The three generations
Rule-based bots. Button-driven decision trees: "Press 1 for shipping." Cheap, predictable, and frustrating the moment a visitor's question is not on a button. Still the right tool for narrow, high-volume flows like order-status lookups.
LLM chatbots. Free-form conversation powered by a large language model, usually grounded in the merchant's own content so answers stay accurate and on-brand. This is the current default for website support and pre-sales questions.
Agents. Chatbots with hands: they combine an LLM with tool access, so they can act — check inventory, start a return, capture a lead, trigger a discount. See AI sales agent and the AI agents for websites guide.
Reactive vs proactive
Most chatbots are reactive: they wait in the corner until clicked. A proactive chatbot opens the conversation itself based on visitor behavior — an exit signal on a full cart, a long dwell on a pricing page. The trigger design is its own discipline; see proactive chat.
Picking one for a store
The evaluation comes down to four questions: what it answers from (your data or generic training), what it can do (answers only, or actions), how it loads (one async script or a 500KB bundle that hurts Core Web Vitals), and what it costs at your conversation volume. Platform-specific comparisons live in the Shopify chatbot guide and the WooCommerce chatbot guide.
Related terms
- AI sales agent — the action-taking evolution
- Proactive chat — bot-initiated conversation
- Onsite conversion agent — the conversion-focused category
Last updated July 6, 2026.