ChatGPT auto-generated conversation titles switch from English to Chinese despite Language being set to English, especially inside Projects

What version of the Codex App are you using (From “About Codex” dialog)?

26.908.70816

What subscription do you have?

Plus

What platform is your computer?

Microsoft Windows NT 10.0.26200.0 x64

What issue are you seeing?

Description:
My ChatGPT language setting is explicitly set to English, and I usually start these conversations entirely in English. However, ChatGPT frequently generates the final conversation title in Chinese.
This does not appear to be random text corruption. The Chinese titles are semantically correct summaries/translations of the English conversation topic. The issue appears to be specifically with the language chosen by the automatic title-generation system.
I have also noticed that the title often initially appears in English, but after the assistant finishes generating the response, the title is automatically
replaced with a Chinese one. This suggests there may be a later/background title-generation or retitling step that is choosing the wrong language.

Language setting:
ChatGPT → Settings → General → Language: English
Browser/UI language: English
Conversation prompts: English

What steps can reproduce the bug?

Reproduction:
I tested the same general behavior both outside Projects and inside a newly created empty Project named “Test for Language.”
The new Project had:

• no previous conversations
• no uploaded files
• no special Project instructions
• an English Project name

I created five English conversations using prompts such as:

• “bloomberg, whats it”
• “show what's hot at Netflix”
• “what's the trend today”
• “who is elon musk”
• “sydney sweeney sports ad”

Results outside Projects:
3/5 titles were English and 2/5 were Chinese.
Examples included:

• Sydney Sweeney Sports Ad
• Elon Musk简介
• 今日趋势汇总
• Show Netflix Hot Titles
• Explain Bloomberg

Results inside the new empty Project:
Only 1/5 titles remained fully English, while 4/5 were Chinese or mixed English/Chinese.
Examples included:

• Sydney Sweeney Sports Ad Details
• Elon Musk简介
• 今日趋势摘要
• Netflix热门推荐
• Bloomberg介绍

This suggests that the problem is not caused only by existing Chinese conversations within a Project, because the Project used for testing was completely new and empty. However, the issue appears to occur more frequently inside Projects.

What is the expected behavior?

When:

• ChatGPT's language is explicitly set to English, and
• the conversation itself is entirely in English,
  the automatically generated conversation title should also be in English.
  Ideally, the title language should follow the current conversation language, or at minimum the explicit ChatGPT language setting.

Additional information

Another important observation
In many cases, the conversation title appears in English first while the response is being generated. After the response finishes, the title changes to Chinese.
Therefore, it appears possible that:

1. an initial/provisional English title is created;
2. a later automatic title-generation process runs;
3. that process incorrectly chooses Chinese as the output language;
4. it replaces the original English title.

Additional possibility:
I sometimes use Chinese in other ChatGPT conversations. I am therefore wondering whether a recent/account-level conversation-language signal is incorrectly influencing the title generator, even when the current conversation is English.
I have attached screenshots showing the comparison between normal chats and chats inside a newly created empty Project.
Please let me know if logs, conversation IDs, timestamps, browser information, or additional controlled testing would help reproduce this issue.

No source files or tests are identified. Start by reproducing the five English conversations both outside and inside a new empty Project, checking the title before and after the response completes. Done means titles remain English when the language setting and conversation are English; collect conversation IDs, timestamps, logs, and browser details if the behavior recurs.