It does not know
Retrieval found nothing relevant. The correct behaviour is to say so and offer a person — not to reason its way to a plausible-sounding guess.
Human handover
The measure of a good assistant is how cleanly it gives up
Every support bot has a boundary. The ones people trust are the ones where the boundary is obvious, deliberate and leads somewhere — not the ones that keep answering past the point of usefulness.
Four reasons to stop answering
Only one of them is a failure. The other three are the assistant working correctly.
It should not decide
Refunds outside policy, contractual questions, negotiations, complaints. The assistant can explain the rule; it should not be the one making the exception.
The person is upset
Frustration is a handover signal in its own right. A visitor on their third rephrasing does not want a fourth answer, they want a human.
They asked for a human
The single most important one. If someone types “I want to speak to someone”, nothing the assistant says next should be an attempt to keep them.
There is a right way and a wrong way to stop
Most of the damage attributed to chatbots is not wrong answers. It is dead ends.
A dead end
- A dead end: “I cannot help with that.” and nothing else
- A loop back into the same failing answer
- A promise of a reply with no way to reach anyone
- A form that discards everything the visitor already typed
A handover
- A plain sentence naming the limit, without apologising three times
- A real route: an email address, a phone number, or details taken for a callback
- The transcript preserved, so nobody has to explain twice
- An honest expectation of when someone will reply
A handover message worth copying
That one needs a person — it is outside what I can decide. If you leave your name and email I will pass this whole conversation to the team and someone will reply by tomorrow morning. If it is urgent, you can call us on 01234 567890 between 9 and 6.
Four things in four sentences: it names the limit, it does not grovel, it gives a real route, and it sets an expectation. Write yours in the same shape and the overwhelming majority of handovers stop being a bad experience.
What happens after it hands over
The conversation is stored in full
Whoever picks it up reads what was actually asked, in the visitor’s own words, including the answers the assistant already gave. No re-qualifying, no “can you explain again”.
It is readable from your phone
The Android and iOS companion app puts transcripts and leads in your pocket, which is where most handovers are actually picked up — in the evening, away from a desk.
It becomes a content signal
Handovers caused by “it does not know” are your training backlog. Handovers caused by “it should not decide” are correct and should not be optimised away.
One thing to be clear about
Handover routes the conversation to your team with the details captured — it is not a real-time agent console inside the widget, and there is no live queue for agents to pick up chats in the moment. For most small and mid-sized teams that is the right trade, because a live queue nobody is staffing is worse than an honest callback. If a staffed live-chat desk is a requirement, it is better to know that now than after you have trained an assistant.
Handover questions
Is handover on by default?
Yes, and you can write the message it uses. The default connects the visitor to a human rather than leaving them stranded; most people rewrite it to name a real contact route for their business.
Does it hand over to a live agent inside the chat?
Handover collects the visitor’s details and routes the conversation to your team rather than opening a real-time agent seat inside the widget. If a live-agent console is a hard requirement for you, that is worth knowing before you evaluate further.
How do I make it hand over more often?
Two levers. The system prompt is yours, so you can name the categories it must not attempt — refunds, complaints, anything contractual. And leaving gaps in the knowledge base unfilled is itself a way of keeping certain topics off the assistant.
How do I stop it handing over too often?
Usually the fix is content, not configuration. Read the transcripts where it gave up: nearly always there is a page that does not exist or a page that is ambiguous. Writing one paragraph often removes a whole category of handovers.
Can it detect frustration?
Modern models pick up on tone reasonably well, and you can instruct it explicitly in the system prompt to offer a human when someone seems frustrated or has asked the same thing twice. It is worth writing that instruction rather than assuming it.
What happens to the visitor’s details?
They are stored with the conversation so your team can follow up, subject to the same retention as other sessions — 7 days on the free plan, 30 days on paid plans. Sessions can be deleted individually or in bulk.
Write the boundary before you launch
Ten minutes on the system prompt and the handover message does more for trust than any amount of tuning afterwards.
Start free