Almost every business has a document library that is more useful than its website: a product manual, a sales playbook, a benefits PDF, a research paper, a compliance booklet. These files do not have URLs, they do not get indexed by search engines, and they certainly do not get read by chatbots crawling your sitemap. Making them queryable by a chatbot used to mean uploading to a vendor's vector store with its own dashboard, its own billing, and its own data residency story to argue about with security.

SleekAI's document mode keeps the documents and the index inside WordPress. Upload a PDF or DOCX through the plugin's Documents tab and it is parsed, chunked, embedded with your configured embedding model, and stored in a SleekAI table alongside source filename, page number, and chunk position. At chat time the retriever ranks chunks against the user's question and the top ones are inserted into the system prompt with their source markers. Citations link back to the document and the specific page, so when the bot says it answered from page 14 of the benefits guide, the user can verify.

The non-obvious part is metadata. SleekAI lets you tag each document with categories, restrict it by user role, and assign it to specific bots. A benefits PDF can be available only to logged-in employees through an internal bot; a public product brochure can be shared with the marketing bot. The same plugin runs both because document scope is part of the bot configuration, not a separate platform. When you replace a document with a new version, the old chunks are removed and the new ones reindexed without touching the bots that referenced the file.

A document is often a more honest source than a web page. It is the version that legal signed off on, the manual the engineers actually maintain, the playbook the sales team prints out before the quarterly offsite. When a chatbot answers from those documents, the answer carries the weight of the source.

When a chatbot answers from a vendor's crawl of your marketing site, the answer carries the weight of whoever last updated that page. The strongest case for document training is content that intentionally does not live on the public web. Internal benefits guides, NDA-bound research, paid product manuals, regulated compliance booklets.

Sending those to a vendor's vector store invites a long compliance review and an annual security questionnaire. Keeping them inside WordPress lets you reuse the access controls and the backup story you already have. Roles and capabilities already determine who sees which content; documents inherit the same model.

Backups already cover wp-content and the database; document chunks and embeddings are inside both. The chatbot becomes a feature of the site rather than a separate service to govern. That is the difference between a tool you can deploy in a week and one that lives in legal review for a quarter.