WordPress AI chatbot that knows your content and runs workflows

FlowMattic AI Assistants give you a WordPress AI chatbot you build, train and place yourself. Each assistant answers visitors from the content you add, can collect leads and meeting requests in the chat, and starts a FlowMattic workflow when something happens in the conversation. It runs on an AI connection you own, chosen from eleven providers.

It suits support teams who want common questions answered on the site, marketers who want leads captured mid-conversation, and agencies running a different assistant for each client. This guide covers the assistant editor tab by tab, how the knowledge base works, how chats start workflows, and where the limits are.

What are AI Assistants?

AI Assistants is a FlowMattic screen at FlowMattic → AI Assistants. Click Create New Assistant, give it an Assistant Name, pick an AI Connection and save. The editor then opens full screen with five tabs, Setup, Content, Actions, Theme and Conversations, and a live preview on the right that you can switch between Desktop Preview and Widget Preview.

You can create as many assistants as you need, each with its own connection, model, instructions, content and look.

How to set up a WordPress AI chatbot

1. Setup: identity and model

On the Setup tab, set the Chatbot Name and Description, choose the AI Connection and AI Model, and write the Instructions (up to 32,768 characters) that shape how it answers. Add a Welcome Message and an Input Placeholder. Temperature is optional; left blank, the provider’s default applies. Load User Session Chats decides whether a returning visitor sees their earlier conversation.

Connections come from Connects, and the assistant can run on OpenAI, Anthropic (Claude), Google (Gemini), Groq, Mistral, xAI (Grok), Perplexity, DeepSeek, OpenRouter, Straico or AWS Bedrock.

2. Content: what it should know

The Content tab holds the assistant’s knowledge sources:

• Upload Files: text-based files such as .txt, .docx, .pdf, .html, .json and .md, picked from the Media Library.
• Crawl Webpages: add URLs and click Crawl All. Each URL is read as a single page.
• Connect Tables: sync a FlowMattic table, manually or automatically when rows are added or updated.
• Sync WordPress Posts: choose post types and click Sync Now. Posts sync in the background and re-sync when you update them.

With an OpenAI connection, sources go into an OpenAI vector store, shown as the Vector Store ID on the Setup tab. With any other provider, uploaded files and crawled pages are split into chunks and embedded into an index stored in your uploads folder, using an OpenAI, Google, Mistral or Voyage AI embeddings connection on your site. When a visitor asks something, the closest chunks are passed to the model as context.

3. Actions: what the chat can do

The Actions tab connects the chat to the rest of your site:

• Event and Action: on Chatbot Response Generated or Chat Started, either Trigger Workflow (pick the workflow) or Send to Webhook (the conversation data is posted as JSON).
• Suggestions: starter questions shown before the first message.
• Collect Leads: an Embedded Form or Chat messages, asked at the beginning, after a few messages or when the chatbot cannot answer, with fields such as Name, Email and Phone.
• Link Button: a button with your text and URL, always shown or when certain keywords come up.
• Schedule Meeting: paste a Calendly, Cal.com or Google Calendar booking link in Calendar Embed URL, or offer built-in time slots.

4. Theme and embed

The Theme tab offers 14 presets plus Custom, light or dark mode, a Brand Color, a background image, Widget Position, Loading Animation, launcher icon, Bot Avatar and Button Label, with fonts, bubble styles and colours under Advanced Appearance. Then click Embed for three options: a shortcode for any post or page, an iFrame for any web page, or a Popup Widget script for a floating chat button.

How a conversation starts a workflow

There are two ways to connect a chat to a workflow. The first is the Actions tab’s Trigger Workflow action, which runs the chosen workflow when the assistant replies. The second is the FlowMattic Chatbot trigger app in the builder, with the events AI Response Received, Lead Captured, Meeting Scheduled and Chat Started.

The workflow receives the visitor’s latest message, the assistant’s reply, the conversation and chatbot IDs, the visitor’s email when they are logged in, and lead or meeting details for those events. Click Capture Response on the trigger and send a test message in the chat to map them. The workflow runs after the reply has been sent, so it acts on the conversation rather than changing the answer.

Use cases

Leads from the chat go straight to your CRM

Collect Leads asks for a name and email when the chatbot cannot answer. The workflow creates the contact, tags the source and tells sales in Slack.

