Skip to main content
Jon Legene Kolstrup

Rødovre, DK2026

I do the marketing and build the tools.

1

Digital marketer, working AI-native. Strategy, SEO and content, with the tooling built to match. Some of it off the shelf, some of it mine.

See my workGet in touch

Tools

Marketing stack

  • Google Search Console
  • Google Ads
  • GA4
  • Klaviyo
  • Meta Business Suite
  • Shopify

Tech stack

  • Next.js
  • TypeScript
  • Tailwind
  • Supabase
  • WordPress
  • WooCommerce
  • PHP
  • Claude Code

Capability

What I do

Marketing is the job. Building the tools for it is how I do the job.

SEO and content
Technical SEO, Search Console analysis, paid search, and content built to be found. Audits, the fixes that follow them, and the measurement that shows whether anything actually moved.
AI-native execution
I run the work with AI agents end to end: research, drafting, auditing, and a verification pass after every batch. Not a novelty. It has run a full working day unattended, off a written work queue, and the queue was empty at the end of it.
Tools I build myself
When the tool does not exist, I build it. Render pipelines, content generators, and an orchestrator that runs many coding agents at once.
Web development
Next.js and TypeScript where a site should behave like an application. Custom WordPress themes and WooCommerce where the client has to run it themselves after I leave. Built, not assembled out of plugins, which is why they stay fast.
Design systems
Visual identity, type and colour systems, and then the part most people skip: encoding them so the tooling can build against them. A design system that only lives in a PDF gets ignored by the third asset. One that agents can read produces work that matches the site without anyone checking.

Analysis

Checking whether the numbers mean what they say

The part of the job that produces no deliverable. It is usually the part that changes the decision.

  1. 01

    A metric measured against reality

    An ad account reported more phone calls from advertising than the entire organisation received from every channel combined. The conversion action was counting menu playbacks at the call centre, not people. Nothing in the platform said so. It came out of putting the account's number beside the company's own call log and noticing they could not both be true.

  2. 02

    Ranking is not demand

    Over a hundred commercial searches, positions in the top three, and almost no clicks across three months. The reporting implied the fix was more content. The measurement said position was never the problem, so more content would have been three months spent on the wrong thing.

  3. 03

    Reading what was inherited

    A negative keyword list handed over with an account turned out to be blocking the exact searches that converted. Nobody had read it since it was written. Auditing something dull that everyone assumes is fine is reliably where the cheap wins are.

  4. 04

    Saying what a method cannot show

    A competitor study across a few hundred posts measured distribution, not sales. That limit went into the document in writing, next to the findings. A study that quietly implies more than it measured is worse than no study, because someone will act on it.

  5. 05

    Refusing to guess

    A few hundred locations geocoded against a national address register, delivered with every record that could not be resolved returned as unknown with a reason. Not one coordinate was guessed. A dataset that is honest about its gaps can be used; one that quietly fills them cannot.

Rule

Agents are fast and confidently wrong. So nothing counts until it has been checked against the source.

Method

How I work

  1. Frame it

    Work out what the actual question is, and what evidence would answer it. Most of what looks like a content problem turns out to be a measurement problem first.

  2. Run it with agents

    The work goes out to AI agents in batches: research, drafting, auditing, fixes. I write the brief, set what done looks like, and build the tool when the one I need does not exist.

  3. Verify every batch

    Nothing counts until it is checked against the source. Agents are fast and confidently wrong, so every batch gets verified before the next one starts. That step is the whole reason this way of working holds up.

Selected work

Who

Jon Legene Kolstrup

Jon Legene Kolstrup

Digital marketing specialist. SEO, GEO and AI-native execution.

I work in-house in digital marketing. Before that: e-commerce, agency, and a stretch teaching Generative Engine Optimization. This site is where the work that does not fit a job description lives.

More about me

Inquiries

Ready to transform your brand?

Book a free 15-minute strategy call. No sales pitch, just value.

Inquiries
Based in
Copenhagen, DK