Control when Yaplet hands off to a human

Updated May 22, 2026

Every way a conversation reaches a person

Sometimes the AI should stop answering and let one of your team take over. Every way that can happen in the chat widget is set in one place — Brand → AI agent, in the Human hand-off section, under the brand the agent belongs to.

  1. The visitor picks Request agent. It sits in the ⋮ menu at the top of the chat window and works at any time.
  2. The visitor taps the support button. A Talk to the support team button appears under the conversation once the visitor has sent the number of messages you set.
  3. The visitor asks Yaplet for a person. Yaplet recognizes phrases like "I want to talk to someone".
  4. Yaplet cannot answer. Nothing in the brand's knowledge answers the question, and you have chosen to connect the visitor to a person in that case.

Each of these can be switched off on its own. Yaplet no longer suggests a person in its own words: a visitor reaches your team only through one of these ways, or through one of the brand's workflows that hands over. And if you switch on Let Yaplet try first, Yaplet first checks whether it can solve the issue itself before anyone is connected — see Let Yaplet try first below.

The Human hand-off section

The section lists its settings in the order a visitor meets them. The first four sit together in a group called Ways to ask for a person:

  • “Request agent” in the chat menu — on, the visitor can ask for a person at any time from the ⋮ menu at the top of the chat. Off, the item disappears from the menu.
  • “Talk to support team” button — on, a button (Talk to the support team in the chat) appears under the conversation once the visitor has been chatting for a while; you set the number in Show it after the visitor has sent … messages in the last 5 minutes. The default is six. Lower it if your visitors expect to reach a person quickly; raise it if Yaplet is handling most questions well. Off, the button never appears.
  • Visitor asks Yaplet for a person — on, Yaplet connects a visitor who tells it they want a person; if nobody is online, the visitor can leave an email. Off, Yaplet never connects anyone itself: when a visitor asks for a person, it points them to Request agent in the ⋮ menu when that item is on, or on Facebook, Instagram and email to the contact details it finds in the brand's knowledge.
  • Let Yaplet try first — on, Yaplet makes sure it knows what the visitor needs before connecting and checks whether it can solve the issue itself: if it can, it answers instead; if not, it connects them to a person right away. Off unless you switch it on. See Let Yaplet try first below.
  • When Yaplet has no answer — a group of its own: what happens when Yaplet finds nothing to answer with. See the next section.

If you switch Visitor asks Yaplet for a person off and want to handle requests for a person your own way, the screen links straight to Brand → Workflows (Open this brand's workflows) and names the recipe: a workflow whose trigger is “User says”, with a request description such as The visitor wants to talk to the support team. The brand's switched-on workflows that hand over to a person are listed under the switches, after Workflows that hand over: — they reach a person whatever the switches say. A last line shows your team hours, with a link to Brand settings → Team hours.

When visitors cannot reach anyone

If every way above is off — the menu item, the button and Visitor asks Yaplet for a person, with the "no answer" choice on Politely decline — and no switched-on workflow hands over, a warning appears at the top of the section: "Visitors have no way to reach a person in this chat." You can save it that way, but then no chat visitor can ever reach your team. To fix it, switch one of the ways back on, or add a Request agent step to a workflow.

What Yaplet does when it cannot answer

In the Human hand-off section, the group When Yaplet has no answer decides what happens when nothing relevant turns up in this brand's knowledge. Two choices:

  • Politely decline — Yaplet tells the visitor it does not have information about that topic. It does not guess and does not make things up. If the visitor asked it to do something for them, such as cancel an order or issue a refund, it says plainly that it cannot do that itself. It mentions a way to reach your team only if it was given one, and never makes up a link or an address.
  • Connect to a person — the visitor is handed to your team right away, with no question asked first. If nobody from your team is available, Yaplet tells the visitor that nobody is available right now and that they can leave a message, and the visitor gets the card described under When nobody from your team is available below.

The second choice is grayed out until Visitor asks Yaplet for a person is switched on, with the hint "Switch on “Visitor asks Yaplet for a person” to use this."

Let Yaplet try first

Many requests for a person are things Yaplet can handle on the spot — a question its knowledge answers, or an order it can look up with one of its tools. Let Yaplet try first lets it do that before anyone is connected. It is off unless you switch it on: brands that never switched it on, and every new AI agent, start with it off. Off, Request agent in the ⋮ menu and the Talk to the support team button connect the visitor to a person in one press, and a visitor who writes that they want a person gets Yaplet's Yes/No question — see What the visitor sees below.

