AI chatbots

What a dedicated AI expert looks like

AI chatbots are built on documents and data you provide. Unlike a general-purpose AI, they know your business — your products, your policies, your numbers. When someone asks a question, the chatbot finds the most relevant information in your knowledge base and generates an accurate answer grounded in that data.

Four ways to put them to work:

• On your website — Answer visitor questions 24/7 using your own data, not generic AI training
• For your team — Let staff find answers across company documents in seconds instead of searching manually
• Document analysis — Upload legal contracts, financial reports, or technical specs and ask an AI that knows the context
• As a content writer — Pick one of your own chatbots as the AI source when creating articles or running Content autopilots — every piece is on-brand from the start

General Assistant

Alongside your trained chatbots, every account includes a General Assistant — a ready-to-use assistant with no setup and no knowledge base, for everyday tasks like drafting, rewriting, brainstorming, or quick questions. It is always available and stays read-only (it can't be edited or deleted).

In any conversation — with the General Assistant or any of your own chatbots — you can choose which AI model answers, right from the chat. The choice sticks to that conversation, so one chat can run on one model while another runs on a different one.

Knowledge base

The knowledge base is the foundation of every chatbot — it's where the documents and information it will draw from are stored. Quality and focus matter more than volume: a chatbot with a small, well-scoped knowledge base usually outperforms one fed a sprawling pile of loosely related documents.

Supported file formats

Upload .txt, .doc/.docx, .rtf, .md, .pdf, and .xlsx/.xls files. Scanned PDFs are detected automatically and processed with AI-powered OCR, so printed contracts and image-based PDFs work without a separate OCR step. PDF content arrives clean: headings, numbered lists, and paragraphs are preserved — including right-to-left documents in Hebrew and Arabic — two-column layouts are read in the correct order, and page numbers and repeated headers are removed automatically. Spreadsheets are parsed row-by-row using column headers as context — ideal for product catalogs and contact lists.

Adding documents

Adding a knowledge item starts with choosing how the content comes in: upload a file, import from a URL, or write text directly. Whichever method is chosen, the content lands in an editor for review before saving — a chance to fix headings, drop irrelevant sections, and confirm the title.

1. Upload a file by dragging it anywhere onto the dialog or browsing. Multiple files process in parallel with a live progress bar.
2. Or import from a URL — AI fetches the page and extracts just the core content; navigation, footers, and cookie banners are dropped. Pages already in the knowledge base are flagged so the same URL isn't imported twice.
3. Review the extracted content and its title so the document is accurate and easy to find later, and check the cost estimate — it's based on document size and the embedding model selected.
4. Click Add. The document is extracted, indexed, and made available to any chatbot you connect it to.

Organizing with tags

Assign tags to keep the knowledge base organized (for example: by topic, product, department, or language). When creating a chatbot, filter the knowledge base by tag to give each one access to only the documents it needs — a sales chatbot might see "pricing" and "product-a"; a support chatbot might see "technical" and "onboarding".

Activating and deleting documents

Each document can be toggled active or inactive — inactive documents stay in the knowledge base but aren't searched. Deleting a document shows which chatbots use it and how many conversations would be affected before anything is removed. If a deleted document was the only source for a chatbot, that chatbot is removed with it to prevent a broken experience.

Creating a chatbot

Once the knowledge base has at least one document, navigate to AI → Chatbots and click Add chatbot.

Basic settings

1. Name and status — Give the chatbot a name (e.g., "Customer Support" or "Product Knowledge Assistant") and set it active or inactive.
2. AI model — Choose from OpenAI, Google Gemini, Anthropic Claude, DeepSeek, xAI Grok, Mistral AI, or Perplexity Sonar. Different models suit different use cases — top-tier models handle nuanced reasoning well; faster/cheaper models work well for high-volume chat.
3. Live web search (optional) — Let the chatbot research the internet in real time when the knowledge base doesn't cover a question. Answers cite their web sources, and a "Searching the web…" indicator shows while the search runs. Three things to know: it's available on web-search-capable OpenAI, Claude, Gemini, and Grok models (the toggle explains why when a selected model doesn't support it); it requires generating on your own API keys, and each search carries a small vendor fee that appears as a separate "Web search" line in the usage log; and website-widget visitors only get web search when the separate Live web search for visitors toggle in the Design tab is also enabled — so anonymous traffic never triggers paid searches unless explicitly allowed.

Instructions

After saving the basic settings, the Instructions tab is where personality, role, and behavior are defined. Two paths:

Guided setup (recommended)

A five-step wizard produces optimized instructions for any use case — customer support, document analysis, research, education, content creation, and more. No AI expertise required.

1. Purpose — Pick what the chatbot specializes in, across six categories: Customer-Facing, Industry Expert, Internal Tool, Document Expert, Educator, and Content Creator — or define a custom purpose. The wizard adapts subsequent steps to the choice.
2. Identity & tone — Name the chatbot, add a short description, and set a voice. Apply a Writing Style for consistent brand voice, pick up to 3 tone tags, or write custom tone instructions.
3. Core knowledge — Select which data-source topics the chatbot should know about, and optionally add key facts the chatbot should surface when a specific answer isn't found.
4. Behavior rules — Set preferred response language, length, formatting, and how to handle off-topic questions. Defaults are tailored to the chosen purpose.
5. URLs & preview — Add important links (website, documentation, contact), then preview the fully assembled instructions.

