Julian Goldie SEO: Who I Am, How I Do SEO, And Why I Run GoldieBench

Julian Goldie SEO is me, Julian Goldie, a Bangkok-based SEO entrepreneur who founded Goldie Agency and who runs this benchmark, GoldieBench.

I've done SEO professionally for over a decade, and today most of my SEO is done by a crew of AI agents.

This page explains who I am, how I got into SEO, what I've built and how I run SEO now.

Then it explains why the person running an SEO agency also runs an AI model leaderboard.

At the end you'll find every funnel I run, marked free or paid, plus the models my SEO agents use and how they score here.

Who is Julian Goldie SEO?

"Julian Goldie SEO" is the handle I use on YouTube and X, so it's the name most people find me by.

Here are the facts in one table.

My longer personal story is on my about page, and the general version of this article is at Julian Goldie: who I am.

My SEO story: from a zero-traffic crypto site to a 70+ person agency

I didn't plan to get into SEO.

While I was travelling around Asia, I built a cryptocurrency website and wrote content for it for months.

It got zero traffic from Google.

That taught me that content on its own doesn't rank, so I taught myself link building.

I applied link building to my own affiliate sites, and the rankings moved.

Once it worked for me, I realised I could sell the same result to other site owners.

I started Goldie Agency completely solo, with no team and no investor.

Over the years it grew into a 7-figure SEO and link building agency with 70+ employees in Bangkok.

I wrote the playbook down in Link Building Mastery, and my Udemy courses have taught 29K+ students.

The agency still publishes its client case studies openly on goldie.agency.

One of them shows a legal services client growing from 263 to 3,157 monthly visitors between June and December 2025, with top-three keyword positions rising from 19 to 108.

How Julian Goldie SEO works today: AI agents do the heavy lifting

In mid-2025 I shifted my focus from SEO tactics to AI agents that do the SEO for you.

That's when I built the Agent Operating System, or Agent OS.

It's a dashboard that runs a crew of AI agents under one shared memory, and it runs on macOS, on Linux and through the web.

Hermes, Claude, GLM and Kimi are the core of the crew, and each one gets the jobs it's best at.

My SEO runs through it as a simple loop.

I set the strategy and check what goes live, and the agents do the repetitive work.

The full walkthrough is in my Claude AI SEO guide on agentos.guide.

The SEO results behind the name

These numbers come from my own Google Search Console, and each one is shown on screen in a public guide.

I've never logged into aiprofitboardroom.com, because my Agent OS documents what we test and publishes it for me.

These are my results on my sites, so treat them as evidence rather than a promise for yours.

The community has its own proof, because the AI Profit Boardroom was named the number one AI community by FatRank after testing 27 communities, and by Autoblogging.ai and a Hermes Agent panel.

Why an SEO person runs an AI model leaderboard

This is the question I get most on this site.

The answer is that my SEO now depends on AI models, so picking the right model is a business decision for me.

The leaderboards everyone reads, like MMLU, HumanEval and SWE-bench, measure narrow things that vendors optimise for.

They don't tell me whether a model can ship a finished thing in one prompt.

So I built GoldieBench to answer that question.

Every model gets the same fixed prompt inside the Agent OS, and it gets one shot with no iteration and no hints.

Whatever comes out on the first try is what you see, and every demo is live and playable.

Each result gets a 0 to 10 score, and the reasoning is posted with it, which you can read on the methodology page.

No vendor pays me, and no bad result gets buried.

The bench is a slice of my daily work, because the same dashboard that runs my SEO is the one I use to test new models.

What I've built, all in one place

Here's everything that sits under my name today.

The Agent OS isn't a public GitHub project, because it's delivered inside the AI Profit Boardroom as an installable system.

How to copy the Julian Goldie SEO approach this week

You don't need my whole stack to start, and you don't need 70 people either.

Here's the small version I'd build if I were starting again tomorrow.

First, export your Search Console queries and find five keywords where you already get impressions but almost no clicks.

Second, record yourself doing real work on one of those topics, even if it's just a screen recording with your voice over the top.

Third, write a one-page skill file with your rules for titles, search intent, voice and schema.

Fourth, give one AI agent the transcript and the skill file, and ask for a post that answers the keyword in the first line.

Fifth, read the draft yourself, cut anything your real work doesn't back up, and publish it.

Sixth, submit the URL for indexing and check Search Console again a week later.

Once one post works, you add more agents, more sites and more keywords, and that's when it starts to compound.

When you pick the model for each agent, use a leaderboard that tests finished output rather than multiple-choice answers.

That's exactly the gap GoldieBench was built to fill.

What GoldieBench is not

I want to be straight about the limits of the bench, because I run it and I also sell training.

GoldieBench scores are my scores on my fixed prompts, so they measure one-shot builds rather than everything a model can do.

A model that iterates well over ten turns can look weaker here than it feels in a long agent session.

Some models have fewer scored tasks than others, so a small sample can move an average a lot.

Every demo is live, so you can play any build and decide for yourself whether you agree with my score.

Where to find me: every Julian Goldie SEO funnel

Here's every official link with one line on what it's for.

The models my SEO agents use, and how they score here

This section comes last on purpose, because it's about the models rather than about me.

GoldieBench doesn't have an SEO writing task, so these scores measure one-shot builds rather than blog posts.

They still matter for SEO, because the same models build my landing pages, tools and site templates.

The closest task to SEO work is the landing page task, where every model was asked for a modern marketing landing page in one shot.

Here's how I read that table for SEO work.

Fusion and Claude Opus 5 sit near the top of the average board, so they're my picks when the output has to be right first time.

GLM-5.2 tied for gold on the landing page task, which is a strong result for a model that's open weights and free for individuals.

MiniMax M3 averages 7.97 at $0.30 per million input tokens, so it's the value pick for high-volume jobs.

Kimi K3 scored 8.6 on the landing page, so it earns its place on builds.

If you want to compare them yourself, open the full model list or the local models board for free options you can run on your own machine.

Also On Our Network

• Agent Operatorshow I run SEO with AI agents every day
• AI Income Deskthe business model behind Julian Goldie SEO
• AI Tool Verdicta checkable fact sheet on whether Julian Goldie SEO is legit
• agentos.guidethe Agent OS workflow that runs my SEO
• aiprofitboardroom.comwhy the AI Profit Boardroom was named the best AI community

Ground every page in real work, test the models before you trust them, and you'll be doing SEO the way Julian Goldie SEO does it.