LINE official account + AI automatic reply: 2026 complete setting tutorial
LINE’s penetration rate in Taiwan exceeds 95%. For most small and medium-sized enterprises, the official LINE account is the main customer communication channel.
However, most merchants’ LINE automatic replies are still at the “keyword comparison” stage - customers type “Business Hours” and the business hours will be returned, and “Address” will be returned to the address. As long as the question method is slightly different, you have to rely on manual reply.
This tutorial will take you from basic automatic response to all the way upgrading to AI smart customer service, so that your LINE can reply to customers at any time.
3 levels of automatic replies from LINE official accounts
Level 1: Basic automatic response (free)
The automatic response function built into the LINE official account does not require any programming capabilities.
Things that can be done:
- Add friend welcome message
- Keyword automatic reply (up to 100 groups)
- Switch between different reply modes according to time period
Setting method:
- Log in LINE Official Account Manager
- Enter “Automatic response to messages” → “Keyword reply”
- Set keywords and corresponding reply content
Suitable: Businesses with less than 30 messages per day and simple and clear FAQ.
Restrictions: Only exact comparisons or inclusive comparisons can be made. When a customer says “Can I return the item?”, it can be triggered if the keyword you set is “return”; but if the customer says “I don’t want the item anymore,” it won’t be triggered.
Level 2: Graphic menu + automatic response (free)
Use graphic menus to guide customers to click instead of typing.
Method: Design a graphic menu and turn the 5-6 most common questions into buttons:
- 📦 Check order
- 🔄 Returns and exchanges
- 📍 Store information
- ⏰ Business hours
- 💬 Contact customer service
Each button triggers a corresponding auto-reply message. The advantage of this method is that customers don’t need to guess “what keywords to type”, they can just click directly.
The result of our actual measurement is that for accounts with graphic menus, the automatic reply processing rate increased from 35% to 58%. Because you guided customers to ask questions in a “way that the system can handle.”
Level 3: AI Chatbot serial connection (requires Messaging API)
This is true AI customer service. Connect the AI model through the LINE Messaging API so that your LINE account can “understand” natural language.
Things that can be done:
- Understand various questions (“I want to return”, “I don’t want it anymore” and “Can I exchange” can all be understood as return and exchange requirements)
- Multiple rounds of dialogue (remember what was discussed previously)
- Find the most relevant answers from the knowledge base based on customer questions
- Automatically transfer to manual if unable to answer
Level 3 complete setting tutorial
Preparation
- LINE official account (skip if you already have one)
- LINE Messaging API: Enable in LINE Developers Console
- AI platform: Botpress / Dify / self-built n8n workflow
- Knowledge Base: Your product FAQs, policy documents
Step 1: Enable Messaging API
LINE Official Account Manager → 設定 → Messaging API → 啟用
After enabling it you will get: -Channel ID
- Channel Secret
- Channel Access Token
These are the “keys” to cascading AI.
Step 2: Establish a knowledge base
Put your FAQ into a structured format. It is recommended to use this template:
| Question Category | Frequently Asked Questions | Standard Answers |
|---|---|---|
| Returns and exchanges | Can I return goods? I don’t want it anymore, how to return it | You can apply for return or exchange within 7 days of receiving the product… |
| Logistics | When will it arrive? Has it been shipped? Tracking | Shipment takes 1-3 working days after payment is completed… |
| Payment | Can I pay by credit card? Payment method, how to pay | We provide credit card, ATM transfer… |
The quality of the knowledge base directly determines the quality of AI responses. The time spent on this step will not be wasted.
Step 3: Connect to the AI platform
Taking Botpress as an example of the serial connection process:
- Create a new Bot in Botpress
- Upload knowledge base files
- Set the brand tone (e.g. friendly, professional, use Traditional Chinese)
- Select LINE in Integrations
- Fill in the Channel Access Token and Channel Secret
- Set Webhook URL
The entire process takes about 30 minutes. No need to write programs.
Step 4: Set transfer rules
Set in the AI platform:
- AI confidence score < 70% → Reply “Let me transfer you to customer service personnel”
- The customer says “find a real person” 2 times in a row → transfer immediately
- Refund amount involved → Transfer to manual processing
- Customer expresses dissatisfaction → transfer to manual + attach summary of conversation
Step 5: Test and go live
- Test at least 50 different questions with your own LINE account
- Ask colleagues to use various “weird questions” to test the AI’s understanding ability
- Check whether the transfer rule is triggered normally
- Start it during non-peak hours first, and observe it for 1 week before fully launching it.
Advanced Tips: Make AI Customer Service Smarter
Tip 1: Analyze open issues
A weekly review of conversations that the AI couldn’t handle and were transferred to humans. These are gaps in your knowledge base.
Add the human customer service answers to the knowledge base, and the AI will be able to do it next time. This is the key to making AI customer service smarter the more you use it.
Tip 2: Active recommendation
AI can’t just passively answer questions. For example:
The customer asked: “Do you sell product A?” AI replied: “Yes! Product A is priced at XXX yuan. By the way, I would like to share with you that many customers who buy A will also purchase product B because…”
This requires adding “associated recommendation” information to the knowledge base.
Tip 3: Data Tracking
Set up metrics to track monthly:
- AI automatic processing rate (target > 60%)
- Average reply time (target < 3 minutes)
- Customer satisfaction (comparison before and after import)
- Transfer rate (too high means the knowledge base needs to be supplemented)
Want to track customer service data more systematically? Reference Customer service data analysis: response rate, satisfaction, conversion.
Cost analysis
| Project | Level 1 (Basic) | Level 2 (Graphic and Text Menu) | Level 3 (AI Concatenation) |
|---|---|---|---|
| LINE monthly fee | Starting from NT$800 | Starting from NT$800 | Starting from NT$800 |
| AI Platform | — | — | NT$0-3,000 |
| Set time | 30 minutes | 2 hours | 1-2 weeks |
| Maintenance time | Low | Low | 1 hour per week |
| Automatic processing rate | 20-35% | 40-58% | 65-85% |
The data tells us that Level 3 has the highest return on investment, but Level 2 also has a good CP value. If you have absolutely no technical skills, starting with Level 2 is the most pragmatic option.
FAQ
**Q: After the Messaging API is enabled, can the original automatic response function still be used? ** Can. You can set the webhook to respond automatically to messages that are not processed by the webhook. Both systems can coexist.
**Q: Will using AI to reply exceed the message fee of LINE? ** There is no additional charge for AI reply messages using “Reply Message”. Only “Push Message” is counted in the free number.
**Q: My customers are older, will they not be used to talking to AI? ** Good AI customer service does not make people feel that it is AI. Set your tone of voice and reply at a moderate pace (don’t respond instantly, wait 2-3 seconds to feel more natural), and most customers won’t notice the difference.
Next step
- If you don’t have an official LINE account → Free application
- Do Level 2 (image and text menu) first, and it will be online in 1 day
- Observe for 2 weeks. If the automatic processing rate is < 50%, upgrade to Level 3.
Free download: GA4 Report Translation Comparison Table - Translate those incomprehensible advertising indicators into human language.
Want to evaluate which level is best for your situation? Book a free consultation.
Further reading: