Troubleshooting Conversation AI

​

Check 1: Bot status

1. Navigate to Settings > Conversation AI and select your bot.
2. Confirm the Bot Status is set to Auto-Pilot. If it is set to Suggestive, the bot will only suggest responses in the inbox — it will not send them automatically.
3. If the status is Off, toggle it to Auto-Pilot and save.

​

Check 2: Channel assignment

1. In your bot settings, check the Channels section.
2. Confirm that the channel the customer is messaging on (SMS, Facebook, Instagram, WhatsApp, Web Chat) is toggled on.
3. If the channel is not assigned, toggle it on and save.

​

Check 3: Contact bot status

1. Open the conversation in the conversations inbox.
2. Check the contact details panel for the Bot Status indicator.
3. If the status shows Stopped or Paused, the bot will not respond to this specific contact even if the bot is globally active.
4. Re-enable the bot for this contact manually or via a workflow.

​

Check 4: Primary bot assignment

1. Go to Settings > Conversation AI.
2. Verify which bot is marked as Primary. The primary bot handles all inbound conversations not assigned to another bot via workflow.
3. If no bot is set as primary, inbound messages on unassigned channels will go unanswered.

​

Check 5: Do Not Disturb

​

Check 6: Phone number or integration issues

• For SMS: Verify your phone number is active and properly provisioned in Settings > Phone Numbers.
• For Instagram/Facebook: Verify the integration is connected in Settings > Integrations. Reconnect if the token has expired.
• For WhatsApp: Verify the WhatsApp Business API connection is active.

​

Knowledge base issues

1. Check training status: Navigate to Settings > Knowledge Base and verify that your knowledge base shows a Trained status. If it shows Pending or Failed, the bot may not have access to the information. See knowledge base for training details.
2. Review content accuracy: Open your knowledge base and verify that the information is correct and up to date. Outdated content is a common source of wrong answers.
3. Check knowledge base assignment: Make sure the correct knowledge base is assigned to the bot. A bot can only use knowledge bases that are explicitly linked to it.
4. Improve content format: Structure your knowledge base content with clear headings, short paragraphs, and explicit Q&A pairs. The AI retrieves information more accurately when content is well-organised.

​

Prompt refinement

1. Add grounding instructions: Include explicit instructions in your prompt like “Only answer questions using information from your knowledge base. If you do not have the information, say so and offer to connect the customer with a team member.”
2. Be specific about boundaries: Tell the bot what topics it should and should not discuss. For example: “Do not provide pricing information. Instead, direct the customer to schedule a call.”
3. Include example responses: Adding 2-3 example question-and-answer pairs in your prompt helps the bot understand the expected response format and level of detail.
4. Test with real questions: After updating the prompt, test with the exact questions that were producing wrong answers.

​

Hallucination prevention

• Add a prompt instruction: “Never make up information. If you are unsure about something, clearly state that you do not have that information.”
• Keep knowledge base content factual and avoid ambiguous language.
• Regularly review conversation logs on the Conversation AI dashboard to catch inaccuracies early.

​

Max messages setting

1. Open your bot settings and look for the Max Messages configuration.
2. Set a reasonable limit (e.g., 10-15 messages per conversation). Once this limit is reached, the bot will stop responding.
3. If max messages is set to unlimited or a very high number, the bot will continue responding indefinitely.

​

Stop bot action

1. Navigate to your bot Bot Goals tab.
2. Add a Stop Bot action with clear trigger conditions (e.g., “Stop when the customer says goodbye”, “Stop after the appointment is booked”).
3. Test the stop conditions to make sure they trigger reliably.

​

Workflow-based stop

1. Create a workflow with an appropriate trigger (e.g., appointment booked, tag added).
2. Add a Disable Bot for Contact action.
3. This immediately stops the bot from sending further messages to that contact.

​

Calendar connection

1. Navigate to Settings > Calendars and verify that a calendar is connected and active.
2. Confirm the calendar is linked to the correct team member or resource.
3. If using Google Calendar or Outlook, check that the integration token has not expired.

