Glossary
Proactive Chat
The chat starts the conversation. A widget opens itself, or an agent sends the first message, when a rule fires: thirty seconds on the pricing page, a second visit, exit intent, or a scheduled check-in.
General definition
Proactive Chat flips the default: instead of a passive launcher waiting to be clicked, the system watches for a signal and reaches out first. In website support it is a chat widget that opens with a targeted question; in messaging products it is an AI agent or bot that posts into a room on a schedule or after a period of silence. The goal is the same: catch a moment when help or a nudge is valuable, not just possible.
- Behavioural triggers: time on page, scroll depth, repeat visits, a particular URL such as pricing or checkout, or an error page
- Intent triggers: exit intent (the pointer leaving the viewport), an abandoned form, a high cart value
- Context triggers: a logged-in customer with an open ticket, a trial nearing its end, a patient with an upcoming appointment
- Time triggers: a scheduled check-in, a follow-up after idle time, or office hours opening
Done badly it is the pop-up everyone closes. Guidelines that hold up: one proactive message per session, a specific question rather than a generic greeting, respect for a dismissal, no proactive prompts on mobile at first load, and a clear path to a human. Measure it: proactive conversations should have a higher acceptance and resolution rate than reactive ones, or the trigger is wrong. In regulated settings, proactive outreach must also respect consent, quiet hours and the rules on what can be said in an unauthenticated channel.
Proactive chat overlaps with marketing automation and with notification design; the difference is that the reply lands in a live conversation, human or AI, rather than an inbox.
In the Ethora ecosystem
Ethora AI agents can open the conversation. In the AI SDK an agent attached to a room or channel can be configured with a heartbeat scheduler so it speaks first on a schedule (a daily summary at 09:00, for example) or after a period of idle time, with a system prompt that defines what it says and a response gate and autonomous-turn cap so agents in multi-agent rooms do not talk endlessly to each other. Per-room transcript memory lets the opening message refer to what happened last time.
On a website, the embeddable assistant widget and the chat widget SDK give you the launcher and the conversation; opening it on a trigger is a few lines in the host page, and the same agent, knowledge base and Trust and Safety rules apply whether the visitor clicked or the widget spoke first. See the AI SDK for agents, RAG and scheduling.
