Is my site’s content sent to OpenAI to create the embeddings?

This is an FAQ entry for the WordPress chatGPT SmartSearchWP plugin.

Yes, when the embeddings are generated. To turn a piece of content into embeddings, SmartSearchWP sends its text to the OpenAI embeddings API, which returns the corresponding vectors.

The text sent includes:

• the content of the post, page or other content concerned;
• the ACF fields you have selected, with the Custom Post Type & ACF addon;
• the text of PDF files, with the PDF addon.

Long content is split into several parts before being sent.

Where are the embeddings stored? #

The vectors returned by OpenAI are stored in your WordPress database, attached to the content concerned. See: Where are the AI embeddings stored?

And during conversations? #

When a visitor asks a question, the question itself is also converted into an embedding by OpenAI. The comparison with your content’s embeddings, which selects the contexts used for the answer, is calculated on your server.

What OpenAI does with this data #

Data received by the OpenAI API is not used to train its models unless you agree to it, and may be kept for up to 30 days in its abuse monitoring logs. Details: Does Open Ai retain the content of requests sent?

What about the cost? #

Each embeddings generation uses your OpenAI credits. That is why, in the free version, regeneration is done manually; the Premium RAG Auto Sync addon can automate it.