Telegram Auto Reply Bot: Setup, Limits and Costs

Almost everything written about Telegram auto replies is out of date, and it is out of date in the same way: it tells you that only a bot can answer automatically, that the bot can only answer people who wrote to it first, and that your own account cannot be automated at all. That was true for years. It stopped being true on the seventh of May 2026, when Telegram gave every user a setting called Chat Automation that connects a bot to their profile and lets it answer on their behalf.

So the first question is not which tool to buy. It is which of four different things you actually want, because they answer different people, under different limits, and one of them did not exist a few months ago. This guide separates them, quotes the limits from Telegram's own documentation, and is honest about the one detail every vendor leaves off the page: the person receiving your automatic reply can tell it was automatic, because the platform marks it.

Every figure below was pulled from Telegram's documentation on the day of writing, against Bot API version 10.2. Where a number is ours, measured on the accounts we run, the text says so.

The question that decides everything: who is doing the replying

Four things get called "Telegram auto reply" and they are not variations of one another. They answer different people.

A bot of your own

A separate account with a username ending in bot, which you or somebody else programs. It answers people who open a chat with it. It cannot open a chat with anybody.

The built in business replies

Greeting messages, away messages and quick replies, set from Telegram's own settings on a Business account. No code, no bot, and they fire from your own account rather than from a bot.

A bot connected to your account

The newest one, and the one that changes the old answer. Your personal or business account keeps its name and its chats, and a bot you authorise reads and answers inside them, within limits Telegram publishes.

A third party autoresponder

An app or service you log into with your account, which then watches your chats and replies for you. Mechanically the oldest approach and the one with the largest downside.

If you only remember one line from this section, make it this: a bot answers strangers only after they speak first, and your account can now be automated but only inside a window Telegram measures in hours.

What changed in 2026, and why most guides missed it

Two changes, both from Telegram, both after most of the advice online was written.

Chat Automation, seventh of May 2026

Telegram's own announcement is short and unambiguous: every Telegram user can connect a bot to their profile and allow it to respond to messages on their behalf, and when connecting a bot you configure which chats it can access, such as excluding all chats with your contacts, or only responding to new chats. The setting sits at Settings, then Chat Automation.

Read that again, because it removes the sentence at the centre of every older guide. Automated replies from a normal account are no longer something you smuggle in with a third party client. They are a setting.

Business chatbots, and how the two relate

Before that, the same capability existed but only for Telegram Business accounts: business users could connect bots that process and answer messages on their behalf, with the same kind of per chat scoping. The May change generalised it to everybody. If you already had a Business account you had this earlier; if you did not, you have it now.

The practical consequence

The old decision was "bot or nothing", and the bot could not reach the people you actually wanted to answer, because they were writing to your account and not to it. The new decision is a genuine choice, and most of the reasons people used to give for the third party route have evaporated.

Route one: your own bot

The classic route, and still the right one for a public front door.

Who a bot may answer

People who already have a chat with it. Someone taps start, or writes to it, or joins a group where the bot is a member, and from then on the bot may reply. A bot cannot pick a username off the internet and open a conversation, and no setting changes that. This is the design of the platform rather than a limit that can be raised.

How the messages reach your code

Two ways, and Telegram documents both. Long polling, where your program repeatedly asks for new updates and each request confirms the previous batch. Or a webhook, where you give Telegram a web address and it sends an HTTPS request there for every update, retrying an unsuccessful request a reasonable number of times before giving up. You choose one; a webhook and long polling cannot both be active.

The part nobody says out loud

Neither of those is an auto reply. They are delivery. Everything a customer would call auto reply, matching a keyword, choosing an answer, remembering where the conversation got to, handing over to a person, is code you write or rent. There is no keyword field in the Bot API, because keywords are not a platform feature.

What that means for the buying decision

When a page advertises a Telegram auto reply bot, it is selling the part after delivery. That is the honest description and it is also the useful one, because it tells you what to compare: not whether it can receive messages, which is free and identical everywhere, but what it does with them.

Route two: the replies built into Telegram Business

Three separate features that people lump together. They are set from the app, they need no developer, and they send from your own account.

Greeting messages

A message sent automatically to somebody contacting you for the first time. Telegram describes it as an instant reply with important information, and it can carry answers to frequent questions or details about the business. There is a setting for the period after which a new message triggers the greeting again, so a customer who returns months later is greeted rather than ignored.

Away messages

A message sent while your business is closed or you are away. Its job is expectation rather than information: the person learns when a human will reply. Telegram also lets you keep automated messages away from friends, family or employees by excluding specific chats or entire chat categories.

Quick replies

Not automatic at all, and worth understanding precisely because of that. Quick replies are shortcuts for preset messages, they can hold several messages at once, and they support formatting, links, stickers, media and files. You type a slash in the message bar and a panel of them appears, ready to send or edit in one tap. Telegram is explicit that this works in private chats and not in groups.

The honest verdict on these three

They are the highest return per minute of setup of anything in this guide, and they solve the two situations that actually annoy customers: nobody answered at all, and nobody said when somebody would. What they cannot do is read what the person wrote. A greeting is the same greeting whether the message says hello or asks your price.

Route three: a bot answering on your account's behalf

The route that changed, and the one with the most documented detail behind it.

