Effective customer-support chatbot scripts identify the bot, explain what it can do, and give customers a clear next step. Use the examples below as templates—not promises: replace bracketed text with your current policies and the information your systems can actually retrieve. Include a route to human help and a fallback for cases the bot cannot resolve.
What a useful customer-support chatbot script needs
A script is more than a friendly sentence. It should guide a customer through one task at a time, make the bot’s limits visible, and keep the conversation moving when the request is unclear or needs a person.
- Identify the bot: Do not imply that a human is replying. Salesforce’s guidance says, “Never leading customers to believe they’re chatting with a human if they’re really chatting with a bot.” Salesforce Help’s transparency guidance makes bot identification an explicit design principle.
- Set a bounded expectation: Name the tasks the bot can handle instead of suggesting it can solve anything.
- Ask for one useful input at a time: For example, request an order number before attempting an order lookup.
- Offer a recovery route: If the bot does not understand, ask a brief clarifying question or show supported options. Make human assistance accessible.
- Close the loop: Confirm the next step or invite the customer to raise another issue before ending the conversation.
These principles align with Salesforce’s conversation-design checklist, which covers greetings, bot identity, capability-setting, turn-taking, follow-up paths, and closure. The scripts here are original templates based on that guidance, not tested scripts or quotations.
Customer-support chatbot script examples
1. Welcome and task menu
Bot: “Hi, I’m [Company]’s virtual support assistant. I can help with order status, returns, or connecting you with our support team. What would you like to do?”
Do these 3 things before closing this tab:
1Repair Windows errors before they cause bigger problems2Fix the driver behind crashes, sound loss and screen glitches3Clear out junk files and repair common Windows errors#1 Best Overall
A good welcome names the bot, keeps its capability claims specific, and offers a clear first move. If the bot supports buttons or suggested replies, use short options such as “Check an order,” “Start a return,” and “Talk to support.” Keep a text-entry option for requests that do not fit those choices.
2. Order-status lookup
Bot: “I can check that. Please enter your order number.”
Customer: “[Order number]”
Bot: “Thanks. Your order is [status retrieved from the order system]. Would you like help with anything else?”
Only report a status that the order system actually returned. If the lookup fails, do not guess, infer a delivery date, or present stale information as current; use the unclear-request or human-help path instead. Salesforce uses order status as an example bot task and lists factual reliability as a response-quality consideration in its agent testing guidance.
3. Return request
Bot: “I can help you check the return steps. What’s your order number?”
Customer: “[Order number]”
Bot: “I found [item]. Based on our current return policy, [accurate next step]. Would you like me to explain the next step or connect you with support?”
Outdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware matchPC Slower Than It Used to Be?
A free scan shows the junk files, broken settings and background clutter dragging Windows down - then fixes them in one click.Free scan · Windows 10 & 11Use the actual policy and eligibility details for the customer’s order. There is no universal return window or eligibility rule to insert here; use “[return window]” only if the business’s current policy establishes one, and state relevant conditions plainly.
4. Clarify an unclear request
Bot: “I’m not sure I understood. Are you asking about an order, a return, or something else?”
Customer: “[Clarification]”
Bot: “Thanks for clarifying. I can help with [supported task], or connect you with a person.”
Keep the clarification short and relevant to the bot’s supported tasks. If the customer remains unclear or the request falls outside those tasks, offer a person or a return to the menu rather than repeating the same question indefinitely. Salesforce’s dialog guidance describes recovery paths for confusion and agent availability.
5. Transfer to a person
Bot: “I can connect you with a support specialist. I’ll pass along what you’ve shared so you don’t have to start over. Would you like me to transfer you now?”
Customer: “Yes.”
Bot: “All right. I’m connecting you now.”
Promise a context handoff only if the support system can pass the conversation details to the agent. IBM recommends carrying relevant context into escalation so customers do not need to repeat the issue; see IBM’s customer-service AI agent guidance. If the system cannot transfer context, say what information the customer may need to provide again.
6. No agent available
Bot: “Our support team isn’t available right now. You can [available next step, such as leave a message or use the support form], or try again [support hours or next available time]. Would you like to return to the self-service options?”
Rank #3
Replace every bracket with a real channel, schedule, or action. Do not claim a response time unless the business can support it. Acknowledge that live help is unavailable while still giving the customer a useful path forward.
7. Close and invite another task
Bot: “You’re welcome. Is there anything else I can help with?”
Customer: “No, that’s all.”
Bot: “Thanks for contacting [Company]. Goodbye.”
Offer the “anything else” question after a task is complete, but let the customer end the conversation without an unnecessary extra loop. Salesforce’s design and dialog guidance both include closure and a follow-up path.
