The search box assumes you already know what you want

Type a word, get a list. It works, as long as you can supply the word.

Often you can. A 60mm lens. A replacement filter. A particular book. The name is the whole query and search does the rest.

When the word is missing

Someone looking for a gift is not thinking "ceramic mug, white, under thirty dollars." They are thinking of something calm and grounding for morning coffee. "I want to spend Sunday working on my car" is the same shape. So is "something cool for my desk setup." None of those is a category and none is a keyword. Each one is a picture of a day, and what belongs in it follows from the picture.

Keyword search does not fail here because gifts lack keywords. It fails when what you want is expressed in terms the catalog never recorded.

Our bet is that people start with the feeling and narrow down afterwards, using size and price and color. Offline they already do. Walk down a row of shops and nobody starts from a category. You see something, you stop, you pick it up, you put it back. What you carry home is often not what you came out for. Online asks for it the other way around: pick a category, type a keyword, apply filters, and hope that what arrives at the end feels right.

Better ranking does not help here. Rank the results for "mug" perfectly and you have perfect results for a guess. The gap is at the input, before ranking gets a turn.

What about asking an AI assistant

Ask an AI assistant for something calm and grounding for morning coffee and it will answer, often well. Assistants and catalogs are not competing here. An assistant is a way of asking, and behind any way of asking sits a catalog, and an answer can only be as good as what that catalog recorded.

Most catalogs record objects: a title, a category, a set of specifications, a price. That is enough to answer "ceramic mug, white, under thirty dollars." It runs out at "something for a Sunday under the car," because no field anywhere says which day a product belongs to. A better question does not fix a catalog that was never asked to hold the answer.

Brands can write those words themselves, and some do it well. The difficulty is that each brand picks its own. One writes calm, the next writes quiet, a third writes understated, and a shopper has no way to know which shop chose which word. A vocabulary only works as a way in when it is shared.

What Okidokes does instead

Okidokes is a product discovery engine that describes every product two ways, how it feels and the moments it suits, drawn from a fixed vocabulary and applied by one pipeline across the whole catalog. Searching runs against that description rather than against the product title.

A search for a mug returns mugs, which is right. A Sunday under the car is a jack, a light you can clip somewhere, a hand cleaner that actually works, and something to listen to. A care package for a friend who is ill is tea, a blanket, something to read, and a soup you can manage one-handed. No category holds either group together. Describing products by feel and by moment lets an answer cross categories.

We are building this now, at okidokes.com. It is a preview and it is small. Nothing on it is for sale: products read "available to shop at launch," and links out to the brands start at beta. The preview is there to test whether the finding works, and that is all it is for. If you try it, the most useful reply is one thing that felt right and one thing you would change.