Point SleekAI at the post type that holds your glossary entries (often a custom 'glossary' CPT, or your default posts tagged with a 'glossary' category). The bot reads title, definition, longer explanation, related terms, and any ACF or metadata you maintain. The data-source wizard maps those fields into named context so the system prompt can reference them by name rather than blob.

Yes. The system prompt instructs the model to end every reply with a link to the canonical glossary entry it drew from, so visitors can verify the definition and see the longer form. For multi-term answers, the bot can link two or three entries. That citation pattern is what keeps the glossary as the authority, with the chat as the conversational front.

Yes. Encode related terms as a meta field or taxonomy on each entry, and the bot can answer 'how is X different from Y' or 'what is the opposite of Z' by reading both entries into the prompt. The reply ends with both canonical links, so the visitor has the full context after the chat closes.

The system prompt can tell the bot to say so explicitly: 'I do not have a glossary entry for that term yet. Would you like me to log it for the editorial team?' The conversation log makes those gaps visible. Most teams find the chat reveals missing-entry requests they would never have got from raw search-console data.

Yes. Multibot lets you run one bot per glossary scope: a developer glossary on /docs/glossary/, a finance glossary on /investors/glossary/, and a clinical glossary on /providers/glossary/. Each bot has its own system prompt, its own source set, and its own tone. Display conditions handle the routing automatically.

Yes. If your glossary has translated entries (via WPML, Polylang, or a sibling CPT per language), point each language's bot at the matching entry set. The bot replies in whatever language the visitor writes in, and quotes the matching translated entry rather than paraphrasing across languages.

For terms with multiple meanings ('handler' in JavaScript versus 'handler' in event sourcing), the system prompt instructs the bot to ask one disambiguating question before answering. The conversation log surfaces those disambiguations, which is often a signal that two glossary entries should exist instead of one. Editorial spots the pattern and adds the entry, the bot improves on the next request.

For regulated domains (legal, medical, financial glossaries) the system prompt should explicitly forbid advice and steer back to definition. 'I can explain what an indemnity clause is, but I cannot tell you whether to sign one. For that, consult a qualified lawyer.' The guideline filter is on by default and helps keep the bot inside that boundary.