We sometimes try short videos and read discussions in the writing community. Under one of the videos on writing, we came across this comment:

“what you discuss in the chat is not a basis for training AI, and I’ll even say more, if you create a new chat from your account, it won’t know what you discussed in your other chats with it. so nothing will be stolen from your chat, but if you post it online, then yes”


The problem is that chat memory and model training are not directly related. A new chat may indeed know nothing about the previous one. This does not prove that the sent messages and files were not used to improve the model.

Regular user chats (ChatGPT, Claude, DeepSeek, Qwen, Alice, GigaChat) may use uploaded data for training. The terms depend on the specific product and account type. In ChatGPT and Claude, this can be limited in settings. Other chats may have different rules, so it’s better to check them before uploading a manuscript.

Specialized services usually work via API. In this case, the text is passed to the model to fulfill a specific request and is not used by the provider for training by default.

You can also connect to the API yourself, but you’ll have to configure keys, payment, requests, and message history transmission. A specialized service handles this part for you.

LitCritic also works via API. Texts are passed to the model only for analysis or response generation.
We’ve put together cards and tried to explain in simple terms the difference between a regular chat and a specialized service.

Sharing cards from our TikTok