Cover of Chat With Dan Petrovic | How AI Search Really Works, Model Bias & What Actually Moves The Needle

Chat With Dan Petrovic | How AI Search Really Works, Model Bias & What Actually Moves The Needle

Chat With Dan Petrovic | How AI Search Really Works, Model Bias & What Actually Moves The Needle

Listen for free

View show details

Season's Savings | 0.99 €/month for 3 months

5.99 €/month thereafter - terms apply. Cancel monthly.
Summary

Welcome back to Search With Sean — another Chat With Sean, and a guest I'd wanted on for a long time: Dan Petrovic of DEJAN, one of the world's most respected SEO experimenters and the person who actually tests how AI search works instead of guessing. This one goes deep.


Dan's current obsession is cutting through the measurement noise — everyone's tracking AI visibility, but then what? He's gone back to on-page optimisation and structured testing: AB-testing content against models like Gemini to find the "model-preference signals" that genuinely move visibility (one team saw a 20–30% lift in AI impressions). We get into why backlinks never actually reach the model, how models only ever see a small snippet of your page, his $2,000 study mapping Gemini's brand memory (including why it quietly buries Chinese brands), the "status climber" problem for challenger brands, and some genuinely clever — maybe slightly evil — tactics. We also cover how often models really update, why search is still ~90% of the influence, agentic commerce, why Dan doesn't trust agents running loose, and how to probe LLMs properly.


Dan discusses:


- Why the measurement space is saturated — and what actually moves the needle now

- On-page optimisation & testing "model-preference signals" (and a 20–30% AI-visibility lift)

- Why backlinks don't reach the model — primary (parametric memory) vs secondary (search) bias

- Extractive summarisation: the model only ever sees a snippet of your page

- His $2,000 Gemini brand study — and why it buries Chinese brands like BYD and Tencent

- Why you have to be in BOTH the model's memory AND the search results

- What content models love (tables, FAQs) and hate (emojis will kill you)

- The "status climber" problem — and how a challenger brand can break in

- How long changes take: model update cadence vs real-time search

- Agentic commerce — watching Cloudflare and Google — and why he won't let agents run loose

- How to probe LLMs properly: 100+ probes, the API, and using Gemini as a classifier

- The three things any brand should focus on to win AI search


🎧 Listen on Spotify:

[ADD EPISODE-SPECIFIC SPOTIFY LINK]


Connect With Dan Petrovic:

https://www.linkedin.com/in/seoguy/

His tool: https://app.dejan.ai/


Thank you to our partner:


Linkavista — A backlink marketplace built for SEOs who want more transparency. With access to thousands of websites, real traffic insights through Google Search Console connections, and an easier way to find relevant opportunities.


Use code SEAN10 for 10% extra credit on your first order:

https://linkavista.com/

adbl_web_anon_alc_button_suppression_t1
No reviews yet