“How are they processed?” - no, there was not an answer provided to that or other questions.
GPTs are search-happy, so you don’t have to specify files, although it can benefit the operation.
Here for example, there is no indication that my knowledge is anything more than toolmaker.txt, and indeed, there is no information related to the question, but a search was (quickly) emitted regardless:
The biggest problem, the same as with Assistants, is that the description of the file search given to the AI is that “the user has uploaded files” - regardless that the GPT maker included them. This gives such confusing results as the text “as used in your tools”, even though there’s only GPT-provided documentation and the user has no tools.
You can’t block a user from still uploading files, and then those uploaded files can distract from the purpose and knowledge specific to a GPT and be commingled with them, especially given that a more advanced GPT may also make use of external actions that rely on file data, can become quite problematic.
Also, another API difference is that any files in shared GPTs basically become OpenAI property, and you cannot revoke their perpetual license to use them as they see fit. Another reason to invest your time in your own API-developed platform.