Lead Captured → FluentCRM create contact → Slack message

Meetings booked and confirmed

A visitor picks a built-in time slot. The workflow adds the event to your calendar and sends the confirmation email with the booker’s name and time.

Meeting Scheduled → Google Calendar create event → Send email

Refund requests escalated to the help desk

When a visitor mentions a refund, a Filter lets the run continue, a ticket is opened with the transcript and the support team is alerted.

AI Response Received → Filter (message contains "refund") → Help desk create ticket → Email

Find the gaps in your docs

Log every question and answer to a table. Once a week, an AI Agent reads the log and lists topics the assistant struggled with, so you know which page to write next.

AI Response Received → WP Tables new record, then Schedule → AI Agent → Send email

One assistant per client for agencies

An agency runs an Acme support bot on Claude and a Northwind pre-sales bot on Gemini, each with its own instructions, content and theme, and routes their leads to each client’s inbox by chatbot ID.

Lead Captured → Router (by chatbot ID) → Email to the right client

SaaS onboarding with next steps

Sync your help articles, add Suggestions such as “How do I invite my team?”, and a Link Button to pricing when someone asks about plans.

Visitor question → Assistant answers from synced docs → Link Button to /pricing/

Conversations into your existing tools

Use Send to Webhook to post each reply to an outside system that already accepts webhooks, with no workflow in between.

Chatbot Response Generated → Send to Webhook → External system

Tips and gotchas

• Post and table sync need OpenAI. Sync WordPress Posts and Connect Tables feed the OpenAI vector store. On other providers, use Upload Files and Crawl Webpages, which feed the local index.
• Add an embeddings connection (OpenAI, Google, Mistral or Voyage AI) before adding content to a non-OpenAI assistant, so files and pages can be indexed.
• The trigger app fires for every assistant. If you run several, check the chatbot ID in a Filter or Router so each workflow only handles its own assistant.
• Workflows act after the reply. A triggered workflow cannot write back into the chat. Send results by email, Slack or a ticket instead.
• Streaming depends on the provider. OpenAI, Anthropic and Gemini stream replies word by word; the other providers deliver the full reply at once.
• Some features are OpenAI-only. The Chat messages lead style and the Chat Started event currently work with OpenAI connections.
• Know where chats are stored. For OpenAI assistants, messages are kept by OpenAI and the Conversations tab reads them from there. For other providers, the full history is stored in your WordPress database.
• Public endpoints are rate-limited to 20 chat requests per visitor IP per minute, and every reply is sanitised before it is shown.
• Refusals end cleanly. If the provider rejects a request, for example over a rate limit or an invalid key, the widget says so instead of leaving an empty reply. This works the same on OpenAI, Anthropic and Google: visitors see a short message and logged-in managers see the provider’s reason.
• Waiting and trying again is opt-in. By default a refused request fails straight away. A developer can return a wait of up to 30 seconds from the flowmattic_chatbot_retry_delay_seconds filter, and the same request is then sent once more, which helps with short rate limits.
• No automatic placement. The assistant appears only where you add the shortcode, iFrame or widget script.

Related features

• Connects: save each provider key once and choose it per assistant.
• AI Agents: reasoning steps inside the workflows a chat triggers.
• WP Tables: log conversations and leads in your own database.

FAQ

Which AI providers can a WordPress AI chatbot in FlowMattic use?

OpenAI, Anthropic (Claude), Google (Gemini), Groq, Mistral, xAI (Grok), Perplexity, DeepSeek, OpenRouter, Straico and AWS Bedrock, each through a connection saved in Connects. Every assistant picks its own connection and model.

Can the chatbot answer from my own content?

Yes. Add files and web pages on the Content tab, plus post types and tables when the assistant uses OpenAI. The closest passages are found for each question and given to the model as context.

How does a conversation start a workflow?

Choose Trigger Workflow on the Actions tab, or use the FlowMattic Chatbot trigger in the builder. The workflow receives the message, the reply and any lead or meeting details.

Can I put the assistant on a site that is not WordPress?

Yes. The Embed dialog gives you an iFrame and a Popup Widget script, both served from your WordPress site, which work on any web page.

Ready to put an assistant on your site? Try FlowMattic, connect a provider, add your content and embed the widget.