Live chat and messaging transcripts
Transcripts of live chat and in-app messaging between customers, chatbots and human agents, from operating companies and sourced to order. They show short, fast exchanges, handoffs from bots to people and what happened next. Labs use them to train conversational agents and to evaluate when an agent should escalate.
What a record contains
One conversation: messages with timestamps and participant roles (customer, bot, agent), handoff events, conversation tags and the outcome.
| Field | What it holds |
|---|---|
conversation_id | Pseudonymous ID |
messages[] | Role (customer, bot, agent), text and timestamp |
handoff_at | When a bot passed the conversation to a person |
tags | Topic and intent tags |
outcome | Resolved, escalated, converted or abandoned |
csat | Rating, where collected |
{
"conversation_id": "C-90117",
"messages": [
{"role": "customer", "at": "2025-06-11T18:02:05Z", "text": "my order hasn't shipped"},
{"role": "bot", "at": "2025-06-11T18:02:07Z", "text": "I can help with that. What's your order number?"},
{"role": "customer", "at": "2025-06-11T18:02:31Z", "text": "[ORDER_ID] and I need it by friday"},
{"role": "agent", "at": "2025-06-11T18:04:10Z", "text": "Thanks for waiting. I've upgraded you to express..."}
],
"handoff_at": "2025-06-11T18:02:40Z",
"tags": ["shipping_delay"],
"outcome": "resolved"
}How AI labs use it
- Conversational agents
- Train on the short, real-time turns typical of chat, which differ from email tickets.
- Escalation decisions
- Handoff events mark the moment a bot passed a conversation to a person.
- Bot evaluation
- Compare a model’s replies with what human agents wrote after a handoff.
Typical preparation requirements
Agreed with the supplier before any work begins. Typical requirements include:
- Personal details redacted, including card numbers and credentials customers paste into chat
- Bot messages labeled so they can be filtered or kept
- Scope, permitted use and de-identification requirements agreed before any work begins
Every dataset has a documented owner and confirmed licensing rights. See data governance on sourcex.si.
What makes a strong package
- Handoff events recorded
- Outcome or conversion tags
- Both bot and human turns kept and labeled
Compared with public datasets
Public sets such as Customer Support on Twitter and Bitext Customer Support LLM Chatbot Training Dataset are useful references, but limited as enterprise training data. The customer support category page compares them with licensed data.
Who typically holds it
- E-commerce brands
- Subscription businesses
- SaaS companies
- Travel companies
- Fintech companies
Know a company like this?
Introduce the company to SourceX. If its data deal closes, you can earn up to $100,000 in referral fees, paid after the buyer accepts the data and SourceX receives payment.
Refer a companyQuestions
Can bot messages be separated from human ones?
Yes. Every message carries a role, so bot turns can be filtered out or kept as labeled examples.
Are pre-sales chats included?
Where the company records them. Sales and support conversations can be scoped separately.
Need this data for a model?
Describe what you need: domain, volume, history, format and licensing terms. SourceX looks for companies that hold it and can license it.