How our kanji names are made and checked

Every name on this site is written two ways: in katakana, the standard Japanese spelling of a foreign name, and in kanji chosen for their sound (ateji — see katakana, kanji and ateji). The kanji renderings and their descriptions are produced with the help of an AI model. What makes them usable is what happens around the model: a dictionary decides what each character can sound like, a fixed list decides which characters are never used, and every published rendering has been audited.

By the numbers

1,059

names published, every one of them covered by the audit below

15,931

renderings audited, character by character

10,423

renderings whose per-character sounds are verified against the kanji dictionary

255

renderings in which an inauspicious character was replaced

2,522

descriptions rewritten

284

flagged meanings examined and cleared

1

rendering withheld from the site after review

How a rendering is made

1. The sound comes first. Before any kanji are chosen, the name is resolved to one katakana spelling — Andrew to アンドリュー. The answer is only accepted if it is pure katakana, breaks down completely into Japanese syllables, and is a plausible length for a name. It is then stored, so every later rendering of that name aims at the same sound.
2. Candidates come from the dictionary. For each syllable, candidate kanji are drawn from the jōyō and jinmeiyō lists — the characters in everyday use and those approved for Japanese personal names — using each character’s dictionary readings. The model is shown real candidates for each sound instead of being left to come up with its own; in our tests that cut the share of characters that do not actually make the intended sound from about 50% to about 6%.
3. Some characters are never used. Twenty characters are excluded outright and four more are excluded for certain themes, from a list reviewed by a native Japanese speaker — characters associated with death, misfortune or cruelty that no Japanese parent would put in a name, whatever their literal meaning. A rendering containing one is rejected before it is stored.
4. Each character’s sound is checked. Once a rendering exists, each kanji is matched to the part of the name it spells, using only the dictionary: 晏 an + 弩 do + 竜 ryu for Andrew. A match is recorded only when exactly one complete alignment exists; if the dictionary does not settle which character carries which sound, nothing is shown rather than a guess. That is why some renderings on the site show the sound of each character and some do not.

How the published renderings were checked

The names published on this site were first generated in June 2026, before the process above was fully in place, and were then audited rendering by rendering. The audit checked that each character’s stated meaning belongs to that character, that it fits the rendering’s theme, that it has a reading, and whether any meaning is negative or any character is on the excluded list. Its findings are the numbers above: inauspicious characters replaced, descriptions rewritten, flagged meanings examined and cleared.

A rendering whose flag has not been cleared is not shown. A name that has not been through review is not listed on the site or offered to search engines, even if someone has generated it.

What this does not promise

Ateji are an artistic spelling, not an official one, and no set of kanji is the single correct rendering of a name. The descriptions are written with an AI model and reviewed, but they interpret the characters rather than state facts about them. If a tattoo or anything else permanent depends on it, read the meaning of every character and, ideally, show it to someone who reads Japanese.