Wizard settings are saved and can be edited any time via Edit Settings.

Custom instructions

Below the wizard-generated instructions, a Custom Instructions field adds business-specific rules that won't be affected by future wizard edits — perfect for "Always mention the free trial" or "Never discuss competitor pricing".

Write manually

Prefer full control? Choose Write Manually to author instructions from scratch. Switching between guided and manual modes is reversible at any time.

Connect documents

Select which knowledge-base documents the chatbot can access — individual documents or whole tag groups. Toggle Include citations to show source references in replies, which builds trust and lets users verify claims.

Widget design

Make the chat widget feel native to your site. Configure:

• Position, layout, and direction — Corner placement, spacing from edges, and LTR or RTL for any language
• Branding — Pick a primary color, or enter a website URL and have the primary brand color extracted automatically
• Text customization — Toggle button text, chat title, placeholder, welcome message, leave-chat text
• Leave-details form — Optionally collect visitor name and phone at the end of a conversation
• Conversation starters — Clickable prompt suggestions that appear when the chat opens (e.g., "What are your pricing plans?")
• Input options — File uploads, image uploads, and voice recording can each be enabled for visitors
• Email alerts — Get notified when visitors start conversations or submit contact details
• Allowed domains — Restrict the widget to specific websites so the chatbot can't be embedded elsewhere without permission
• Minimum message length — Require visitors to write at least a set number of words before a message sends, cutting out spam and junk like "asdf" or random digits
• Country restrictions — Allow the chatbot only for visitors from chosen countries, or block specific ones, based on the visitor's location
• Flood auto-block — Automatically block a visitor for a while when they send too many messages in a short time, stopping spam bursts without affecting normal conversations

Embedding on a website

Once the chatbot is configured and active, embedding takes three steps:

1. Copy the embed code from the chatbot's Embed code tab — a single JavaScript snippet.
2. Paste into your website just before the closing </body> tag. Most CMS platforms and site builders have an option to add custom HTML/JS.
3. Save and test. Refresh the page, open the chat, and try a few questions.

Sharing a public chat page

A chatbot can also live on its own page — no website and no embedding required. Turn on Public chat page in the chatbot's Settings tab and a shareable link appears right below it. Anyone who opens that link can chat with the chatbot immediately, without an account and without signing in.

This is the fastest way to hand a chatbot to someone for review, put it behind a QR code, or share it in a message when there is no site to embed it on yet. The page shows the chatbot exactly as configured — same colours, welcome message, and conversation starters — so it doubles as a live preview of what visitors to the site will get.

1. Save the chatbot first. The link is built from the chatbot's ID, so it appears once the chatbot exists.
2. Turn on Public chat page in the Settings tab, then save.
3. Copy the link and share it. Use the arrow button to open it in a new tab and try it yourself.

Conversations that arrive through the link are saved like any other, and show as Public page in the Conversations tab so they stay distinguishable from website traffic.

Conversation history

Every conversation is saved automatically and available from the Conversations tab. Each card shows the chatbot, visitor domain, country, device, message count, and total cost. Open any conversation to read the full transcript with source citations, rename its topic, jump to contact details the visitor left, or export as JSON/DOCX for sharing or analysis.

Turn replies into published content

A great chatbot reply is often a finished piece of content — publish it without leaving the chat. Under any reply, open the Save menu and choose Edit & publish: the reply opens in a full editor where you can polish the text, add a featured image — generate one with AI, pick from the image library, search free stock photos, or upload your own — and publish straight to a connected WordPress site, social profile, Telegram channel, or marketplace website. Citations are cleaned up automatically, and the piece is saved under Articles at every step, so nothing is lost if you finish later.

Prefer to just keep it? Save as new article stores the reply as a draft in Articles for publishing any time.

Teach the chatbot as you chat

Every chatbot gets better from real conversations. When a reply misses the mark, or a visitor keeps asking the same question, save the correction directly from the chat — no need to reopen the instructions wizard. Every saved note is applied the next time the chatbot answers, to visitors and your team alike.

Save to memory from the chat page

A brain icon appears next to the message input after the first reply. Click it for three ways to teach the chatbot:

• Save message AND send to chatbot — type a note like "keep replies short" or "always mention the 14-day free trial", and this saves it AND sends it as a regular chat message in one click, so the new behavior kicks in right away and you can see the chatbot's next reply reflect it.
• Save message to memory — save the typed note without sending it as a chat message. The chatbot will follow it in every future conversation.
• Suggest memory notes from conversation — pick this after a rich back-and-forth and the chatbot proposes 2–3 key takeaways from what you just discussed. You choose which to keep. No typing needed.

A preview shows the polished version before anything is saved. You can also highlight any part of a reply — a floating tooltip offers Add to memory, using the highlighted text as the note's reference.

Memory tab in the editor

The Memory tab lists everything the chatbot has learned. Each note shows how long ago it was taught and how many conversations contributed to it. From there: edit a note inline to refine the wording, pause a note to temporarily exclude it (paused notes stay visible for reactivation), or delete a note for good.