Improving Korean Character Counting Accuracy in ChatGPT

Hi OpenAI team,

I’m writing as a Korean-speaking user to report an ongoing issue with Korean character counting in ChatGPT. While English text is counted very precisely, Korean character counts are often inaccurate due to Hangul’s syllable block system. For instance, “가” is counted as one character by humans but is often miscounted by the model. This is not a technical limitation—platforms like Naver handle this perfectly—it’s an issue of prioritization. And it matters, because accurate character counting affects resume writing, contests, social media, and more.

This is not just a feature request. It’s a usability concern tied directly to user trust and practicality.

Please consider this feedback for future updates. Korean deserves the same precision as other languages.

(P.S. This message was written with the help of ChatGPT itself—aka “Monday”.
Sure, it still counts Korean characters wrong… but it’s still the best partner I’ve got.)

1 Like

I think this occurs for similar reasons as most models can’t answer correctly “How many r’s are in strawberry?”.

They “see” things as tokens, and not as characters.

However, if I’m not wrong, these are more like syllables. Maybe if you ask for how many syllables and not how many chars, it may answer differently.

I know that you meant that the model “should” know this, but they don’t necessarily know.

AI is not good to count letters or words.
You should use python.

Prompt:

Count the number of visible Hangul syllables in the following Korean sentence. A Hangul syllable is one complete block such as "가", "한", or "의", and each block should be counted as 1 character, regardless of internal jamo.

Text: "안녕하세요. 저는 인공지능입니다."

Rules:

- Only count complete syllable blocks.
- Do not count individual jamo (ㄱ, ㅏ, etc.).
- Return only the character count, no explanation.
- Use python