​

Bot action configuration

1. In your bot Bot Goals tab, verify that a Book Appointment action is configured.
2. Check that the correct calendar is selected in the action settings.
3. Review the trigger conditions — the bot needs clear instructions on when to offer booking (e.g., “When the customer expresses interest in scheduling a meeting”).

​

Timezone issues

1. Verify the account timezone in Settings > Business Profile.
2. Check that the calendar availability is set in the correct timezone.
3. If the bot is showing wrong available times, the timezone mismatch is likely the cause.

​

Availability settings

1. Open the calendar that the bot is trying to book on.
2. Verify that availability hours are set and not blocked.
3. Check for conflicts — if all slots are full, the bot cannot offer any times.
4. Make sure the booking buffer and minimum scheduling notice settings are not overly restrictive.

​

Prompt instructions

When the customer wants to schedule an appointment, collect their
preferred date and time, then use the Book Appointment action.
If no times are available on their preferred date, suggest
alternative dates.

​

Expected delay

​

Knowledge base size

1. Break large documents into smaller, focused sections.
2. Remove outdated or irrelevant content.
3. Use clear headings and structured formatting to help the AI find information faster.

​

Prompt complexity

1. Keep your prompt focused and concise — aim for under 2000 words.
2. Remove redundant instructions.
3. Use clear, simple language in your prompt rather than complex conditional logic.

​

Concurrent load

​

Escalation triggers in the prompt

• Your prompt may have overly broad escalation triggers. Narrow them down. Instead of “escalate if the customer seems unhappy”, use “escalate only if the customer explicitly requests to speak with a human agent or if the customer expresses strong frustration after two or more attempts to resolve their issue.”
• Make sure the bot has enough information in its knowledge base to handle common questions without escalating.
• Add instructions like: “Try to resolve the issue using your knowledge base before escalating. Only escalate as a last resort.”

• Add explicit escalation triggers to your prompt: “If the customer mentions a legal issue, a safety concern, or requests a refund over $500, immediately escalate to a human agent.”
• Include a catch-all: “If you cannot confidently answer the question after checking your knowledge base, escalate to a human agent.”
• Make sure a human handover action is properly configured in your bot goals.

​

Monitoring escalation rates

• Under 10%: The bot is handling most conversations well (verify it is not missing cases that should escalate).
• 10-25%: Normal range for most businesses.
• Over 25%: The bot may need a better knowledge base or more detailed prompt instructions.

​

Prompt language specification

Always respond in English, regardless of what language the customer
writes in. If the customer writes in a language other than English,
respond in English and politely let them know that you can best
assist them in English.

Detect the language of the first message and respond in that same
language for the entire conversation. Supported languages: English,
Spanish, French. If the customer writes in an unsupported language,
respond in English.

​

Multi-bot language routing

• A prompt written entirely in the target language
• A language-specific knowledge base
• Tone and cultural nuances appropriate for that audience

​

Knowledge base language

​

Training status

1. Go to Settings > Knowledge Base and find the knowledge base assigned to your bot.
2. Check the Status column. It should show Trained.
3. If the status is Pending, wait for training to complete. If it shows Failed, delete and re-add the content.

​

Knowledge base assignment

1. Open your bot settings.
2. Verify the correct knowledge base is selected in the Knowledge Base section.
3. A bot will only reference knowledge bases that are explicitly assigned to it.

​

Content relevance

• Use clear, descriptive headings that match how customers ask questions.
• Include FAQ-style content with explicit question-and-answer pairs.
• Avoid jargon unless your customers use it too.
• Keep content concise — long, rambling documents make it harder for the AI to find the right information.

​

Format issues

• URLs: If you added a URL as a knowledge source, verify the page is publicly accessible and not behind a login wall.
• File uploads: Ensure uploaded files are in a supported format (PDF, DOCX, TXT) and the text is selectable (not scanned images).
• Content length: Very short knowledge base entries (a sentence or two) may not provide enough context. Aim for at least a paragraph per topic.

​

Grounding instructions in the prompt

Always base your answers on information from your knowledge base.
Do not use general knowledge to answer customer questions. If the
answer is not in your knowledge base, tell the customer that you
will find out and get back to them.