From today onwards, searching on Atom is a conversation. Tell it “more abstract”, “make it a word”, “something sounds quiet” and it understands. No filters for wrestling, no keywords to guess. You describe what you are building and refine in simple language until it is accurate.
This is something we have long believed in and finally developed the technology to act.
The search for the 25 years that the world has known it has one dimension. You type and other search engines look for it. Type “travel” and you get all the names with travel in it. This box does not know whether you are building a luxury Safari or a budget flight app. It’s not knowing you want something that feels like an adventure, but never say “excursion”. It can not be. It is a meaningless match.
For most websites it is very good. When you search for a flight or stapler, the term is intentional.
The naming is different. When you name a company, you almost never know the word. You know the feeling. You know how excited customers are that you are chasing after how you want people to feel when they say it out loud. The hardest part of naming is not choosing between options. It’s perfect name is a word you have never thought of. A search box that matches a string is the worst tool for that, as it can only return what you already know to request.
What we built and where it stopped
It took us years to close that gap. Atom Search has long stopped matching exact keywords. We put in a sense so the search for “fitness” shows the energy, movement and stamina, not just the names with the “fit” stamped in them. That work is real, and it makes our search smarter than string matching.
But there is still a wall and the wall is a filter.
All markets improve in the same way: checkboxes. A word. Short. Brand. Under $ 5,000. Hard little boxes that assume your taste fits a predefined container. True taste does not exist. “Make it feel more premium” is not a checkbox. “A little abstract, but still talkative” is not a drop. When your intentions differ from one type to another, the device becomes deaf.
So you did what everyone else did: modify the filter, scan the grid, uncheck it, try again, and gradually grind your taste according to what the filter has come to support. You adapt to the machine. It should be the other way around.
What’s new: You say it, listen to it, it ranks again.
Now refining your search is a conversation. You describe what you are building, say, “an expensive brand of skin care products for busy parents” and then guide it the way you would direct the name counselor sitting next to you:
“More abstract.” “Make them a word.” “Something that sounds more believable.” “A little. I do not want the word skin in it.”
It reads the reader’s resume, all the strong candidates against it, and re-ranks the whole, so that the name deserves to go up to the top. Then it tells you what has changed: the name that just popped up that has changed. Correction is visible, not magic.
And it works. Since we shipped it, attendance has increased by about a third and departures have dropped by about a third. People are in search of improvements instead of jumping, which is exactly what happens when the device meets you at your taste buds.
The part I love the most: It understands the meaning behind the request, not the literal words.
“More abstract.” Search for “fitness” the old-fashioned way and you get the obvious: FitHub, FitZone, GoFit. Literally, cramped, forgettable. Say “more abstract” and it gets what you want it to mean. You want a feeling of energy and movement without having to write it down. It stops naming by “fitting” in them and starts creating attractive names, great brands are created. It knows that “abstract” means “do not give me a dictionary, give me something new.”
“A little. Drop the word theme.” This is where the filter has nothing to offer. You can ask it for “one word” or “under six letters”, but you can not tell the checkbox to “keep calm, the nobility that is attached to skin care, just stop naming me with the skin in them.” That is the concept of subtraction, while the whole subject remains the same. Speak it in plain language, and the literal play declines as names that capture the same world without clear words rise in their place. Filters can match strings. Only conversation can keep an idea.
“Shorter” “Four letters” “Easy to say”. Length stops as a slide you fight. Say “shorter” and the name is too long to fall off even though they are a perfect topic match. Say “four letters” and it respects it. Say “easy to say” and it weighs on pronunciation. These are the preferences expressed in terms of such.
And because it’s the conversation you put them: more abstract, then a little shorter, then something more expensive. Each turn improves the last. You do not reset the filter. You are having a conversation with a name.
Why is this important?
The problem with naming is that words never form. Anyone can create a thousand strings, and most of them are worthless when you check .com or that trademark. The hard part is finding a large inventory of usable real names and checking names that fit the feeling that you can not name them. That’s the problem we set to solve, because that’s the problem between the founders and their names.
And there are quieter changes below. Solid filters do not just determine what the buyer can request. They determined what could be found. Names that do not fit the checkbox are effectively invisible, no matter how good they are. By turning search into conversation, we have opened up many more ways for names to come up with the right answer: the right tone, the right length, the right feel, the right business’s where it comes in demand. Now each name in the catalog has several more paths to the surface instead of one. The great name by which the filter is kept is finally seen.
That’s what conversation improvement does. It turns a flat string search into a multi-dimensional reading of intent (sound, structure, length, literally the business itself) and allows you to express it all the way you want to speak to a person.
Search for a spoken language: the exact words you typed. Naming speaks volumes. For the first time we talked together.
This place goes next
There is a bigger reason we created it this way. The interface for name search is becoming a conversation, and not just on Atom. The founders are starting to create an internal agency, describing what they want in plain language and expecting the right answers to come back. Finding a name that already understands the intent, the juice, and the clarity is really the kind of ability that is in that conversation.
So we open it up. The same improvements that power search on Atom are being made available to agents through our APIs, including the MCP, so that the founding assistants are already creating access to Atom’s catalog of real names that are available and displayed correctly in the flow of conversations. You will not have to leave your agency to find your name. Your agent will find it for you.
We started by discussing our own search as a person. The next step is to let it talk to other things.
Try
To Atom, Search for anything. You are working and instead of reaching for the filter, tell it what you want. Ask for more abstractions. Ask for a word. Ask for something that feels like your brand. Then look at the reorganized name to meet you.
This is live on Atom today and it is just the beginning of what conversation search can do.
Posts Introduction to AI Refinement: Intuitive Search First appeared Atoms.
