How to get as relevant results via API such as using ChatGPT whatever in 3.5 or 4?

First of all, I’m glad to join your community !
With the same formulation of the prompt, I still get very bad results while asking dietary classification of a list of recipes via API with model GPT-3.5-turbo-instruct regarding the results from ChatGPT 3.5 or 4.
I use french language.
Advices are very welcome…
Can’t wait from hearing from you,
Take care,

Hi and welcome to the Developer Forum!

Please give some examples of prompts you have used and replies you have received and what you think is wrong with them.

Are you using a RAG (retrieval augmented generation) implementation sending context documents (I assume your recipe documentation) along with the question, or just querying the model to us it’s own training for the recipes?

I can tell you from experience, in either case, translating from one language to another sometimes causes a little to a lot to be lost in the translation. This would be the case in particular with the OpenAI models if you are relying on their training as it, I assume, is mostly in English.