A chatbot is software that responds to messages automatically. Simple bots follow scripted menus and keyword triggers; AI-powered bots generate answers from your information. Both can help, as long as they handle the right tasks and hand over to people smoothly. The Academy's glossary explains DM and community management.
What chatbots handle well
| Good for chatbots | Better for people |
|---|---|
| Opening hours and locations | Complaints |
| Order tracking links | Refund disputes |
| Frequently asked questions | Sensitive or emotional situations |
| Booking links | Complex product advice |
| Collecting details before handover | Anything involving judgement |
| Out-of-hours acknowledgements | Escalations and crises |
Designing a good bot
- Say it is a bot: be upfront that customers are talking to an automated assistant.
- Offer a human route: always make it easy to reach a person.
- Keep menus short: three or four options at most.
- Use your voice: friendly and clear, not robotic.
- Set expectations: explain when a person will reply.
- Test thoroughly: try every path before launch.
AI-powered bots
AI bots can answer a wider range of questions by drawing on your website, FAQs and policies. They need careful setup: restrict them to approved information, prevent them from making promises such as refunds or discounts and monitor answers regularly. AI bots can state incorrect information confidently, so start with a narrow scope.
Handover to people
The handover is where many bots fail. Pass the conversation history and any details collected, so customers do not repeat themselves. Notify staff quickly and set response time expectations. The Academy's guide to social media customer service covers human response.
Platform rules
Platforms set rules for automated messaging. Meta's messaging policies limit when businesses can send messages, often within a window after the customer's last message. WhatsApp Business includes business profile, catalogue, quick replies, labels, greeting/away messages — free for small businesses✓, and larger businesses use the WhatsApp Business Platform for automation. Use official tools and approved providers.
Data and privacy
Chatbots collect personal data such as names, order numbers and contact details. Explain how data is used, collect only what is needed and store it securely. Do not ask for card details or passwords through chat. The Academy's guide to UK GDPR for social media explains obligations.
Keyword automations
Many social teams use simple keyword automations: comment a word to receive a link by DM. These suit campaigns and lead generation, but they are not customer service bots. The Academy's guide to auto-reply DMs explains setup.
Tone
Bots should sound like your brand but not pretend to be human. Avoid jokes in situations where customers may be frustrated. Apologise simply when things go wrong and move quickly to a person.
Measuring success
Track resolution rate, how often customers ask for a person, response times, satisfaction ratings and repeat contacts. If many customers abandon the bot or ask for a human immediately, simplify the bot or reduce its scope. The Academy's guide to social media response time offers benchmarks.
Keeping content current
Bots give wrong answers when information changes, such as new opening hours, delivery times or prices. Assign someone to update bot content whenever policies change, and review conversations monthly.
Accessibility and inclusion
Write bot messages in plain language, avoid relying only on buttons or images and make sure the bot works with screen readers on each platform. Some customers find bots difficult to use, so the human route must be easy to find.
Starting small
The best first bot answers your five most common questions and offers a person for everything else. Review conversations after a month, add answers for questions that come up often and remove options nobody uses. Growing a bot gradually keeps quality high and lets you learn how customers actually use it.
Multilingual support
Some bots can answer in several languages. Test translations with native speakers before launch, and make sure the handover to people works for customers writing in other languages too.
Testing with real customers
Before a full launch, run the bot with a small share of conversations and read every exchange.
Escalation triggers
Set rules that pass conversations to a person automatically, such as words like complaint, refund, cancel or urgent, repeated questions or signs of frustration. Escalation triggers catch conversations the bot should not handle before they go wrong.
When not to use a chatbot
Businesses with low message volumes, highly personal services or customers who strongly prefer human contact may be better served by quick human replies and saved responses. A bot that handles three questions a week adds complexity without much benefit.
A worked example
An online shoe retailer receives hundreds of Instagram DMs weekly, mostly asking about delivery and returns. It sets up a bot that answers delivery and returns questions with links, collects order numbers for anything else and hands those conversations to staff with the details attached. Response times fall sharply, and staff spend their time on the conversations that need them.
Common mistakes
- Bots pretending to be human.
- No easy way to reach a person.
- AI bots making promises they cannot keep.
- Outdated information.
- Asking for sensitive data in chat.
Learning customer service on social
The Academy's community management module in the second level covers customer service, and the AI and automation at scale module in the leader level covers chatbots and automation design. The free assessment shows where you would start.
