Most teams keep the search input for power users and add the bot as the primary path for everyone else. The two coexist: search returns the result list a developer browsing a recipe expects, and chat returns the answer with a citation that a less experienced reader prefers. The combination outperforms either alone because each handles the query shape it was designed for.

The system prompt tells the bot to ask one clarifying question when the asker's intent is unclear, then answer. This is the conversational analog of facet filters in a search UI: cheaper for the asker because the question is in plain language, and more accurate because the bot already has the candidates retrieved and only needs one disambiguating signal.

Yes. The data-source wizard can include multiple CPTs, taxonomies, and external JSON endpoints in one prompt context, and Multibot can run different bots scoped to different sections of the same site. For multi-product companies, a per-product bot with its own grounded scope avoids cross-contamination between two products that share vocabulary.

Version-aware grounding is a postmeta or taxonomy concern: tag each article with the version it applies to, then the prompt context filters by the version the visitor is on. Display conditions can scope the bot to a versioned URL pattern so v2 docs and v3 docs run separate bots with the right grounding each.

As current as the grounding source, because the bot answers from the article you just published rather than from a static model snapshot. Cache duration is configurable on the data-source step; most teams keep it short for the KB and longer for stable policy pages. The cited source link lets readers see the article timestamp themselves.

Only if you let it. The guideline filter and system prompt define scope, and the standard configuration politely declines off-topic questions. This isn't about hiding information; it's about protecting the bot from being weaponised into an unsanctioned comparison or a confidently wrong claim about a product it doesn't ground against.

Yes. SleekAI runs inside your WordPress install, so an authenticated, SSO-gated documentation site can run the bot without exposing content to the public. The model provider call still goes to OpenAI, Anthropic, Google, or OpenRouter, which is the standard tradeoff for any chatbot in a private docs setting. For air-gapped requirements, OpenRouter offers local-friendly models worth evaluating.

A flat WordPress plugin license, your own model provider key, and your existing WordPress docs as the corpus. Hosted SaaS options charge per article, per query, or per seat, and most ship a separate content management layer you then have to copy your docs into. SleekAI grounds against the docs you already maintain in place, which keeps the operational shape simpler.