Study Shows LLM Conversion Rate Is 9x Better — AEO Is Coming

Forbes

5 hours ago

Bing, OpenAI, Microsoft and Google logos displayed on a phone screen and a laptop keyboard are seen ... More in this multiple exposure illustration photo taken in Krakow, Poland on February 8, 2023. (Photo by Jakub Porzycki/NurPhoto via Getty Images)
Some predict that by 2028, more people will discover products and information through large language models (LLMs) like ChatGPT and Gemini than through traditional search engines. But based on research I conducted with Cornell Master's students, that shift is happening much faster. LLM-driven traffic is already starting to outperform traditional search — not in volume, but in value.
Traffic from LLMs converts at nearly 9x higher rates than traditional search. This is the biggest disruption to search since the dawn of the internet. If you're a brand or publisher, now is the time to adapt your SEO playbook. Oh, there is no 'S' — it's now called Answer Engine Optimization (AEO)
Back in January, I predicted that traditional search was on its way out. Just six months later, the shift is already visible. In my UX research, I classify shoppers into three categories:
It's easy to see how all these needs can now be met through a conversation with LLMs like ChatGPT, Claude, Gemini, or Perplexity.
Say you're looking for an isotonic drink powder. Instead of scanning blogs, watching videos, or scrolling endlessly, you now ask ChatGPT — and it responds with direct recommendations:
Ask about ketogenic-friendly options, and it will go even further — offering details on ingredients, comparisons, and alternatives.
Staff Sergeant Alex Mackinnon from the Royal Electrical and Mechanical Engineers holds a sachet of ... More isotonic drink, Tuesday September 20, 2005, at Bramley Training Area near Basingstoke, where the Army announced it will be including the sports drink in its ration packs. The powdered drink will be incorporated in 24-hour ration packs after the its producer, GlaxoSmithKline, won the three-year contract in a tendering process. See PA Story DEFENCE Drink. PRESS ASSOCIATION Photo. Photo credit should read: Chris Ison/PA (Photo by Chris Ison - PA Images/PA Images via Getty Images)
This isn't search — it's advice. And when users follow those links or act on suggestions, they convert at dramatically higher rates compared to normal search traffic. In my studies, LLM-generated traffic behaves more like a personal recommendation than a keyword query. But here's the catch: if your brand isn't listed, you're invisible. The customer won't even consider you.
Good numbers are hard to come by. LLM traffic, like what comes from ChatGPT, doesn't always leave a clean trail — users might just copy and paste a product name and head to Amazon or another site. To get better data, we created a ChatGPT-style experience inside the site search of several e-commerce stores.
In A/B tests, we compared regular keyword search with an AI-guided, conversational search experience. The difference was stunning: almost 9x higher conversion. Yes, nine times.
But it's not just conversion that's changing — the way people search is evolving, too. In the past, users typed one or two words like 'camera.' Now, when they're shown more natural and detailed responses, they respond in kind. We're seeing queries like:
'What's a compact camera for wildlife photography that fits in a carry-on?'
Semrush backs this up with broader data:
In our interviews, shoppers said they felt more 'understood' and 'better about their purchase.' It didn't feel like a search engine. It felt like getting advice from a knowledgeable friend.
If you scale that behavior to external LLM traffic — not just on-site — the value of that traffic already rivals what you get from SEO.
For brands, this means it's time to rethink how you show up in these conversations. That's what AEO — Answer Engine Optimization — is all about.
Brands need to act. If you're not being cited by LLMs, you're becoming increasingly invisible. To get picked up by an LLM, you need to understand how these models learn from content.
Masking in ML Training
LLMs are pattern-completion engines. I often use the example of 'Life is like a box of ___' in my online certificate from Cornell.
Correct. The answer is Chocolate. Machines learn the right answer through trial and error. This approach is called masking. To show up in an LLM's response, your content needs to become part of its masked training data.
LLMs look for authoritative, helpful, and authentic content. Since they predict the next word in a conversation with a user, they favor content written in a conversational or Q&A format.
For brands a new playbook is emerging AEO. I outlined all what brands need to know. AEO is just the beginning. Two even bigger shifts are on the horizon — and both will deeply impact how brands show up in the age of AI: Paid Ads in LLMs and Model Context Protocol and agents that act on behalf of the LLM. The future is already underway. Ping me on LinkedIn if you want to continue the conversation.