Switched on, when a visitor asks for a person in the chat widget — from the ⋮ menu, with the button, or by writing it — Yaplet first checks whether it can solve the issue itself. Nothing is written in the chat while it checks.

  • The visitor sees "Preparing your request…" and then "Just a moment…" for a few seconds. If the check takes too long, the visitor is simply handed over.
  • If the conversation has no concrete issue yet — for example, the visitor tapped the button straight away, only said hello, or just wrote that they want a person — a card titled Request a person asks: "Tell us briefly what you need, so we can help faster." The visitor sends their answer with the check mark or with Enter, and it shows in the chat as their own message. The visitor cannot skip the card: choosing Request agent or tapping the button again only puts the cursor in the card's box (the × still closes it). Yaplet then checks with what they wrote.
  • If Yaplet can settle the issue, it answers right away in the chat, as a normal answer, with no buttons for reaching a person next to it.
  • If Yaplet cannot settle it, the visitor is handed to a person at once, and your team gets what Yaplet found as a private note in the Inbox. That happens when the visitor wants a status changed or only your team can change it (a stuck or late order, a held payout, a pending review, a refund), or when the knowledge only says to contact support. If Yaplet's answer then finds nothing after all, it hands over at once too. When the visitor wrote that they want a person and Yaplet judged that a person is needed, it keeps its Yes/No question — see What the visitor sees below. If nobody from your team is available, the visitor gets the card for leaving your team a message instead — see When nobody from your team is available below.
  • If one of the brand's switched-on workflows hands over to a person and fits the visitor's issue, the hand-off happens at once.
  • Asking for a person again after one of Yaplet's answers always reaches a person. If Yaplet answered the visitor within the last 30 minutes and they have written nothing new since, a request for a person — from the ⋮ menu, with the button or by writing it — goes straight to your team, with no card and no check.
  • Also straight through: a visitor Yaplet can't answer right now — your plan's AI allowance is used up with no wallet balance, their AI is paused, or they reached the daily answer limit.

Why switch it on: it spares your team the conversations Yaplet can close by itself, and a visitor whose issue Yaplet can't settle still reaches a person right away. Each check is billed like an AI answer and counts toward the visitor's daily answer limit, and so does the answer Yaplet then gives.

To use it, switch on Let Yaplet try first in the Human hand-off section. Switch it off again and requests for a person go straight to your team. Facebook Messenger, Instagram and email are not affected either way: there a request for a person is handed over directly, as before.

When nobody from your team is available

A team member counts as available when their status is Online, they are not disabled in your workspace, and they have access to this brand. If nobody is available when a visitor asks for a person, the conversation is still queued in your inbox as unassigned and waits there — nothing is lost, and your workspace's owner is alerted. Instead of telling the visitor that someone is coming, the website chat shows a card titled The team is away right now. If you have set team hours and the request comes outside them, the card adds when your team is usually back, for example "Usually back Monday 09:00".

  • The card says "Write everything you need below — it goes straight to the team, and they'll reply in this chat." The chat itself is the message: the box to type in stays open under the card, and everything the visitor writes waits there for your team.
  • Leaving an email address is optional: Leave your email so they can also reach you by email, with a Save button. The field is already filled in when the visitor's address is known. A visitor who closes the card without one reads "You'll see the reply here next time you visit."
  • Or keep chatting with the AI appears only when Yaplet can answer this visitor right now.

The same card also appears when someone was available but nobody from your team has replied or taken the conversation 5 minutes after the request. Its title is then The team hasn't been able to reply yet, and your workspace's owner is alerted at the same moment. The 5 minutes are the same for every brand. A brand with a switched-on workflow that starts on the Nobody picked up trigger keeps that workflow instead of this card.

Visitors are never promised a reply time, and there is no setting for one.

What the visitor sees

When a request goes ahead to a person — Yaplet could not settle the issue, the visitor asked again after one of its answers, or Let Yaplet try first is off:

