Chatbot vs. live chat: when should AI hand over to a human?

Chatbot vs. live chat sounds like a choice between two tools. In practice, the best support setups use both: an AI chatbot takes the first answer for repeat questions, and a person takes over the moment the conversation needs judgment, empathy or access the bot doesn't have. The real question isn't which one – it's where the line between them sits.

This guide explains what each does well, which questions belong on which side, and how to design a handover that customers don't notice as a break.

What's the difference?

	Live chat	AI chatbot
Who answers	A person on your team	Software, usually a language model reading your content
Availability	When someone is online	Around the clock
Speed	Depends on how busy you are	Seconds
Good at	Exceptions, complaints, advice, sales conversations	Repeat questions with a documented answer
Weak at	Being there at 3 a.m., answering the same question for the 40th time	Anything not in its sources, judgment calls, goodwill decisions
Cost	Your team's time	Software plus AI usage – see what an AI chatbot costs

Older rule-based chatbots worked with buttons and decision trees. Modern AI agents read your help articles and answer in free text. That makes them much more useful – and it makes the handover more important, because a fluent answer can still be a wrong one.

Which questions should the chatbot answer?

A good rule: if the answer is written down and is the same for everyone, the bot can take it. Typical examples:

• Opening hours, delivery times, shipping countries
• Return and exchange policy
• Payment methods, invoices, VAT questions with a standard answer
• "How do I…?" questions covered by a help article
• Sizes, specs, compatibility – when they're documented
• Where to find something on your website

These questions make up a large part of the daily inbox for most small teams, and nobody enjoys typing the same answer again. That's where a bot saves real time.

When must the AI hand over to a human?

The bot should step aside in these situations – and you should configure it to do so, not hope it will:

1. The customer asks for a person

Always honour it, immediately. Nothing frustrates people more than a bot that keeps answering after "I want to talk to a human". Treat words like "human", "agent", "person", "call me" as a direct handover trigger.

2. The bot isn't confident

If the answer isn't clearly covered by your sources, a handover beats a guess. A wrong answer about a refund or a delivery date creates a second, angrier conversation.

3. The conversation goes in circles

If the customer has asked twice and the bot has answered twice, a third bot answer rarely helps. Limit the number of AI turns per conversation.

4. Money, complaints and exceptions

Refunds outside the policy, damaged goods, a missed deadline, a wholesale enquiry – these need someone who can decide. The bot can collect details, but a person should answer.

5. Emotion

An upset customer wants to feel heard. Even a perfect bot answer reads cold when someone is angry.

6. Personal data and account actions

Anything that requires looking into an order, changing an address or verifying identity belongs with your team, unless your bot is explicitly connected to those systems.

How to design a good handover

A handover fails when the customer has to start over. A good one has four parts:

1. Tell the customer what's happening. "I'm passing you to the team – they'll pick up right here." Set expectations if nobody is online right now.
2. Hand over the context. The person taking over should see the whole conversation and ideally a short summary, so they don't ask "How can I help?" again.
3. Stay in the same window. The customer shouldn't be sent to an email form or another channel.
4. Keep the bot out once a human has answered. Two voices answering at once is confusing.

When nobody is online, the handover still needs an answer: collect an email address so you can reply later, and say when someone will be back. Our post on away and offline messages has templates.

How asktrix handles the handover

In asktrix, the AI agent Trix and your team share the same conversation in the shared inbox. The handover options are settings, not guesswork:

• On request: when a visitor asks for a person, Trix hands over. You can add your own trigger keywords.
• When unsure: if Trix finds no confident answer in your sources, it hands over instead of guessing.
• After a number of turns: you set how many times Trix may answer before a person takes over.
• With a summary: your team gets the conversation with a short summary, so nobody asks twice.
• Stays out afterwards: once a teammate replies, Trix doesn't answer in that conversation again.

Every answer shows the help articles it came from, and you can test changes in a playground before switching them on. On the Team plan, learning notes capture what your team answered after a handover; once you approve a note, Trix knows the answer next time. Quiet hours let Trix answer on its own outside working hours and give your team a morning summary of what happened and where they're needed.

Setting it up: a practical order

1. Start with live chat only. Watch which questions come in for a week or two. The live chat widget is free on the Free plan.
2. Write the top answers down. Turn the ten most frequent questions into short help articles – one question per article. A help center does double duty: customers read it, and the bot answers from it.
3. Switch on the bot with strict handover. Hand over on request and when unsure, and keep the turn limit low at first.
4. Read the handovers. Each one shows a gap: a missing article, an unclear policy, or a question that really needs a person.
5. Loosen gradually. As your articles improve, let the bot take more – but keep "on request" handover forever.

Signs your balance is off

Too much bot: customers repeat "human" several times, ratings drop after AI conversations, or your team keeps correcting bot answers.

Too little bot: your team answers the same five questions every day, evenings and weekends pile up, or customers wait hours for answers that are on your website.

FAQ

Will an AI chatbot replace live chat? For repeat questions, it can take most of the load. For complaints, advice and exceptions, customers still want a person – and a chatbot without a human behind it loses trust quickly.

Should customers know they're talking to a bot? Yes. Label AI answers clearly. People accept a bot for quick answers; they don't accept being misled. In asktrix, AI messages carry their own sender name, which you can set per language.

What if nobody is online when the bot hands over? Ask for an email address and say when the team is back. The conversation waits in the inbox and your reply reaches the customer by email if they've left.

Want to try the combination? Start free with live chat, then test Trix on your own help articles with a 14-day trial of the Start or Team plan.