Namethistle

How our tools work

If a tool gives you a number, you should be able to see how it got there. Here is the whole method.

Last updated September 9, 2026

Most tools that score things do not say how. We would rather publish the method, partly because it is the honest thing to do and partly because people find our mistakes that way.

Step one: breaking a name into syllables

Everything starts here. A name is lowercased and stripped to letters, then we locate its vowel groups. Each group of adjacent vowels forms the nucleus of one syllable.

The consonants between two nuclei then have to be divided between the first syllable's ending and the second syllable's beginning. We use the Maximal Onset Principle, the standard rule in phonology: give the following syllable the longest opening it can legally take, and leave whatever remains as the previous syllable's ending. So jennifer divides as jen-ni-fer, not jenn-if-er.

One English-specific adjustment: a final silent e preceded by a single consonant is absorbed into the previous syllable rather than treated as its own. Grace is one syllable, not two. But marie keeps both, because the vowels sit adjacent.

Syllable structure of the name Jennifer Jennifer divides into three syllables: jen, ni and fer. Each syllable has an optional opening consonant group, a vowel at its core, and an optional closing consonant group. Jen is j plus e plus n. Ni is n plus i with no ending. Fer is f plus e plus r. j e n n i f e r syllable 1 syllable 2 syllable 3 opening consonants vowel core closing consonants Consonants between two vowels go to the syllable that follows, as far as English allows. So it is jen-ni-fer, never jenn-if-er.
The same split the engine performs on every name you enter.

Step two: recombining

Six methods run over the syllable structures. Syllable splice takes whole syllables from each name in turn. Sound overlap finds the longest shared run of letters between the end of one name and the start of another and fuses them there. Alternating interleaves syllables. Vowel swap keeps one name's consonant frame and substitutes the other's vowels in order. Head and tail cuts at syllable boundaries. Nested inserts one name's core syllable inside another.

Each method runs in both directions, and with three or more names it also runs as a fold, blending the first two, then blending that result with the third, and so on.

Step three: scoring pronounceability

Every candidate starts at 100 and loses points for specific, named problems.

  • No vowel at all: −70. The name cannot be said.
  • Vowel ratio below 22 per cent: −22, marked "consonant-heavy".
  • Vowel ratio above 68 per cent: −14, marked "vowel-heavy".
  • Illegal consonant cluster of three or more letters: −16 each.
  • Illegal two-letter cluster: −5 each.
  • Awkward letter pair such as a doubled vowel or a clash like qz: −9 each.
  • Under three letters: −25. Over twelve: −14.

A cluster counts as legal if it appears in our list of consonant combinations English permits at the start of a syllable, or the separate list of those it permits at the end. The onset list includes items like bl, chr, scr, spl, thr. The ending list includes ck, ng, nts, rst and similar. Both lists are visible in the source of the tool.

Separately we measure rhythm: the proportion of adjacent letter pairs where one is a vowel and one is a consonant. Names that alternate evenly read fastest. The final sayability figure is 72 per cent of the penalty score plus 28 per cent of the rhythm figure.

Step four: scoring the blend

A name can be perfectly pronounceable and still be a bad blend, so this is measured separately. For each name you entered, we compute the longest run of characters it shares with the result. That tells us how much each source actually contributed.

  • Two or more source names appear whole and together account for the result's length: −55, marked as joined rather than blended. These are hidden by default.
  • One source name survives almost intact (within two characters): −30.
  • A source contributes fewer than two characters: −22 for each such name.
  • Imbalance between the largest and smallest contribution costs up to 18 points.

Step five: ranking

The number shown on each card is 55 per cent sayability and 45 per cent blend quality. Sayability is weighted higher because an unpronounceable name is useless regardless of how elegantly it merges its sources. Ties break toward the shorter name.

How the score on each result is calculated The final score is 55 per cent sayability and 45 per cent blend quality. Sayability is itself 72 per cent cluster and vowel penalties and 28 per cent rhythm. Blend quality penalises plain concatenation, a source name left almost intact, a source contributing fewer than two letters, and imbalance between sources. Sayability · 55% Blend quality · 45% Can a person say it? Illegal consonant cluster −16 Too few or too many vowels −22 Awkward letter pair −9 Very short or very long −25 Then blended with a rhythm measure: how evenly vowels and consonants alternate. Is it really a blend? Names simply joined −55 One name left intact −30 A name barely present −22 Uneven contribution −18 Measured from the longest run of letters each name shares with the result. Above 85 reads naturally. Between 60 and 85 usually needs a small tweak. Below 50 has a named problem, and the card tells you which.

Known limitations

We would rather state these than have you discover them.

  • The rules are English rules. Names from languages with different phonotactics — Polish, Vietnamese, Xhosa, Georgian — will score lower than they deserve, because the clusters they use legitimately are not in our English lists. We are working on per-language rule sets.
  • Spelling, not sound. We work from letters, not phonemes. English spelling is irregular enough that this occasionally misjudges a syllable boundary, particularly with names of Irish and Welsh origin.
  • No semantic check. The tool has no idea what a result means. It cannot tell you that your blend is an existing word, a brand, or something unfortunate in another language. Always search a shortlisted name before using it.
  • Deterministic. The same inputs always produce the same output. This is intentional, so a shared link reproduces exactly what you saw, but it means results are not unique to you.

Where our data comes from

Several tools are backed by published statistics rather than by lists we wrote ourselves. That is what lets them weight results by real frequency instead of treating every name as equally likely. All of it is processed into compact files when the site is built, and nothing you type is ever sent to any of these sources — or anywhere else.

  • US given names, 1880–2025. The Social Security Administration’s record of names on card applications, covering roughly 375 million births. Public domain. Names given to fewer than five babies in a year are excluded at source. Used by the Name Popularity Explorer and the Random Name Generator.
  • US surnames, 2020 Census. Frequently occurring surnames from the 2020 Census, published by the US Census Bureau: every surname borne by 100 or more people, with counts. Public domain. We ship the 5,000 commonest and draw in proportion to their counts. The file also breaks each surname down by race and Hispanic origin; we do not read those columns, because weighting a name generator by the ethnicity of the people who bear a surname would tell you nothing useful about whether a name is plausible.
  • Baby names in England and Wales, 1996–2025. Birth registrations published by the Office for National Statistics. Source: Office for National Statistics licensed under the Open Government Licence v3.0. Contains public sector information licensed under the same licence. Names given to two or fewer babies in a year are redacted at source. Covers England and Wales only; Scotland and Northern Ireland register births separately.
  • English word list. Derived from SCOWL and its hunspell dictionaries, used under a permissive licence that requires its copyright notice to travel with it. That notice ships alongside the data. Used by the Username Generator and the Name Scrambler.

Everything else — the regional name corpora, the fantasy culture packs, the word banks behind the team and nickname tools — is hand-curated by us. Those lists aim to reflect what is common rather than what is interesting, and corrections to them are welcome.

Corrections

If you find a name our syllabifier splits wrongly, or a cluster we treat as illegal that plainly is not, please tell us through the contact page. Include the specific name. Those reports are the main way this gets better.