Can You be Too Early to a Gold Rush?

At a recent dinner, the head of ecommerce at a swimwear brand pulled up a dashboard on her iPhone and turned it towards me.

It was an AI search tracking tool. And her competitor had twenty percent more share-of-voice in an AI search result.

“I’m thinking about stopping everything else and going all in on AEO,” she said half-way laughing. But I could tell she was stressed about it.

Her CEO had access to the same dashboard. And she said he was already asking why they were behind.

I asked her one question.

“How much of your traffic is actually coming from AI search?”

She paused as I could see her inwardly reflecting on how her answer would sound out loud. She then answered, “less than one percent.”

That moment captures where most ecommerce teams are right now.

Every ecom leader has an executive mandate to show up in AI search. Almost nobody has a revenue number that justifies it though.

The pressure is real. The numbers are real. But the revenue is small.

Like the swimwear ecom leader, the question brands face right now is how do you make a rational investment plan in AI without overreacting to the hype or underinvesting in what might matter later?

We get the narrative already: AI will change how people discover products.

LLM search is growing at triple and quadruple digit rates. The charts are steep as some reports cite 1,200% year-over-year growth.

At the same time, most brands are still seeing AI search traffic show up more as a rounding error in their analytics.

Both of these things can be true.

And that cognitive dissonance between the two is what makes deciding what to do next hard for brands.

Early-stage channels create a frenzy of gold-rush behavior among marketers. No one wants to be the one who shows up late to an arbitrage party.

But do we really think hacky, gray hat SEO-type tricks like restructuring collection pages to match chat prompts are going to drive in massive amounts of traffic that generates sales?

You may get some sales, but then how long is that tactic going to last?

People forget that back when gray hat SEO was working, Google’s algorithm wasn’t updating every few weeks. Claude and ChatGPT roll out a new model faster than it takes for us to understand their previous models.

So what may work right now might not work in thirty days.

Instead of building a strategy that navigates volatility, what you’re actually doing is executing a strategy of volatility.

I have seen this gold-rush pattern before when I was building Amazon Alexa.

In 2016, I worked at a company called Graphiq. We built a knowledge graph that turned structured data into visual answers for publishers. Amazon acquired us to power Alexa’s question-answering system.

Like AEO, the vision was clear. Voice would become a commerce interface. People would ask Alexa what to buy. Alexa would answer and complete the purchase.

The industry consensus said voice commerce was inevitable.

Reports like Backlinko’s infamous Definitive Guide on Voice Search reported explosive adoption.

Amazon invested heavily. Entire teams were built around voice-to-purchase.

The data was not fabricated. Adoption was growing (there have been over 500 million Alexa-enabled devices sold since 2014).

But the mistake was assuming that adoption curves would translate into purchasing behavior on the same timeline.

In 2018, a report showed that only 2% of consumers used an Alexa-enabled device to make a purchase. And of that 2%, 90% refused to make a second voice-purchase. Big flop.

In 2022, Amazon reduced Alexa-related teams by thousands of roles. The capital allocation had outrun the revenue reality as Alexa is currently known as the largest failed product in technology history at over $10B.

But…AI search today is stronger than voice was.

LLMs are not a novelty interface. They are becoming the default way people ask questions.

However the voice-search lesson still applies.

Adoption curves and revenue curves don’t always move at the same speed.

There’s a proliferation of AI tools in commerce right now and it’s incredibly difficult to differentiate between signal and noise.

So where should we look to get a sense of what’s real?

Let’s start with Shopify. Their partnerships with Google on Universal Commerce Protocol (UCP) and with OpenAI on ChatGPT Shopping are two very clear signals.

Product tags. Structured metadata. Clean catalog data. Detailed product descriptions. FAQ sections that answer real customer questions in plain language. These are the inputs agents read when they generate answers.

When Shopify publishes guidance on how to structure product data for AI agents, that is closer to a sound strategy roadmap than a prompt matching hack. It is a long-term infrastructural bet that compounds across AI search and traditional SEO (no it’s not dead, yes it still matters).

From there, your investment is shaped by your product category.

Some products map naturally to LLM behavior. Supplements, cooking, tech gear/gadgets, and health-related products lend themselves to expert-style search queries.

There’s an expectation in the search result for technical specificity. Someone might ask, “What dairy-free protein powder is best for women training for a marathon?” An LLM can generate a structured answer that includes specific brands.

On the other hand, if you’re a fashion or apparel brand you have to ask yourself what percent of your ideal customers are actually searching that specifically for a pair of loafers or a knitted jacket?

It’s unlikely that if you were to win share-of-voice in AI search for those prompts that it would be enough to drive a meaningful amount of traffic or revenue.

Leveraging an onsite AI tool that helps customers choose a pair of jeans based on fit and relevant customer reviews feels like a better lever to pull.

Telling your boss you increased first-time customer conversion rate by 15% by leveraging onsite AI sounds like a big win to me.

Competitor visibility in an AEO tracking dashboard is one scoreboard.

Revenue contribution from AI search is another.

If you cannot tie the investment to a number that matters to your business, you are reacting to someone else’s scoreboard.

And look, I get it. The executive mandate to “do AI” is legitimate..

If you are the person responsible for translating this pressure into a plan, the work is not about chasing the newest tactic.

It is about deciding what number would prove the investment is working, and then building toward that with discipline.

It’s still early innings, meaning we have the opportunity to explore and test a lot of different places to invest in AI in addition to search as well.

That is the conversation I am having with brands and agencies alike right now.

If you want to start having that conversation, shoot me a DM!