#136 The Next Arbitrage Window Is on ChatGPT | Mike Mallazzo & Tye DeGrange

Let's get into the world of AI.

You're in this intersection of AI, media, commerce.

What is something that's changed your mental model around...

You referenced at the beginning, a customer acquisition.

Yeah.

In particular, over the last 12 months, give or take.

Yeah, so I think the most interesting question in our profession right now that really nobody has an answer to, and I'm trying to kind of figure out my own worldview day by day, is really around what marketing to agents will look like and what does our profession look like as agents start to sit between a marketer and the end customer.

You know, I would say 12 months ago, I was very skeptical that the idea of marketing to agents would even exist.

At Mike McNerney's Mark event last summer, I interviewed the CEO of Profound, James Cotterwalder, and he was kind of espousing this world where the job of a marketer is really to persuade an algorithm.

And I didn't quite call bullshit on it on stage, but I certainly had a very palpably skeptical view to the audience.

And while I don't agree with James on everything, I think he was directionally right about this idea that a real part of the job now is persuading an LLM as much as it is persuading an end person.

What I would just kind of say on that, though, is, you know, I think there's this sort of fallacy in the marketing ecosystem that marketing to agents or marketing to LLMs is going to be this sort of very rational endeavor where... you just create really clean markdown files and really clean catalogs and you pursue whatever the latest aeo arbitrage hack is on reddit and you'll be able to kind of convince the models i think that's wrong for a number of reasons i think you know with each model release and this obviously was in the news this week with dorkesh's post on ai developing civilizations and all that but With each model release, the sort of LLMs are becoming a little bit more anthropomorphic, a little bit more human-like.

They're learning to kind of look for heuristic shortcuts that don't use as much compute.

And kind of what that means is that AI and LLMs are going to have brand preference the same way that people do.

I'm actually writing a blog post right now that's called How to Sell Lucky Strike to an LLM, kind of drawing on the Madman illusion.

But I think that's the really interesting question is, how do you make an LLM or an agent develop brand preference? I think that's the highest order thing that everybody's trying to accomplish with AEO and GEO and a lot of the sort of affiliate tactics.

But I haven't seen a lot of marketers kind of zoom out and think about it that way.

Yeah.

I had a lot of things going through my head, but maybe for people to understand how much of that is happening now, just to get people up to speed, in your opinion, and based on what you're seeing currently.

Well, I think it's two parts, right? I think, obviously, consumer usage of LLMs goes up every day.

Google, obviously, is effectively becoming its own zero-click interface, right? So the idea of the entire web being kind of contained with an answer engine is very, very real for a lot of people today.

Yep, yep.

What is obviously more nascent is any real sense of agentic commerce, where an agent is autonomously purchasing something end-to-end.

But an agent already mediating a lot of what you see between a prompt and between when you actually reach a human set of eyeballs, that is already very prominent.

If you think about the percent of Google searches that are strictly zero-click now, plus the kind of aggregate usage of chat GPT cloud perplexity, most, I would say, brand touchpoints at this point are touching some sort of an AI system before they are getting to publisher or brand content.