When it fires
Once per post in every sync of the knowledge base, after the post’s text was extracted (from its rendered content, or from the file for a PDF) and before it is split into chunks. What the filter returns is appended to the text after a blank line and is indexed with it: it can be found, it is quoted as a source, and the AI reads it.
KimaAI hooks it itself at priority 10 for the content templates (Embedding, Content templates), which write custom fields, taxonomies and ACF fields into the text per post type. Add to $text instead of replacing it, or their text is lost.
The post’s chunks are embedded again on its next sync when the added text changed.
Parameters
| Name | Type | Description |
|---|---|---|
text | string | Text appended to the post’s text; empty unless a content template added some. |
post | WP_Post | The post being prepared for the knowledge base. |
Returns
string: The text to append.
Example
add_filter(
'kimaai_embedding_extra_text',
function (string $text, WP_Post $post): string {
if ($post->post_type !== 'product') {
return $text;
}
// The delivery time is a custom field the product page shows outside the content.
$delivery = (string) get_post_meta($post->ID, 'delivery_time', true);
if ($delivery === '') {
return $text;
}
return trim($text . "\n\nDelivery time: " . $delivery);
},
10,
2
);
Related
kimaai_embedding_chunk_text: changes what is embedded per chunk, without changing the stored text.kimaai_embedding_vector_metadata: stores data with each chunk instead of in its text.