What does a searcher actually want when they type one word? You’d think SEOs had that one figured out by now. Then Vinod Kumar gave me the cleanest example of search intent I’ve heard in a while, and it came from a clothing store’s search box, not from Google.
Vinod is the co-founder and CEO of Syntheum AI, an agentic merchandising startup that works with large brands and retailers on e-commerce. We talked on the Unscripted SEO Podcast about on-site search, agents and AI shopping. You can watch the full episode here or listen on Castos.
One letter separates two different shoppers
Here’s his example. You’re on an apparel site and you type “khakis.” You want pants, and if those pants come in five colors you want to see all five. Now type “khaki” without the s. Khaki is a color, a kind of brown, so you want the brown shirt, the olive pants, maybe a camo hat.
“Now imagine khaki and khakis, one character difference, but two very distinct results.”
Somebody has to teach the site that difference, and the product records might not even contain the word khaki. They might say tan, brown or earthy. That’s the merchandiser’s job. And as Vinod pointed out, shoppers who use the search box are the ones most ready to buy, so getting it wrong costs real money.
Google has been doing this to us for years
This is the part that clicked for me. Disambiguation is exactly what Google does to figure out intent, and it’s what we try to influence from the outside, without direct control. Vinod is doing the same work from the inside, with the controls in hand.
I’ve had this conversation with clients more times than I can count. I had an Atlanta law office whose homepage said “law office” and never said which kind of legal services they provide. To them it was obvious. To a search engine it’s ambiguous, and an ambiguous page doesn’t get matched to anybody’s query. You never made the semantic connection, so the engine can’t make it for you. If you’ve ever ranked for the wrong half of your audience, it’s usually the same mistake.
“Not very different from how you optimize your search results for organic search. Similar principles and concepts apply to optimizing results for on-site search.”
Science underneath, art on top
I asked Vinod whether the vector and relevance work my friend Michael McDougald geeks out about shows up in his mapping. It does. He split merchandising into two halves.
- The science. Vectorization and semantic relationships between words, keywords, descriptions and products. That’s math: matrix multiplication, linear regression and the like.
- The art. What the brand stands for. Maybe it’s sustainability, maybe it’s choice, maybe it’s color. A merchandiser might just have a hunch that one color should show first.
His coffee mug example makes it concrete. The brand team won’t call the mug blue, because blue is boring, so they call it ultramarine. Shoppers don’t search for ultramarine. So an agent maps ultramarine to every blue a shopper might type, and the merchandiser drags the colors into the order the brand wants. The math runs underneath and the brand preferences sit on top.
That’s a good model for your own site, too. Let the tools handle the mapping. Spend your own time on the parts only you know, like which product you actually want people to see first.
We used to do this with end caps
I told Vinod our generation is a little spoiled. Thirty years ago these conversations happened in department stores: light blue jeans on the second row, the good stuff on the end caps. Now the digital shelf is so big no human team can keep it stocked.
He agreed, and pointed out it’s about to get stranger. Shoppers now research and compare inside ChatGPT, Gemini and Perplexity, and a chat answer isn’t a shelf anyone can arrange. Vinod’s word for it was probabilistic. I wrote about what that means for agencies over on SEO Arcade.
Check what the machines think you sell
If you only do one thing after reading this, read your own homepage like a search engine would. Does it say what you sell, for whom, in plain words? Then run it through AVA, the free AI Visibility Auditor Vinod’s team built, and see whether ChatGPT and Gemini tell the same story you do. If they don’t, you’ve got a khaki problem.