Reviewing results from Freename's early beta sessions revealed a clear pattern. Founders advanced only a small portion of the generated names, while rejected candidates showed recognizable weaknesses. "Proxelo" sounded pharmaceutical. "Gridvast" suggested infrastructure rather than a design tool. "Clarmify" joined familiar stems so plainly that it felt assembled instead of discovered.
The names that made it through those first reviews shared qualities we have since incorporated into Freename's scoring approach. This article explains those qualities, their importance, and the limits that generation tools still face.
First test: phonetic coherence
A name that is difficult to pronounce at first sight starts at a disadvantage, whatever else it offers. The issue is often less obvious than it seems. Some names look easy to say but become unclear aloud. "Claev" might be pronounced "clayv" or "clev." "Loewe" is a familiar European brand, yet its pronunciation depends on cultural knowledge. "Quellix" can prompt disagreement over its stressed syllable.
Phonetic coherence means a name suggests one clear pronunciation and feels natural in at least two major languages spoken by its intended users. A developer tool for English, German, and Spanish-speaking engineers should avoid sound combinations that feel awkward in German or cannot be spoken in Spanish. This is a quality indicator, not an absolute rule. Many successful products are pronounced imperfectly in some languages, but the factor still deserves a place in scoring.
Across European languages, useful phonetic patterns often include consonant-vowel alternation, such as CVCVC or CVVC structures, few consecutive plosives, and first-syllable stress when the intended impression is grounded rather than elevated. These are tendencies, not laws. "Spotify" does not follow the first-syllable pattern, yet its rhythm makes the name work. As an initial screen, however, phonetic coherence removes many structurally weak options.
Name length and syllables
Freename session data supports an observation linguists have made about brand names for years: 2-3 syllables are the most consistently memorable. One-syllable names may sound abrupt or be harder to trademark. Four or more syllables increase cognitive load. The ideal name can be spoken in one breath and has an unmistakable rhythm.
That does not make every two-syllable name the strongest choice. A three-syllable name with a firm internal cadence, such as "Palantir," "Asana," or "Zendesk," can be just as memorable. The goal is a pleasing rhythm, not the lowest possible count. Weak candidates often sound as though they need one more syllable or have one too many. "Plexivio" trails on longer than its intended idea requires, while "Brix" feels unfinished without another word.
Characters and syllables are related, but they measure different things. A six-character name may contain two or three syllables, depending on its construction. Our scoring gives syllables greater weight because names spread socially through speech. The written version affects domain availability and readability at small sizes, while the spoken version influences the word-of-mouth growth that carries a name.
Distinctiveness versus competition
Distinctiveness depends on distance from the closest competitor in the same category. "Notion" succeeds not because it is an invented word, but because no other major productivity tool uses it. Its users have effectively assigned the common English word to one product. If three other companies used "Notion," that advantage would vanish.
In generating and scoring candidates, Freename compares each proposed name with existing trademarks and known products in the founder's stated category. This is a practical screening step as well as a legal safeguard. A name can perform well on sound, length, and memorability, but an existing association in the same market makes it unsuitable.
Crowded naming categories make distinctiveness especially difficult. Developer tools, productivity products, and B2B SaaS already contain many names ending in "-ly," "-ify," "-hub," or "-base." New candidates using those formulas begin with less separation. The patterns can still produce good names, but they demand more differentiation than a name that breaks away from them.
The meaning layer
A name may sound clean, have good proportions, and stand apart, yet still feel dull. The meaning layer helps distinguish names people find intriguing from those they merely accept. It is also the quality current generation systems have the greatest difficulty producing consistently.
Meaning does not require a name to describe its product. "Stripe" says nothing literal about payment infrastructure, and "Loom" does not explain async video messaging. Both carry a secondary association that becomes richer after the product is known. Stripe suggests the precision of a clean line. Loom suggests communication threads being woven together. The link is loose enough to evoke, yet specific enough to feel deliberate.
Generation systems create this quality from time to time, but not dependably. Names founders call a "click" in Freename sessions usually contain that secondary resonance. Names described as "fine but not exciting" generally do not. We can assess explicit meaning, including whether a name evokes the right category or creates unwanted associations. We cannot consistently measure the elusive feeling that a name was somehow waiting for this product.
Where generation tends to fail
Generation failures tend to follow structural patterns rather than chance. The most common is the portmanteau trap, where two descriptive ideas are joined correctly but produce an obviously constructed name. "Trackvault" for analytics, "FormFlow" for workflow software, and "DataNest" for a database all contain relevant parts. None contributes meaning beyond those parts, so the whole does not rise above the components.
Another problem is suffix clustering. A system may produce several candidates with the same ending because its training data contains many successful names built that way. A session yielding "Verido," "Flexido," and "Proxido" illustrates the issue. Each might be acceptable alone, but together they show that the generator is searching within a narrow pattern range.
These patterns do not rule out usable names. "Typeform" combines two descriptive elements and has become a successful product name. The key is that when the same patterns produce weak candidates, the weakness is usually predictable. Recognizing those tendencies helps you assign the right weight during evaluation.
Applying the criteria to a shortlist
When Freename, or another generation process, gives you candidates, applying these criteria creates a structured path from a list of 20 to a shortlist of 4-5 for testing with real people.
Rate every name for one obvious pronunciation, a 2-3 syllable structure, no close competitor in its category, and secondary resonance beyond its literal parts. Use yes/no for pronunciation and competition, yes/partial/no for syllable structure, and yes/no/uncertain for resonance. Candidates with four yes ratings deserve serious attention. Those with three can remain on the shortlist. Names with fewer than three are rarely worth defending, even when the domain is available.
This is an initial screen, not a final verdict. A name may miss two criteria and still suit a particular product and market. These measures describe qualities associated with durability, not a formula that guarantees success. The most dependable final test remains the founder saying the name aloud 50 times over two days and noticing the reaction.