Sara & Nik’s SEO Podcast:
Dan Petrovic

About this conversation

This episode of Sara & Nik’s Podcast welcomes Dan Petrovic, founder of DEJAN. He has been in SEO for roughly 20 years, and his research on AI search is circulating across the industry.

The topic is what happens between a user’s question and an AI model’s recommendation, and how much of that SEOs can influence. Dan’s answer has two sides. AI search opens up a new layer of on-page work, but it all rests on a foundation SEOs already know:

“You can’t ignore traditional SEO because it drives the AI.”

From Accidental Black Hat to Midlife Retirement and Back

Dan never really planned to pursue a career in SEO. Science was his first passion, and when he arrived in Australia in 1998, a guidance counselor steered him toward multimedia. Around the same time, he was performing live electronic music and running a website about it.

That website changed everything. A stranger from Germany offered him $500 a month to place a link on it. Curious, Dan installed the Google toolbar and found his site had a PageRank of 7. He recollects that event as the moment he began his SEO career as a black hat.

What followed was a network of 1,000 domains, a paid-off mortgage, and an agency that grew to 60 people, which he eventually shrunk down because “it’s not a good thing to have a big agency”. Nik agreed, explaining that he has kept Get Stuff Digital deliberately lean for the same reason.

In January 2013, Dan published an article predicting that people would one day chat with Google instead of searching it. Then he stepped away for about a decade. He spent those years as a full-time dad, working on science projects and gardening, until an invitation to speak about AI at SMX Munich brought him back.

With LLMs, he no longer needed a dev team to execute his ideas, which brought Sara to sum up why nobody should be retiring right now:

“Who would retire in the right mind in this time? It’s so fun, right?”

How to Judge SEO Research

Before speaking about AI SEO on stage, Dan spent three years learning how models work by building one from scratch. That meant collecting training data, training a tokenizer, picking a loss function, and making many mistakes along the way.

The result was a model that wasn’t very good, but at least there was no mystery about LLMs left for him. Compared to the old days of SEO, he sees this as a gift:

“Google’s complete blackbox. What we have now is a dream come true truly for all SEOs.”

His frustration is with people who scratch the surface, draw conclusions, and walk away. For clients trying to tell real experts from pretenders, his advice is simple:

“Just keep asking how. And if they break down, if they don’t have an answer for the building blocks, the core mechanics, the first principles of it, then you know it’s smoke and mirrors.”

Dan also welcomes criticism when it’s respectful and backed by evidence. This prompts Sara to push back on how the community handles disagreement in practice. She describes it as cliquey, a place where good research gets ignored while popular voices win by default:

“It really depends on how popular you are right and that’s how truth becomes defined”

Dan’s answer is a moderated comment system on his blog. It uses an LLM to score each criticism, rejects personal attacks, and even flags commenters who shift the goalposts mid-argument.

Grounding Snippets, Links, and the Originality Paradox

Nik brought up a conversation with John Mueller at a Google event in Barcelona: start with fundamentals, because complex tactics won’t save a rotten foundation. Dan agrees. For him, 90% of AI visibility is good SEO.

The remaining 10% happens after search. Dan describes the process as follows:

• A complex prompt is split into fan-out queries.
• Google pulls the most relevant chunks from each ranking page.
• Those chunks are stitched together into a grounding snippet, and the model works only with that snippet.

Dan now tests this by removing sentences from the snippets and measuring the impact:

“I sometimes remove a sentence and the rank jumps up. So these are literal negative signals in your grounding snippet that drag your page down in the generative rankings.”

On links, Dan says they still work when they’re authentic. His team runs a red team / blue team workflow. One AI agent writes the content, and a hostile reviewer tries to guess who wanted the link. If the reviewer guesses right, the article gets thrown away.

As for PR campaigns aimed at getting into training data, Dan is skeptical, since labs now train on curated and synthetic data:

“You’re adding a few drops to an ocean.”

His advice is to worry less about Reddit and external sources and focus on your own pages. That leads to the paradox Dan finds hardest to accept – Content that is too original can underperform, because “it always gravitates towards the average of its training data.”

The SEO’s job, in his view, is to strike a balance between three audiences: users, search engines, and now LLMs.

To close the episode, Dan offered a bold prediction, delivered “just for fun”: by 2037, AGI will be a reality, and it will come out of China.