Why the AI agent didn't reply

When the AI agent stays quiet, the conversation shows why. Open it, read the activity lines between the messages, then find your case below.

What the activity lines tell you#

• No line about the AI agent: it never picked the conversation up. Go to "The AI agent never joined the conversation".
• "[AI agent name] was automatically assigned to this conversation" and nothing after it: the AI agent has the conversation but hasn't sent anything. Go to "The AI agent joined but sent nothing".
• "[AI agent name] was removed from the conversation: no matching guidance detected": go to "Restrict AI responses to enabled guidances only".
• "[AI agent name] was removed after [teammate] responded": a teammate replied, so the AI agent stepped out. See "The AI agent joined, then left".
• "The conversation was handed over to a support agent by [AI agent name]", or "[AI agent name] was removed from the conversation", with a See reason link: the AI agent handed the conversation over. Hover over See reason to read why. If the reason refers to your guidance, check that guidance for an instruction to hand over.

The AI agent never joined the conversation#

The AI agent joins a conversation at the moment the conversation is created. Check these in order:

1. The AI agent is on and the inbox is selected. Go to AI agent → Settings. The page is shown to administrators. Enable AI agent must be on and the inbox must be in the list below it. Add it with Select inboxes, then click Update settings. Nothing changes until you click that button.
2. The conversation is new. Conversations that existed before you switched the AI agent on, or before you selected the inbox, are not picked up. That stays true when the customer writes again. To put the AI agent on one of them, open the conversation and choose the AI agent as assignee in the right sidebar. It answers the customer's next message, as long as Enable AI agent is on.
3. The inbox type can be selected. On this page you can select email, chat widget, Facebook and Instagram DM, WhatsApp and Judge.me inboxes. Other inbox types can't be selected here. On Facebook, Instagram and TikTok comments the AI agent doesn't post replies, it only prepares drafts for your team. See Facebook and Instagram comments: find, reply and moderate.
4. The AI agent isn't paused for chat. If Pause AI when agents are online is switched on for a chat inbox, new chats skip the AI agent while at least one agent is online.
5. You have AI agent messages left. The Free plan includes 300 AI agent messages per month and the AI agent stops there. On Plus and Scale plans it continues at $0.075 per extra AI agent message until your spending cap is reached. With a cap of $0 it stops at the plan limit. The spending cap is in Settings → Usage & billing. When the allowance is used up, the AI agent stops joining new conversations and no note is added to them.
6. The sender isn't blocked. Conversations from blocked senders are skipped.

The AI agent joined, then left#

• A teammate replied. A reply from a person takes the AI agent off the conversation. The one exception is while an AI needs your input card is waiting for your team. Then the AI agent stays on the conversation.
• The conversation was assigned to a teammate, by hand or by an automation rule with the Assign to agent action. A conversation has either the AI agent or a person as assignee.
• An auto-label with "Unassign AI agent" matched the first message. The default labels Spam, Marketing & Sales and Auto-responder have it ticked. Change it per label in Settings → Labels. See Labels and auto-labels.
• No enabled guidance matched. See "Restrict AI responses to enabled guidances only" below.
• Someone other than the customer wrote into an email thread, for example an address in CC. The activity line says the message came from someone other than the contact.

Once the AI agent is off a conversation it doesn't come back by itself. Later messages from the customer wait for your team. To hand the conversation back, choose the AI agent as assignee.

The AI agent joined but sent nothing#

• It's an email and only a few minutes have passed. On email the AI agent waits up to 3 minutes before it starts on a reply. Chat, DMs and WhatsApp are answered right away. There is no setting to change the delay.
• "Require approval before AI actions" is on. The reply waits in AI agent → Needs approval until someone approves it.
• The AI agent asked your team a question. With Ask human agent on, the conversation waits in AI agent → Needs approval until someone answers the AI needs your input card, clicks Handover on it, or the Ask human timeout runs out.
• The email is automatic. Out-of-office replies and other auto-replies are not answered. Neither are emails sent from the inbox's own address, such as store notifications that use your support address as the sender. Unless an auto-label takes the AI agent off, these conversations stay under AI agent → Open. Resolve them or assign them to a teammate.
• An automation rule resolves the conversation. If an active rule with the Resolve conversation action matches the message, the AI agent doesn't reply.

