Nobody Cares About Your Search
You want to be the store that shows the customer exactly what they’re looking for. Faster than the competition. Better than the competition. You want them to come, find, buy — and come back, because it was easy.
So you invest in how customers search on your site. And I don’t just mean the search bar at the top. I mean the entire journey — how they click through categories, filter by parameters, sort results by price, scroll through dozens of products before finding the right one. All of this is search. And you’re fine-tuning it. Optimizing listings, measuring conversion on every query, handling synonyms and autocomplete. Paying for indexing tools so results load in milliseconds.
And it makes sense. Whoever controls search controls what the customer sees. And whoever controls what the customer sees controls the purchase.
Except the customer opens their AI assistant and says: “I need a winter jacket.”
Nothing more. No filters. No size. No budget.
The model already knows. It knows they live in Brno and walk to work. It knows they’re slim, wear a medium, but have a long torso. It knows they bought a Patagonia last year and loved it — but it was too short. It knows they prefer minimalist Scandinavian style. It knows how much they’re willing to spend, because they discussed their family budget together.
“I’ll just put ChatGPT on my site and get the same thing.” No. You won’t. Because the customer shared this information with their assistant — not your website. You have their purchase history with you. The model has their life. Where they live, how they think about money, what they dealt with last week. The customer will never give you this. Not because they don’t want to — but because they have no reason to. They shop with you. They live with their assistant.
What you’re paying for
The customer types “winter jacket” — and you don’t know if they want down or synthetic. You don’t know if it’s for the city or the mountains. You don’t know their body type, what style they prefer, or how much they want to spend. So you show them four hundred jackets and hope they’ll click their way to the right one.
A license for a site search tool. Someone tuning the rules — what shows up for which query, in what order. An analyst measuring what customers click on and where they leave. A developer integrating and maintaining it all.
For an e-commerce business turning over tens of millions, that’s hundreds of thousands a year. Not on one invoice — spread across licenses, salaries, developer time. That’s why nobody adds it up. But the money is there.
And all of it solves one problem: the customer didn’t tell you what they actually need. Your entire search stack exists to compensate for one thing — that you know nothing about the customer. It’s a sophisticated guessing system.
And you’re paying for the guessing.
Two types of search — and only one of them is yours
A customer who knows what they want will still use your search. They type “North Face McMurdo,” find it, visually confirm, select color and size, buy, leave. Fast, no thinking. Nobody’s taking this from you — and here, your search works today and three years from now.
But this isn’t where decisions are made. This customer decided long before — somewhere else.
On Amazon, customers who use search convert 6x higher than those who don’t. On Walmart, 2.4x. People want to search and find the right product. The question is — where will they do it.
The interesting — and expensive — part is the second moment. The customer doesn’t know exactly what they want. They know they need “a winter jacket.” But which one? Down or synthetic? For the city or the mountains? What cut for their body? This is where you spend the most today — filters, sorting, recommendations, rules for what appears and in what order. The whole machinery exists to guide the customer from “I don’t know” to “I’m buying.”
And this is exactly the part LLMs are taking over. Not because they’re faster. Not because they have a nicer interface. But because the customer already had that conversation with their assistant. They told it they walk to work and hate bulky jackets. That they have a long torso and last time the Patagonia ended above the waist. That they want something clean, Scandinavian, under five thousand. They got three recommendations with explanations — one for daily commuting, one for a ski trip in March, one as a compromise of both. They decided. And they come to your site with a clear intent — find the specific model, click, buy.
But this goes against everything you’ve learned in e-commerce: the customer doesn’t mind waiting ten seconds for that answer. You’re paying hundreds of thousands for your search to return results in fifty milliseconds. And the customer will happily wait ten seconds for the assistant’s answer. Because it’s an answer — not a list. Speed of response matters less than quality of response. And that’s the end of the game you’ve been playing for the past ten years.
The numbers confirm it. Gartner predicts that by 2026, traditional search engine volume will drop 25%. By 2028, many brands will lose 50% or more of their organic traffic. Adobe Analytics shows the other side — traffic to U.S. retail sites from generative AI sources grew 1,200% in eight months. These visitors browse 12% more pages and have a 23% lower bounce rate. They arrive prepared. They decided elsewhere.
The valuable search — where decisions happen — will disappear from your site. Not overnight. But gradually and irreversibly.
Today, only a fraction of customers use AI assistants for shopping. But a third of Gen Z and a quarter of Millennials already prefer AI platforms for product selection over traditional search. 46% of them use AI platforms daily. In three years, this will be the norm — just like nobody today thinks twice about whether to Google something. Privacy? People traded it for convenience long ago — they share location, browsing history, biometric data. They’ll do the same for personalized advice from an assistant. And hallucinations? Models are improving at a geometric rate. What was a problem this year won’t be one next year.
What to do — specifically
First: simplify your search. Your search doesn’t need to be smart. It needs to be fast and accurate. The customer types a product name, gets the product. It handles typos, partial names, finds North Face even when they type “nort face mcmurdo black.” That’s all you’ll need it to do.
But that’s a commodity every basic search tool handles today. You don’t need to pay hundreds of thousands for it. According to Baymard Institute, 41% of e-commerce sites fail to properly support even basic search query types. Invest in reliable product lookup — not in sophisticated recommendations that someone else will soon do for you.
Second: shift budget to product data — but think in decisions, not filters. As I wrote in a previous article, data granularity determines whether a model recommends you. Here I’ll go more specific. Your filter says “material: down / synthetic.” An LLM needs to know “down, 800 fill power, extended cut for tall frames, waterproof, minimalist Scandinavian design, rated to -15°C.” This granularity is what matters now. Go through your catalog and ask for each product: if this were the only information a model had — would it be enough for a recommendation?
Third: make your data accessible to models. It’s not enough to have good data in an internal search index. If an LLM can’t access it, it doesn’t exist. Schema.org markup on every product page. Structured product feeds. Publicly available specifications that a model can read and interpret. Today you decide what the customer sees — you push products up, hide ones you don’t want to sell. Tomorrow the model decides. And it doesn’t care what you wanted to show first. It cares about one thing only: which product data best matches what the customer needs. The quality and accessibility of that data is your new lever. And it’s the only one you’ll have left.
Fourth: measure where decisions come from. Start tracking what share of your traffic arrives with clear purchase intent — direct PDP visits, branded queries, exact product names — versus discovery traffic through categories and filters. When you see discovery declining and direct rising, you’ll know the thesis of this article is playing out. And you’ll be prepared — not surprised.
Conclusion
Search on your site won’t disappear. But the part you’re paying hundreds of thousands for? Nobody will care about it.
You’ll keep product lookup — customer types the name, gets the result. Fast, simple, commodity. The other part — the advising, selecting, deciding — will be handled by the AI assistant. And it’ll do it better, because it knows the customer better.
You can fight it. Invest more in smarter search, more sophisticated filters, better recommendations. Or you can move that money to where it’ll determine whether the assistant recommends you at all — into your data.
Try it today. Open ChatGPT and ask it about a product you sell. Does it recommend you?