AI Chatbot Analytics Reporting in HitKeep

AI Chatbot Analytics shipped in HitKeep 2.2.0. The dashboard is built on top of structured custom events, so teams can instrument support bots, docs bots, shopping assistants, and AI search experiences without deploying a separate analytics pipeline.

The report is built on top of normal custom events, which keeps the instrumentation model simple and portable.

This page is not about inbound AI traffic. It measures your own on-site conversational assistant — the widget you shipped — from assistant.* events your code sends. AI agents crawling your site and humans arriving from ChatGPT or Perplexity are the opposite direction of traffic and are reported on the AI Agents page instead. See AI Visibility Analytics for that side.

HitKeep AI chatbot analytics with KPI cards, conversation activity chart, and chatbot breakdown panels — Conversation KPIs, timeseries activity, and chatbot-specific breakdowns on the Chatbots page.

What this report measures

The Chatbots page (labeled Chatbots in the sidebar, route /ai-chatbots) focuses on on-site chatbot usage and outcomes:

conversation starts
prompts sent
responses rendered
citation clicks
handoff requests
assisted conversions

Because this is event-based analytics, the report supports the same time ranges, audience context, and share links as the rest of the dashboard.

Recommended event schema

Use these event names consistently:

Event name	When to send it	Recommended properties
`assistant.chat_started`	User opens or starts a chatbot session	`bot_id`, `provider`, `model`, `surface`
`assistant.message_sent`	User submits a prompt	`conversation_id`, `message_index`, `intent`
`assistant.response_rendered`	A model response is shown to the user	`conversation_id`, `message_index`, `response_ms`, `tool_count`, `citation_count`
`assistant.citation_clicked`	User clicks a cited source or linked answer	`conversation_id`, `citation_url`, `citation_index`
`assistant.handoff_requested`	User asks for a human or the bot escalates	`conversation_id`, `message_index`, `reason`
`assistant.goal_assisted`	A conversion happens during or after the chatbot flow	`conversation_id`, `goal_name`, `goal_value`

Property guidance

Keep the payload metadata-focused.

Use stable identifiers such as bot_id and conversation_id.
Prefer short categorical values like provider=openai or surface=support-widget.
Store performance and usage metrics like response_ms, tool_count, and citation_count.
Do not send raw prompt or response bodies by default. Structured metadata is usually enough and is much safer from a privacy perspective.

Browser example

<script>
  const conversationId = crypto.randomUUID();

  window.hk?.event?.('assistant.chat_started', {
    bot_id: 'support-bot',
    provider: 'openai',
    model: 'gpt-4.1-mini',
    surface: 'help-center'
  });

  window.hk?.event?.('assistant.message_sent', {
    conversation_id: conversationId,
    message_index: 1,
    intent: 'billing'
  });

  window.hk?.event?.('assistant.response_rendered', {
    conversation_id: conversationId,
    message_index: 1,
    response_ms: 842,
    tool_count: 1,
    citation_count: 2
  });
<\/script>

What the dashboard shows

The Chatbots page groups these events into a dedicated workflow:

KPI cards for conversation volume, prompts, responses, assisted conversions, handoff rate, and citation CTR
timeseries chart for conversation starts, responses, handoffs, and assisted conversions
breakdown panels for intents, providers, and surfaces
audience context such as top pages, sources, devices, countries, city, provider, and ASN

You can also scope the report by provider, bot_id, surface, or model.

Recommended implementation pattern

Start by sending assistant.chat_started and assistant.message_sent.
Add assistant.response_rendered once you can measure latency and citations.
Add assistant.handoff_requested and assistant.goal_assisted to connect usage with support and conversion outcomes.
Keep property names stable so dashboard filters remain useful over time.