What it does

AnswerRails puts a chat assistant on your WordPress site that answers visitors’ questions from your own content, shows where each answer came from, and says so when the answer isn’t there.

On the visitor’s side

A visitor sees a small launcher in the corner of your pages. Opening it gives them a chat window with your name on it, a welcome message you wrote, and a box to type in. On a phone the window takes over the screen; on a desktop it can be dragged and resized. It follows the visitor’s light or dark preference, and it is built to be operated by keyboard and screen reader as well as by mouse.

They ask a question. The assistant looks for the answer in the material you have given it — your pages and posts, documents you have attached, and structured data — and replies in plain language. Under the reply is a short Sources list naming the pages it drew on, so the visitor can check the answer against the original. When your content doesn’t hold the answer, the assistant says so rather than improvising; When it says “I don’t know” explains exactly how that decision is made.

A conversation lasts as long as the visitor keeps it going. Two hours without a message and the session is dropped; what the browser keeps between page loads, and for how long, is a setting you control (see Where your data goes).

Where the answers come from

The assistant answers from a knowledge base you build inside WordPress. You choose which pages and posts belong in it, attach documents, and add JSON for structured facts such as price lists or part numbers. The knowledge base is stored and searched in your own WordPress database; only the excerpts relevant to a question are shared with your AI provider to answer it.

Alongside the knowledge base, every answer is written with a standing list of the real, published pages on your site in front of the model — titles and addresses only. That is what lets the assistant link visitors to the right page, and what stops it inventing a page that doesn’t exist.

The model itself is one you choose from OpenAI, Mistral or Anthropic, running on your own account and your own key. AnswerRails is the layer in between your site and that provider; it does not sit in the middle of the conversation and it does not resell anyone’s API. Your key, your bill covers who pays whom.

What happens to every message

Each visitor message is checked by your provider’s moderation service before any answer is attempted, on every plan, with no off-switch. Abusive or harmful input is deflected rather than answered, and repeated abuse ends the session. Whatever the model writes back is treated as untrusted until it has been checked: HTML is stripped, links are verified against the pages that actually exist on your site, and an answer that claims more than your content supports is replaced by an honest one.

What you see

In the WordPress admin, AnswerRails gives you an Overview that shows each stage of the pipeline — visitor question, input guardrails, knowledge search, the chat model, and output guardrails — with a live indicator on each and a one-click test of every service the assistant depends on. Errors are translated into plain language (the provider is busy; the key is wrong; the account is out of credit) with the action to take.

Chat Logs, if you turn logging on, show each conversation with the path every turn took and the sources it used. The same screen holds the privacy tools: find a visitor’s records, export them, anonymise them or delete them, and see the audit trail of who did what. Knowledge is where you add and manage sources, with a bulk picker for pages and posts and a sync status on every item.

Free and Pro, in one paragraph

Everything above is in the free plugin: the widget, any provider and model on your key, the knowledge base with local search, citations, the honest refusals, moderation, the health screens, chat logs, and the complete set of consent, retention and data-request tools. Pro adds five things the conversations can do for your business afterwards — your WooCommerce catalogue and external web pages as sources, a Business Intelligence screen, capture of product ideas and sales leads from real conversations, and human-handoff escalation by email — plus the control surfaces for retrieval depth and moderation tuning. Free vs Pro draws the line precisely.