A demo request that looked like it came from the Philippines turned out to be a family-owned hat business in the United States. When I asked how they found us, the answer was one word: Gemini. Here is what that taught me about the future of product discovery.
It started the way most of my days start: with a notification.
A demo request had come in for Discount Prime. I opened our analytics, curious about who it was. The visitor location said Philippines. I made a few quiet assumptions in my head, the kind we all make without noticing, and moved on to prepare for the conversation.
Then we actually spoke. And the assumptions fell apart almost immediately.
The person was working for a family-owned business in the United States. The Philippines location was just where part of their team operated from. The company itself had a story I did not expect.
They had started with something wonderfully specific: cowboy hats.
Not hats in general. Cowboy hats. A narrow, almost stubborn niche. Over the years that niche had grown into a much broader and more interesting hat business, with a catalog full of products that clearly came from people who cared about what they made.
The size of the business was interesting. Their requirements were even more interesting.
They wanted wholesale and B2B pricing. They needed the kind of capabilities that serious merchants ask for when they are past the experimentation phase and into real operations. So I did what I always do: I sat down and mapped their requests against what Discount Prime already does.
Most of what they needed already existed inside the product. Not as something we would have to build, but as features that were already there. I still had a few follow-up questions, so we scheduled a demo.
I joined the call expecting one or two people. Instead, six people from the hat company were on the screen.
That detail mattered to me. A six-person team showing up for a demo of a Shopify pricing app is not a casual thing. These were people who had clearly done their homework and wanted to make a real decision together.
We went through their use cases. The questions were sharp. The fit was good. And near the end, I asked the question I always ask, almost out of habit:
"How did you find us?"
I expected the usual answers. A Google search. A Shopify app store listing. A recommendation from another merchant.
The answer was one word.
"Gemini."
Gemini.
Not Google. Not the app store. Not a referral. A team of six people in the United States, running a growing hat business, had described their problem to an AI system, and that system had pointed them toward us.
I sat with that for a while after the call.
For years I have thought about discovery in a very particular way. Let me describe it with the image that lives in my head.
I am sitting in a small corner of Toronto, almost like someone alone on a remote island.
Every day, I write things down on pieces of paper. Our address. What our product does. The use cases we handle. How we compare to alternatives. The answers to questions people ask us. The reasoning behind the decisions we made.
Then I roll up those pieces of paper, put them inside bottles, and throw them into the enormous ocean of the internet. And I hope. I hope that someone, somewhere, will find one, open it, understand what it says, and come to our address.
Sometimes that actually happens. A bottle gets found. A person reads it. They show up.
For a long time, the only thing reading those messages was a person, usually with a search engine in between. My job was to make sure Google could find the bottle and that a human could understand it once they opened it.
The cowboy hat call made something obvious that I had only half understood before.
The bottles are no longer read only by people.
There is a new reader in the ocean now. It does not just find the bottle. It opens dozens of them, reads them all, summarizes what they say, compares them to each other, and then makes a recommendation to a real human being who never sees the bottles at all.
That reader is the LLM. Gemini, ChatGPT, Perplexity, and the AI layers now sitting inside search itself.
So the question I used to ask has changed.
The old question was: "Can Google crawl this page?"
The better question now is: "Can an AI system understand who we are, what we do, who we help, and when to recommend us?"
Those are not the same question. A page can be perfectly crawlable and still be useless to an LLM trying to decide whether to recommend you. Crawlable is not the same as understandable. Indexed is not the same as recommendable.
There is a lot of terminology floating around right now, and most of it is more intimidating than it needs to be. Here is how I think about it in plain language.
SEO is the one we all know. Optimizing your content so traditional search engines can find it and rank it. This still matters. It is not going away.
AEO, or Answer Engine Optimization, is about making your content answer specific questions clearly. Not "here is a page about pricing" but "here is the direct answer to the question this person actually asked." Answer engines reward content that resolves a question instead of dancing around it.
GEO, or Generative Engine Optimization, is about making your content understandable and usable by generative AI systems. The goal is for an AI to be able to read your material, understand it correctly, and use it accurately when it generates an answer for someone.
LLM optimization is the broadest version of this idea. It means making your product, your documentation, and your public content easy for a language model to interpret and recommend. It is less about keywords and more about clarity, structure, and honesty.
You do not need to memorize the acronyms. You need to internalize one shift: you are no longer writing only for humans and crawlers. You are writing for systems that read everything, understand meaning, and make recommendations on your behalf.
Here is the part that took me the longest to accept.
Most software teams pour enormous effort into product UX. The onboarding flow, the dashboard, the empty states, the micro-interactions. All of that still matters.
But there is a second surface that most teams treat as an afterthought: the public knowledge surface. The collection of everything you have written about your product where the world, and now AI systems, can read it.
If your product is brilliant but your public knowledge surface is thin, vague, or inconsistent, then the AI reading your bottles has very little to work with. It cannot recommend what it cannot understand.
This realization got sharper for me recently while I was writing a comparison article about Shopify B2B and wholesale apps. I was thinking carefully about how to explain trade-offs between products, and it hit me: a comparison page is not just a marketing asset anymore. It is a machine-readable decision surface. It helps a customer decide, and it helps an AI system understand where each product fits in the stack.
That reframing changes how you write everything.
Here is the checklist I now keep in mind. It is not theoretical. It is what I am actively trying to do for our own products.
The hat company wanted wholesale and B2B pricing, and that is exactly the territory I have been spending a lot of time in lately. The more I look at it, the more I believe the quote, checkout, and wholesale buying flows are where the next interesting layer of B2B commerce is being built.
That is part of why we are thinking about things like TradeQuote AI, our way of exploring how those B2B workflows should actually feel. I am not going to turn this into a pitch. I only mention it because the cowboy hat call sits right at the intersection of two trends I care about: B2B commerce getting more sophisticated, and AI becoming the way buyers find the tools to run it.
So here I am, still in my corner of Toronto, still writing things on pieces of paper, still rolling them up and throwing them into the ocean.
That part has not changed. What has changed is who picks them up.
For years I wrote those messages for a person who might one day find them. Now I write them knowing that long before a person ever reads one, an AI system will have opened it, read it alongside a thousand others, decided whether it understood me, and chosen whether to point someone in my direction.
A cowboy hat company in the United States found a small team in Toronto because an AI system read our bottles and understood them well enough to recommend us. That is not science fiction. That happened on a Tuesday.
So I am still throwing bottles into the internet. I am just learning to write them so that both people and machines can understand what they say. Because the next six people who show up on a demo call might also answer that old question the same way.
"How did you find us?"
"Gemini."
AI & Development
Most of what we read about AI in software comes from clean demos and tidy lab examples. The real story starts when AI code meets thousands of users, messy data, and production pressure. Here is what actually happens, and where engineering still matters.
Development
To get scalable code from ChatGPT, Claude, or Gemini, you need to understand how these models actually think. Here is the math behind why they write beautiful but non-performant code, and how to prompt your way out of it.
Insights
After 4 failed startup attempts as a non-technical founder, I learned the hard way that great ideas are not enough. Now AI has changed everything. Here is what I wish I knew earlier.