Restrict AI responses to enabled guidances only#

This setting in AI agent → Settings makes the AI agent answer only when the customer's message matches a guidance with the status Enabled.

• Where it applies: all email and social inboxes, including Judge.me. Chat and WhatsApp are never restricted. The Playground isn't either.
• What happens when nothing matches: the AI agent is removed and the line "[AI agent name] was removed from the conversation: no matching guidance detected" is added. The conversation goes to the agent chosen under Handover to a human agent → Assign the conversation to, or stays unassigned. If the inbox has auto-assignment on, an agent can be assigned at that moment.
• What the customer gets: nothing. The Fallback message is for handovers and isn't used here.
• Where the conversation shows: in your regular lists. It isn't listed under AI agent in the sidebar and isn't counted as an AI agent ticket in Reports → AI agent.
• Labels: auto-labels are applied before the AI agent decides, so a conversation can get a label and still get no reply.

The setting does nothing while Enable AI agent is off.

A conversation should have matched a guidance#

Open the guidance in AI agent → Guidance and check:

1. Status is Enabled. New guidance starts in Testing, which is active in the Playground only. That includes guidance an AI assistant created for you.
2. "Apply to channels" includes the channel. A conversation in an email inbox needs Email, and so does a Judge.me review. Facebook and Instagram messages need DMs. Chats need Chat. Guidance with no channel ticked applies nowhere, and guidance created by an AI assistant has no channels unless the assistant set them.
3. Your changes are published. After you edit an enabled guidance, customers keep getting the published version until you publish the draft.
4. The trigger describes the customer's situation. Matching reads the guidance name and the trigger. It doesn't read the instructions. One guidance is picked per message.
5. No guidance in Testing has a similar trigger. A guidance in Testing can be picked for a live conversation. It isn't applied there, and with this setting on the AI agent leaves the conversation. Disable that guidance or enable it.

To see what matched, look at the customer's message in the conversation: the matched guidance name is shown on it. In AI agent → Guidance, the Conversations column counts the unique conversations that matched each guidance in the last 7 days.

For help with triggers, see Writing guidance.

Let the AI agent answer everything#

Switch the setting off and click Update settings. The AI agent then answers every new conversation in the selected inboxes from your knowledge and your enabled guidance. Conversations it already left stay with your team.

What isn't possible#

• Restricting only some inboxes. The setting covers all email and social inboxes at once. Instead, leave it off and use an auto-label with Unassign AI agent for what the AI agent should skip (auto-labels are on Plus and up), or write guidance that hands those topics over.
• Restricting chat or WhatsApp. Instead, remove that inbox from the AI agent, or switch on Pause AI when agents are online for a chat inbox.
• Sending the customer a message when nothing matches. Instead, switch the setting off and write guidance that tells the AI agent to hand over the topics you want your team to answer. On a handover the customer gets a message based on your Fallback message.

The Playground replied but a live conversation got nothing#

The Playground doesn't mirror your live setup:

• It works while Enable AI agent is off.
• Guidance in Testing is active there. Live conversations need Enabled.
• Restrict AI responses to enabled guidances only doesn't apply there, so the Playground answers from knowledge when no guidance matches.
• It uses the channel you pick in its channel selector, and a guidance is used only if it lists that channel.

See Testing the AI agent.

The chat on your store doesn't answer#

1. Check that the chat widget shows on your storefront. See How to add the chat widget to your Shopify store.
2. Select the chat inbox in AI agent → Settings, switch on Enable AI agent and click Update settings.
3. Look for the chats under AI agent → Open in the Views section of the sidebar. While the AI agent handles a chat, it isn't listed in the chat inbox.

Where these conversations are listed#

Conversations the AI agent is handling are under AI agent in the Views section of the sidebar, not in All conversations, your inboxes or your views. See Understanding conversation views.

Still stuck?#

Contact support from the chat inside the app. Send the link to the conversation and tell us which activity lines you see in it.