Fourteen of 195 names on the shelf split into two real words cleanly; the other 181 are read by a marketplace indexer as a single unknown string. What that does to discovery, and the three changes we now make before listing a coined name.
A listing page shows one thing a buyer can act on: a price. It hides the only other number that decides an exit. Behind every card on a shelf, the name has to survive a marketplace's search engine before it survives a negotiation, and most of the work happens in a place nobody looks: the way the marketplace breaks the name into words.
I wrote about root words and listing scores in October, reading five of our own cards field by field. That post asked what a marketplace does with a coined name. This one asks the harder question on the same shelf: how many of the names actually split into two words at all, what happens to the ones that do not, and what we are changing in how we buy and write.
The honest answer is smaller than the marketing suggests, and the count is the first thing to publish.
The count, read from our own shelf
I pulled the 195 names on the live portfolio page and put them through the same test a shopper applies in the first second of reading: can the string be sliced into two dictionary words, each at least three letters, with nothing left over?
The answer for 14 of them is yes, with no argument:
- ShopUna reads as Shop + Una — and only the first half is a real word.
- GigaPay , GetLogo , AutoMation Ops and AutoMationStack split cleanly, each half carrying its own meaning.
- GlassRepair , IceShop , DevelopEnglish , AlertPeople , PaintedColors , PoPPut , GuruPeople and HomeStays are two ordinary words pressed together.
- StreamOra is the interesting one, because Ora is a word in Italian and not in English, so the split works for one market and fails for another.
Fourteen out of 195 is seven percent. That is the number I would like to be larger, and it is the number that governs how the rest of the analysis should go.
The failure mode is not what the name says — it is what the indexer reads
Take ShopUna. A person sees "shop" and a soft ending. A search engine reading a root-word field sees one token: Shopu . That is not a word in any language, it is not a category, and it competes with nothing. The listing score reads the root as a name brand rather than a category term, and a buyer searching for a shop platform never meets it.
This is the part that most sellers who buy coined names miss: the failure is silent. The name sounds good in a voice note. The seller's own pitch writes it as "Shop Una" and the brain hears two words. The marketplace's indexer writes it as a single unknown string and files it under nothing. Two readers, two different names, one card.
The same thing happens in reverse for names that look invented to a human and split perfectly for a machine. AutoMation is a common typo for automation and a terrible brand, but an indexer that splits it into Auto + Mation lands on a live root with real demand behind it. Ugly names sometimes index better than elegant ones. That is uncomfortable and it is what the field shows us.
Why coined names still make up most of the shelf
If two-word names index better, why is our shelf full of one-word inventions? Because the two goals pull in opposite directions, and both are real:
- A coined single word is defensible. It has no existing meaning to fight, no trademark crowd nearby, and it can be the whole brand.
- A two-word name is discoverable. It inherits search intent, it explains itself, and it survives being heard once.
Neither is wrong. What is wrong is pricing a coined name as if it will be discovered by search and pricing a two-word name as if it will not. When a name's split matters to discovery, the listing copy has to carry the split for the indexer. When it does not, the listing copy has to carry the meaning in plain language instead, because nothing else will.
The three things we now do with a name that does not split
The first is not to sell it as a category. A name that does not split has no category to inherit, so a listing that says "ShopUna — a name for e-commerce platforms" is selling a promise the search layer cannot keep. The copy has to sell the name as a name: its length, its sound, its spacing, who could own it. That is a different sentence and it reads differently to a buyer.
The second is to write the split into the text where a human will read it, every single time. If the split is not obvious, the description says it once, early, in a normal sentence: "Shop Una — two words, one brand." A reader who hears the name in a voice note two days later still has the pieces. This costs one line and it is the cheapest improvement available on any listing.
Third, and the one that changes the buying side: a coined name now has to earn its place with a second test, not just a sound test. Before we buy one, we ask what its root field will be read as. If the answer is a single unknown string, the name is being bought as a brand-only asset, and it has to be priced like one — which usually means no more than a band down from the two-word names next to it, because the demand it can reach is narrower.
What this costs at the price sheet level
The shelf reads 195 names, a total ask of $249,263, a median of $1,170 and a range from $120 to a single name at $5,300. By price band: 6 names at $199 or under, 49 between $200 and $499, 27 between $500 and $999, 82 between $1,000 and $1,999, 30 between $2,000 and $4,999, and 1 above $5,000.
Set the 14 clean splitters against those bands and something uncomfortable appears: they are not concentrated at the bottom. Several of the cleanest two-word names sit in the $2,000 and up range, and several single invented words sit right beside them in the same band at the same price. On the listing page, a name an indexer can categorize and a name it cannot are shown with equal weight. The buyer pays for that equality whether or not he notices it.
The fix is not to reprice the shelf today. We do not have the click data to prove what a split is worth in our own market, and pricing on a theory is how shelves get ruined. The fix is to record the split on every card going forward, so that in a year the click data can answer the question instead of my reading of it.
The parts of a name a marketplace cannot see
Three things decide whether a coined name is discovered, and only one of them is on the card:
- the root word the indexer reads, which we can control by how we write the listing;
- the category the buyer is searching, which we cannot control at all;
- the memory of the name after one hearing, which depends on whether it splits in the ear.
The second one is the reason a coinage can live for years at a fair price and get no views. Not because the price is wrong, but because nobody searching gets near the card. That is a distribution problem wearing a pricing disguise, and every seller who has ever cut a price 20% on a one-word invention has felt it.
The two-word rule we now use before listing
We do not split every name. We run one question per card before it goes out:
If a stranger saw this name once and had to type it a week later without seeing it, which pieces would he remember?
If the answer is a pair of real words, the listing says those words in the description and the title order stays natural. If the answer is one invented word, the listing stops pretending discovery will do the work and starts selling the name directly: a lander that reads the name, a sentence that defines it, and a price that reflects a brand-only asset.
That is the whole change. One question, one extra line in the copy, and a lower ceiling on what we will pay for a name the indexer cannot read.
What I would do with a shelf of fifty coinages
If I were starting again with fifty invented names and no traffic, the order would be this. Write the split for every name that has one, on the card and in the description, in the first week. For the names that do not split, build the smallest possible lander that explains the name in three sentences and captures an email — a name that cannot be found by search has to be found by someone, and a captured address is the only proof that it was. Then wait one quarter and read the numbers instead of the name.
The names that produce a view and no email are priced wrong. The names that produce no view at all are invisible, and no discount fixes invisible. That distinction is worth more than any relisting fee, and it is the one thing a marketplace score cannot tell you.
Buying a name this week?
193 exact-match, brandable and geo domains — buy-now price, escrow-protected, same-day transfer. See the portfolio · the contact page
Top comments (1)
ShopUna is the perfect example. A person hears shop and a soft ending, and a search engine hears one unknown string. Your test about typing it a week later is the one I would keep, because memory after one hearing is where a coined name either lives or dies. If you are weighing a coined name for a real product, NamerGuide at namerguide.com checks domain and username availability together, so you can see at once whether the split you want also survives as an address and a handle.