How to write a script from scratch
- Choose one customer task. Start with a discrete need such as checking an order, understanding return steps, or reaching support. Avoid a single catch-all response that tries to cover unrelated tasks.
- Define what the bot can actually do. Identify the system lookup, policy information, or transfer capability the flow depends on. If a capability is unavailable, do not write the script as though it exists.
- Map the turns. For each step, specify what the bot asks, what customer input it needs, and what happens after a valid answer, an incomplete answer, or a failed lookup.
- Write the recovery and handoff paths. Include a concise no-match response, a useful clarification when possible, and an accessible route to a person. Specify what happens when no agent is available.
- Use plain, consistent language. Keep each message focused on one conversational turn. Offer concrete choices when they make the next step easier, while allowing text where a choice menu would be too limiting.
- Test varied messages and revise. Try the expected wording, synonyms, incomplete requests, vague messages, and unrelated questions. Check whether replies are factual, clear, respectful, context-aware, and helpful in moving the customer forward.
- Review after changes. Update scripts when policies, support hours, product capabilities, or knowledge sources change. Zendesk recommends testing and updating guidelines; IBM advises maintaining knowledge sources and monitoring feedback.
Keep the bot’s required behavior separate from its voice. Zendesk’s documentation team writes: “Clear, specific rules, grouped by topic and supported with examples, help improve results for greetings, closings, channel-specific language, and message length.” See Zendesk’s guidance on writing AI agent instructions. For a more formal test goal, Salesforce Trailhead says: “The goal is to check if the agent’s instructions–the prompts and guidelines you wrote–are effective at generating appropriate responses to user utterances.”
Common script-writing mistakes to avoid
- Passing the bot off as a person: Identify it in the opening instead of using language that creates a false impression.
- Claiming more than the bot can do: Describe supported tasks, not an unlimited ability to help.
- Inventing policy details: Replace placeholders such as “[return window]” or “[accurate next step]” with the business’s current rules; do not make up deadlines or eligibility promises.
- Guessing when data is missing: A failed order lookup is a reason to clarify or escalate, not to fabricate a status.
- Creating an endless fallback loop: After an unsuccessful clarification, show a menu or offer human help.
- Ending at a dead end when agents are offline: State the real next available channel or timing and leave a self-service route.
- Discarding conversation context on transfer: Pass relevant details to the agent where the system permits, or tell the customer what they will need to repeat.
When a script should hand off to a person
Automation should stop when it cannot give a reliable answer or take the action the customer needs. The exact threshold depends on the task and the system, but a sound flow offers escalation when an answer is uncertain, a lookup fails, the request is outside the bot’s supported scope, or repeated clarification does not resolve the issue. Make the handoff explicit: tell the customer what is happening, whether context will be carried over, and what to do if a representative is unavailable.
There is no evidence here that one vendor’s bot performs better than another. The relevant design criteria are task coverage and clear next steps, access to a human, preservation of context during handoff, clarity about bot identity and limits, and how easily scripts can be tested and updated.
The Tool Desk
Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →Frequently Asked Questions
What should a customer-service chatbot say first?
It should identify itself as a bot, name a few tasks it supports, and offer a clear next action, such as choosing order help, returns, or a support representative.
How should a chatbot respond when it does not understand?
It should acknowledge the misunderstanding, ask a short relevant question or offer supported options, and provide a way to reach a person if the issue remains unresolved.
Should a chatbot tell customers it is a bot?
Yes. The greeting should make clear that the customer is interacting with an automated assistant, not suggest a human is replying.
What should a chatbot say when no agent is available?
It should state that the team is unavailable, give a real alternative or the actual support schedule, and keep an available self-service route open.
How often should chatbot scripts be reviewed?
Review them when policies, support hours, system capabilities, or the information the bot relies on changes, and test them regularly against varied customer wording.
Frequently Asked Questions
What should a customer-service chatbot say first?
It should identify itself as a bot, name a few tasks it supports, and offer a clear next action, such as choosing order help, returns, or a support representative.
How should a chatbot respond when it does not understand?
It should acknowledge the misunderstanding, ask a short relevant question or offer supported options, and provide a way to reach a person if the issue remains unresolved.
Should a chatbot tell customers it is a bot?
Yes. The greeting should make clear that the customer is interacting with an automated assistant, not suggest a human is replying.
Recommended Free Tools
What should a chatbot say when no agent is available?
It should state that the team is unavailable, give a real alternative or the actual support schedule, and keep an available self-service route open.
How often should chatbot scripts be reviewed?
Review them when policies, support hours, system capabilities, or the information the bot relies on changes, and test them regularly against varied customer wording.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
