Adding content to the index

The four ways to add content to the Aimogen Pro embeddings index, automatic indexing of new posts, and how the index is organised.

Aimogen Pro › Advanced Tools › AI Embeddings

Four ways in, each on its own tab.

1. Add a new embedding manually#

Type or paste text directly. Good for facts that exist nowhere else on your site — policies, FAQs, internal knowledge.

Each entry is stored as an aiomatic_embeddings post, so it can be edited later.

2. Upload from a file#

Upload a document and index its contents. Good for manuals, specifications and existing documentation.

Larger documents are chunked automatically.

3. Scrape data from a URL#

Give a URL; the page is fetched, cleaned and indexed. Good for indexing your own documentation site, or public reference material.

4. Index existing posts#

The most-used option. Select post types and index your existing content in bulk.

For a content site, this is usually the starting point: index your posts, and the chatbot can answer from your archive.

Automatic indexing#

Settings › Embeddings › Enable Embeddings Auto Indexing For Newly Published selects post types that are indexed automatically as they are published.

Indexing is deferred to the aiomatic_index_single_post cron event rather than done during the publish request, so publishing stays fast.

The embedding template#

The template decides what text represents a post in the index. It accepts placeholders:

Title: %%post_title%%
Categories: %%post_cats%%
Tags: %%post_tags%%
URL: %%post_link%%

%%post_content_plain_text%%

Including the title and taxonomy matters: retrieval matches on the indexed text, so a chunk that includes its title is easier to find than a bare paragraph.

Including the URL is useful — the chatbot can then cite where an answer came from.

Rewriting content before indexing#

Why bother: raw post content contains navigation text, calls to action and formatting noise that dilute the embedding. A rewrite that condenses a post into its factual content produces sharper retrieval.

The cost is an AI call per indexed document, which is significant for a large archive. Worth it for a knowledge base; usually not for a blog.

Namespaces#

A namespace partitions the index. Content indexed into support-docs is not returned when searching product-pages.

Set them so that:

• A support chatbot searches documentation
• A sales chatbot searches product content
• Content generation searches reference material
• Each agent has its own memory space (agents_<id>, automatic)

Each context under Settings › Embeddings › Enable Embeddings For has its own namespace field, and the embeddings_namespace shortcode attribute overrides it per placement.

The aiomatic_embedding_namespace filter lets code choose the namespace dynamically. See Filter reference.

Managing the index#

List Added Embeddings shows what is indexed, and lets you edit or delete entries.

Because entries are aiomatic_embeddings posts, they are included in your database backup.

Keeping the index current#

This is the part that gets neglected. An index reflects content as it was when indexed.

• Auto-indexing handles new posts. Enable it.
• Edited posts need re-indexing. Re-index the affected posts.
• Deleted posts leave orphaned vectors. Delete the corresponding entries.
• After a bulk edit, re-index everything that changed.

A stale index is worse than no index, because the chatbot answers confidently with outdated information.

Related#

• Vector stores
• Embedding models
• Retrieval troubleshooting