and take changes from a developer ticket to the recrawl that confirms the fix.

Content with quality control

I run a refresh program for old articles and an AI-assisted content process where every number needs a source and every change is measured in Search Console before and after.

Content map and internal linking

I keep an inventory of every page on the site with its intent, traffic, and links, so cannibalization is caught early and internal linking is planned on data.

Multilingual sites

I work on sites in English and Polish: content localization, hreflang, and separate keyword targets for each market.

What I’ve learned

Three lessons that changed how I work

  1. 01

    A traffic drop isn’t always a ranking drop

    When I dug into a large organic traffic drop, most of the lost clicks were on keywords where rankings hadn’t moved. AI answers above the results took the traffic. Since then, I check what actually changed before I start fixing.

  2. 02

    A recommendation without a sign-off test comes back next audit

    I read three audits of the same site spanning six years. The same problems came back in each one, because nobody could check whether the previous fixes were done. That’s why every recommendation I make has a test you can run.

  3. 03

    AI writes fast, and makes things up with confidence

    AI-assisted drafts can contain invented anecdotes, quotes, and statistics that sound credible. I learned to treat every number as a claim to verify: it has a source or it goes. I apply the same rule in client content audits.

Ground rules

What I don’t do

  • I don’t guarantee rankings or AI citations – nobody controls those.
  • I don’t sell links. I only take on link acquisition in ongoing engagements, and only for pages where the data shows links are really missing.
  • I don’t invent numbers for reports. If I don’t know something, I say which data is missing.
  • I don’t hand over reports that end up in a drawer – every recommendation has an owner and an acceptance criterion.