Splitting names with AI? It quietly dropped Dr. and a middle name.
The prompt most of us type
Tested in GPT-5.5 (OpenAI API), Oct 2026
We pasted 12 fictional names, including Dr. Priya Raman and Nguyen Thi Lan, under the kind of prompt most people would type.
What went missing
Real output · GPT-5.5 (OpenAI API) · Oct 2026
'Dr.' and 'Thi' are gone from the split columns, with no note. It also swapped the order of two names (probably correctly) without saying so. A second run also left out the Full name column.
Names don't share one shape
Per W3C (the web standards body), family name comes first in China, Japan, Korea and Hungary. Vietnamese names run family-middle-given, and Spanish speakers often have two family names.
Source: W3C Internationalization, Personal names around the world
Keep the name, flag the doubt
Tested in GPT-5.5 (OpenAI API), Oct 2026
The fix is to keep the original full name and ask for a Check column that says why a row is unsure. 'Best split' plus a flag is the point: it guesses, but tells you where.
Same names, with a Check column
Real output · GPT-5.5 (OpenAI API) · Oct 2026
The full name stays, and unsure rows say why. 'Thi' is back in the first name. 'Dr.' and 'Jr.' are still left out of the split, but now the row says so.
A to-check list, not a fix
It flagged 10 of 12 rows, even a simple hyphenated name, so you still check by hand. W3C also suggests asking whether you need separate name fields at all.
Source: W3C Internationalization, Personal names around the world
Before you paste a real list
Names are the data here, so strip everything else first (emails, phone numbers, account numbers, notes), and check the tool's data policy or your employer's approved tools before pasting a real contact list.
Try it: Run it on 10 rows, then check every NEEDS CHECK row.
Sources and assumptions
- W3C Internationalization: Personal names around the world: Family-name-first order (Chinese example 'Mao Ze Dong'; also Japan, Korea, Hungary); Vietnamese order familyName-middleName-givenName; Spanish speakers commonly have two family names; short words like 'de' between family names; prefixes/suffixes such as de, von and Jnr/Jr; the advice to ask whether separate given/family name fields are needed and to use the full name as the user provides it where possible. The page shows no publication date. (checked 2026-10-28)
Assumptions:
- All 12 names are fictional sample data written for this test, not real contacts.
- Tested only in GPT-5.5 via the OpenAI API (Oct 2026). The Claude runs failed to execute in the research environment and produced no output, so the post must not claim any Claude result.
- The plain prompt was run twice. Both runs dropped 'Dr.' and 'Thi' and reordered Wang Xiuying and Kim Min-jun. The second run also left out the Full name column.
- The reorders of Wang Xiuying and Kim Min-jun match W3C's description of family-name-first order, so they are probably correct guesses. The point is that the model made them without saying so, not that they were wrong.
- Results can vary between runs and model versions.
The short version
- A plain split prompt dropped Dr. and Thi.
- Keep the full name column.
- Ask it to mark NEEDS CHECK with a reason.
- Expect a to-check list, not a finished job.


