How to show up in AI search.

No one knows how LLM search works. Here’s probably how it works.

“I said I want to touch the earth. I want to break it in my hands. I want to grow something wild and unruly.” – Natalie Maines

Your boss bounds into your office, radiating big golden retriever energy. This is all hypothetical, by the way. You might not have an office. Your boss might not radiate golden retriever energy. You might be the boss. It’s rude of your hypothetical boss not to knock, though, right?

You’re busy reading the SRH Empirical Marketing Dispatch, and your hypothetical boss just waltzes in like he owns the place. Sure, maybe he actually owns the place, but that’s rude. Also, why is your boss a man? Never mind. He’s not real. Don’t read too much into it. Let’s call your entirely fictional boss Gary.

Gary says, “Hey. I’ve been, uh, thinking.” Gary thinks a lot, and usually he’s really good at it, but sometimes he isn’t. And when he isn’t, he usually leads with, “Hey. I’ve been, uh, thinking.”

The “uh” is his tell. The “uh” sends a shiver down your spine into your soul, where it rattles around like a handful of loose change in the dryer. You smile pleasantly.

“So, you’ve heard of, like, ChatGPT, right?” he asks. Of course you have. Everyone has. You nod politely.

“Did you know people use that thing for search?” You do know that. You know that somewhere around 5.6% of people in the US use an LLM for search. You know that AI overviews appear in around 50% of Google searches — sometimes less, sometimes more. You also know that less than 1% of total website traffic comes from AI referrals.

(You also know that 70% of Gen Zers are using TikTok for search, but that’s another Dispatch for another day.)

And here’s something else you know: According to a recent McKinsey report, around 55% of consumers (it depends on the category) are now using AI search to make buying decisions. A brand’s website, however, is not the only thing LLMs use to generate an answer. And the robot probably won’t give your website as much authoritative weight as you might expect. More on that in a moment.

“ChatGPT is the new Google!” says Gary. “My nephew in business school told me all about it. It’s really exciting. Google is dead! ChatGPT is the future!”

It’s nice to see Gary this excited about something, but you already know that the other shoe is about to drop. You can see it hovering like Wile E. Coyote right after he runs off a cliff and just before he remembers gravity is a thing.

“I want our brand to show up in AI Search. So if you could just, uh, hack the algorithm or whatever, that’d be great.”

And then he leaves. The other shoe stops hovering and lands with a resounding thud. Gary’s not wrong, right? It’s always good to be part of the consideration set, even if it’s robots doing the considering. Except it’s not robots.

So how do you “hack the algorithm or whatever?” Focus on the “or whatever.”

You might already know this, but just in case you don’t, here’s some stuff about the robots:

AI means a lot of things. We’re talking specifically about Large Language Models (LLMs) like ChatGPT, Claude, Gemini, etc.
LLMs, very generally speaking, are probability engines. They are calculating the statistical probability of an answer. When you ask it a question, the robot looks at all the information it has been trained on and everything it can search for and asks, “Based on all of this, what is the most likely acceptable answer?”
In other words, they’re not looking for the truth. LLMs only reflect what people are already saying.
Share of search is critical because the human mind is still the most important search engine.
Every model is different, and no one really knows how they work.

Unlike search engines, LLMs are not looking for keywords. They’re looking for signals. The internet, unsurprisingly, contains a lot of noise.

Let’s say you’re searching for new golf clubs. You’re probably not, but let’s say you are. If you type “best golf clubs to help with my slice” into Google, you’ll get a lot of results. Some results will be golf club brands. Some will be coaching programs. Some will be videos about how to adjust your swing. You have to sort through them.

If you ask ChatGPT, “I’m working on my swing, but are there any clubs out there that will help me fix my slice?” ChatGPT will search the internet to see if anyone else is talking about this particular problem.

It will look at Google Search to see which brands other people are searching for by name. Depending on where it’s allowed to go, the LLM will also scour blogs, golf gear review websites, Reddit, YouTube transcripts, golf discussion boards and golf podcast transcripts and any other sources it can find to see if there’s a brand of golf club out there that will help you solve your slice problem.

The LLM will also look at golf club brand websites, but it will give those results less weight. The robot is looking at a brand’s website for facts, but it’s looking to the rest of the world for proof. Though on-site social proof can also help.

LLMs seem to be looking for conversations and consensus, for clear signals that agree with each other.

So if a brand of golf clubs gets mentioned in the same sentence as “fix my slice” on Reddit, that’s great. If the brand is also mentioned with “fix my slice” in a podcast transcript, a YouTube transcript and a golf gear review website, the LLM’s confidence score for the brand increases dramatically.

The LLM decides on the most statistically safe answer and then gives you the statistically safest sources. Maybe it actually found the answer on Quora or Reddit, but it’s sending you a NY Times article because it has been trained not to look like an idiot.

To be fair, if you told your friend to buy a certain brand of golf clubs because some Reddit user named KingBananaDork1173 swears by them, well … you wouldn’t actually do that.

Is an LLM giving you the right answer from the best source? Maybe. How do the robots decide which brands to choose? Kinda the same way people do. Fame.

If you want your brand to be noticed by the robots, you’ve got to get noticed and remembered by as many people as possible.

Andrew Ehrenberg’s Double Jeopardy Law applies to LLM search as much as it does everywhere else. Share of market correlates with share of search, which correlates with share of voice. Big brands have more customers and those customers are slightly more loyal. Big brands are more famous and they get searched for, talked about and blogged about more. Their products get more reviews. For LLMs, this means the big brands get more mentions, and those mentions are more authoritative.

Hang on. Consensus and conversation … that sounds like what social media influencers do, no? No. Well, yes and no. Influencers are often great, but social media is a bit of a walled garden when it comes to LLMs, though the walls have more cracks in them than they used to.

For example, Meta has its own LLM, so it doesn’t want Gemini or ChatGPT poking around Instagram. So if hundreds or thousands of people are raving about your brand on Instagram, ChatGPT won’t know about it unless the conversation spills over the walls into the other parts of the internet that it can see. TikTok is way more transparent, though not entirely.

Generative Engine Optimization (GEO) is not about throwing elbows over keywords. It’s about being found and verified. Great website content matters because all the LLMs seem to be optimizing for certainty, but being talked about everywhere else matters more.

If you want to get noticed by the robots, you’ve got to be noticed by as many humans as possible. Online and in real life. Your brand needs to be front and center. Your distinctive brand assets need to be on point. Your creative must be big, bold and relentlessly consistent. The same with your media plan. In order to be noticed, you need to get people talking, blogging, writing, reviewing and otherwise posting about you.

You need to grow something wild and unruly. That’s how you “hack the algorithm or whatever.” If you would like some help with that, give us a call. And tell Gary we said, “Hi.” We’ll see you next time.

Sources!

“New front door to the internet: Winning in the age of AI search” – McKinsey
“AI Demystified: Introduction to large language models” – Stanford University
“ Large Language Models as Optimizers” – Chengrun Yang* Xuezhi Wang Yifeng Lu Hanxiao Liu Quoc V. Le Denny Zhou Xinyun Chen
“GEO: Generative Engine Optimization” – Pranjal Aggarwal, Vishvak Murahari, Tanmay Rajpurohit, Ashwin Kalyan, Karthik Narasimhan, Ameet Deshpande