How to build a Telegram customer support bot with AI
By the Botable team
A Telegram customer support bot answers the questions you get most, sorts the rest into topics, and escalates anything it cannot handle to a person. Describe your topics and your canned answers to Botable and it builds the whole flow.
What is a customer support bot?
A Telegram support bot works as a triage layer in front of a human team. The first message offers a small set of topics as inline keyboard buttons — order status, billing, technical, talk to a person — because a button is unambiguous where free text is not. Each branch either answers from a fixed set of replies or collects the details a human will need before handing over. The escalation path is the part most people underspecify: decide whether an unresolved conversation forwards to a private admin chat, opens a ticket row in the bot's own database, or simply hands out a support link. Support bots also need a memory of who asked what, so a returning user is not asked for their order number twice; on Botable each bot gets its own SQLite database for exactly this, and every conversation stays inside your bot rather than in a shared vendor inbox.
What commands does it have?
| Command | What it does |
|---|---|
/start | Greets the user and shows the topic buttons |
/help | Lists everything the bot can do |
/status | Looks up the user's open request |
/human | Escalates the conversation to a person |
/faq | Answers the most repeated questions |
What do you need before you start?
- A Telegram bot token from @BotFather
- Your topic list and the answer for each — the bot can only be as good as these
- A destination for escalations: an admin chat ID, a group, or a support link
How do you build it?
What should you write down before building a support bot?
The five questions you answer most, word for word, and what you want to happen when the bot cannot answer. Those two things determine the whole flow. A support bot built without them turns into a menu that leads nowhere.
How do inline buttons make triage better than free text?
A button is a closed set. The user cannot phrase a topic in a way the bot has not seen, so routing is exact and nobody hits a dead end. Reserve free text for the final step, where the user describes their actual problem to a human.
How does escalation to a human work?
Tell Botable where escalations should land. The common pattern is a private admin group: the bot forwards the conversation plus the user's handle, and your team replies from there. Ask for a ticket reference in the message so two staff do not answer the same person.
Can the bot answer in the customer's language?
Yes — describe which languages you support and the bot can detect the incoming language and reply in it. Keep the escalation message in your team's language so your staff are not translating on the fly.
A prompt you can use as-is
Paste this into Botable and adjust the details. It is specific on purpose — a vague description produces a vague bot.
Build a customer support bot. /start greets the user and shows inline buttons for Order status, Billing, Technical issue, and Talk to a human. Each branch answers from a set of replies I will give you. Talk to a human collects the user's question and forwards it with their @handle and a ticket number to my admin group, then tells the user someone will reply. /status shows the user their open ticket.
Who is this bot for?
- Small teams whose support already happens on Telegram
- Shops that want order-status questions handled without a person
- Anyone drowning in the same five questions every day
What goes wrong with this kind of bot?
- A bot that cannot escalate is worse than no bot — wire the human path first
- Do not have the bot promise response times your team cannot meet
- Keep the topic menu to four or five buttons; more and people stop reading it
Questions about customer support bots
Can it read from our existing help centre?
It can fetch from any public URL or API, so a public help centre can be a source. A private knowledge base needs an endpoint the bot can call; describe that endpoint and it will be wired in.
How do we see what people are asking?
Every conversation lands in your bot's own database, and you can ask the dashboard questions about it in plain language — for example which topic was picked most this week.