  1. From the ⋮ menu or the Talk to the support team button, the visitor is not asked anything more. When the visitor wrote that they want a person and Yaplet can't solve their issue, Yaplet asks first — Yes, connect me, or No, keep chatting to stay with Yaplet. If nobody from your team is available, Yaplet instead asks briefly whether they would like to leave a message for your team, and the buttons are Yes, leave a message and No, keep chatting; Yes shows the card. Typing works too: a visitor who writes "yes" instead of tapping is connected just the same. The question stays in the conversation, so your team sees it too. After one of Yaplet's answers, a visitor who writes that they still want a person is not asked again. Every button the visitor taps shows in the conversation as their own message, so your team sees what they chose.
  2. Yaplet confirms the handoff. After a tap the chat says "A team member will join this chat." — a fixed line, said in the visitor's language; after a typed yes, Yaplet says it in its own words. When nobody from your team is available, that line is left out and the visitor gets the card straight away. It is Yaplet's own line, so there is nothing here for you to write.
  3. The conversation appears in your inbox as unassigned. If Yaplet checked the request first, it leaves your team a private AI summary note in the conversation, right under the request: one line on what the visitor needs, plus "The assistant found: …" when Yaplet found something that may help. When the visitor's need is already in their own words — written into the Request a person card, or the question Yaplet had just answered — the note holds only what Yaplet found. The note stays in the conversation's history.
  4. Someone picks it up and carries on in the same thread — the whole Yaplet conversation is visible to them.

On Facebook, Instagram and email conversations, Yaplet skips the Yes/No confirmation and connects the visitor directly. Facebook and Instagram have their own tappable controls instead: a persistent chat menu with Talk to a human (and Talk to the AI bot to switch back — see Going back to the AI below), plus a Talk to a human quick reply that appears on Yaplet's answers once the same message threshold is reached. Facebook and Instagram cannot show cards, so when nobody from your team is available the visitor there gets a short offline notice instead, asking them to leave their message; with team hours set, it also says when your team is usually back. If nobody replies within 5 minutes, nothing more is sent to the visitor there, but your workspace's owner is still alerted. On email conversations every AI reply carries a "Talk to a human" button the customer can use at any time. No separate email confirms the handoff: the customer's next email is your colleague's reply.

Going back to the AI

While nobody from your team is actively answering, the visitor can return to Yaplet with the Talk to AI Bot chip under the conversation or AI Mode in the ⋮ menu. Both appear:

  • while the visitor waits and nobody from your team has joined yet (on the card, the same choice reads Or keep chatting with the AI);
  • when a team member has the conversation but has not replied for 5 minutes since the visitor's last message. Switching back then takes the conversation off that person, and your team sees a private line in the thread: "The visitor went back to the AI".

They never appear when Yaplet can't answer this visitor: Yaplet is not answering on the website chat, the AI is paused for them, or they have reached the daily answer limit. When the visitor switches back, the chat shows "Agent request canceled". Your team has no button for handing a conversation back to Yaplet.

On Facebook and Instagram, the menu's Talk to the AI bot follows the same rule, including after 5 minutes of silence from the team member who has the conversation. If Yaplet can't answer right now, the visitor reads "The AI assistant can't answer right now — our team will reply to you right here."

When the team member handling a chat is offline

A conversation normally stays with the team member who has it. But if the visitor writes while that person is not available — set to Offline, disabled, or without access to the brand — and they have not written in the conversation for 5 minutes, the visitor is not left waiting for someone who is gone. This works on the website chat, Messenger and Instagram:

  • If Yaplet can answer the visitor, it takes the conversation and answers that message. Your team sees a private line such as "Anna was offline — the AI took over".
  • If Yaplet can't, the conversation goes back to your team as a new request for a person — with "A team member will join this chat." or the card, the usual alerts and the 5-minute card. The private line then reads "Anna was offline — the chat is waiting for the team".

A team member who wrote in the conversation less than 5 minutes ago keeps it whatever their status says, because people often answer without switching to Online. Email conversations always stay with their team member. Visitors never see these private lines.

When someone hits the daily answer limit

There is one more, automatic way a conversation reaches a person. Every website visitor, e-mail address and Messenger or Instagram user gets a fixed number of Yaplet answers per 24 hours (30, 10 and 30 by default — set on Brand → Brand settings → Usage limits). At the limit Yaplet sends one short closing message, pauses for that person for a day and hands the conversation to your team straight away, as an ordinary request for a person, and your team is notified once. A private "AI paused" line in the conversation tells your team why. Details: Yaplet usage and credits.

Phone calls work differently

On a phone call there is no handing the caller over to a person, so all of the above collapses to a plain refusal when the agent finds nothing: it tells the caller it does not have that information. That is true for every phone agent, whether or not the brand has a chat AI, and there is no field in the dashboard to change it.

Who can pick the conversation up

Access is decided per brand. A teammate only sees this brand's conversations if they have been granted it. You do that from the avatar at the bottom-left of the dashboard → Manage team, on that person's row under Brand access. Owners and administrators always have access to every brand.

Next step

Once handoff is set up, check how often Yaplet actually escalates: Find knowledge gaps Yaplet couldn't answer. A high escalation rate almost always means missing knowledge rather than a bad setting.

Did this article answer your question?