What you are actually granting

When a bot is connected to an account, the Bot API describes the permissions as a set of rights, and they are worth reading one at a time rather than as a blob. Whether the bot may reply. Whether it may mark incoming private messages as read. Whether it may delete messages it sent. Whether it may delete all private messages in the chats it manages. Whether it may edit the account's name. Each is separate, which means the interesting question is not whether to connect a bot but which of these to hand over.

The limit that matters most, and it is a clock

The right to reply is defined in the documentation as the bot being able to send and edit messages in the private chats that had incoming messages in the last 24 hours. That single clause decides what this route can and cannot be used for.

It means the bot can answer somebody who just wrote to you, all day and all night, which is exactly the job. It also means the bot cannot start a conversation, cannot chase a lead that went quiet three days ago, and cannot send a broadcast from your account. The window opens when they write and closes a day later.

How the bot hears about the messages

Through its ordinary update stream, with separate update types for a connected account: one for the connection itself being created, changed or removed, one for a new message from the connected account, one for an edited message and one for deleted messages. Individual messages carry an identifier for the business connection they arrived through, so a bot serving several accounts can tell them apart.

Scoping, which is the part to spend time on

Telegram lets you configure which chats the connected bot may touch, and gives two examples in its own wording: excluding all chats with your contacts, or only responding to new chats. Those two settings between them cover most of what goes wrong. A bot that answers your suppliers, your family and your accountant is a bot you will disconnect within a day.

Route four: third party autoresponder apps

The oldest route, and the search suggestions show it is still the most searched: autoresponder for iOS, for desktop, the pro version, the downloadable package.

What they actually are

Two different things sharing a name. Some are ordinary bots with a friendly panel, in which case they are route one with a subscription. Others run your account, either as a modified client or through the user API, which means they hold a live authorisation for your account rather than a permission for one job.

Why this route is weaker than it was

It used to be the only way to make a personal account answer by itself. That was a real reason, and it is gone. Since May 2026 the platform does this natively, with a rights list you can read and a scope you can set, and without anybody else holding your login.

When it is still the answer

When you need something the platform's rights list does not cover, or when you are automating across several accounts that are not yours to connect one by one. Both are real cases. Neither is the common one.

The detail every vendor leaves out: automatic replies are labelled

This is the single most useful thing in this guide and it comes from one field in the Bot API.

What the platform marks

Messages carry a flag indicating that the message was sent by an implicit action, and the documentation names the examples: an away or a greeting business message, or a scheduled message. In other words Telegram itself distinguishes a message a human typed from one the system emitted on their behalf.

Why it matters even if the customer never sees the flag

Because it tells you how the platform thinks about these messages, and because anything reading messages programmatically on the other side can see it. Write your greeting as if the reader knows a machine sent it, because in the only sense that is checkable, they can.

What to do with that

Stop writing automatic messages that pretend to be a person typing. A greeting that says the truth plainly, tells them what happens next and when, outperforms one that says "hey, just saw your message" and then goes quiet for six hours. The second one is a small lie that the next six hours expose.

The limits, in one place

Numbers Telegram publishes, so you can size a setup before building it.

The first three are generous for replying and irrelevant to it: a reply bot answering one person at a time will never approach thirty messages a second. They matter when the same bot also broadcasts, which is a different job covered in our guide to running a Telegram broadcast bot.

What "AI auto reply" means on Telegram now

The phrase turns up throughout the search suggestions, so it is worth saying what the platform actually shipped rather than what the phrase implies.

Bots you can summon into a chat

Telegram now lets a bot be mentioned by username in any private or group chat even when it is not a member, and it replies in the chat where it was tagged. The privacy boundary is stated plainly: such a bot sees only the message it was tagged in and any replies to that message, cannot see who is in the chat, and cannot read other messages.

Streaming replies

Bots can send a response as it is generated rather than waiting for the whole message, which is why some bots now appear to type an answer out in front of you.

Bots talking to bots

Bots can now respond to other bots, not only to people, which is what makes chains of automated steps possible without a person in the middle.

The honest reading

None of that answers your customers by itself. It lowers the cost of building something that does, and it means the sentence "we use AI" tells you nothing about whether a tool will answer a pricing question correctly at two in the morning.

Writing the messages, which is where most of the value is

The routing is a decision you make once. The wording is what every customer actually meets, and it is where a working setup and an irritating one diverge.

The greeting has one job, and it is not to greet

Its job is to remove the reader's uncertainty about what happens next. Most greetings spend their words on warmth and none on that, which is why so many read pleasantly and help nobody. Say who this is, say what to send so the answer comes faster, and say when a person will be here.

A greeting that says "Hi, thanks for reaching out, we'll get back to you soon" has told the reader nothing they did not already assume. One that says what to include in their first message shortens the whole conversation, because the second message is now the useful one instead of the fourth.

The away message lives or dies on a number

"We are currently away" is not information. "We are back at nine in the morning, UTC plus three" is. The reader is deciding whether to wait or go elsewhere, and only the second version lets them decide. Put the actual hours in the text, including the time zone, because half the people reading are not in yours.

If you also want the message to work while you sleep, add the one thing that does not need you: a page, a catalogue, a form. An away message with a working next step is the cheapest sale in this whole guide.

Quick replies are worth auditing, not inventing

The best set of quick replies is not designed, it is discovered. Go through the last fifty conversations you answered by hand and count which answers you typed more than twice. That list is your set. Anything you invent instead will sit unused while you keep typing the real ones.

Because a quick reply can hold several messages with formatting, links and files, the ones that earn their place are usually the awkward answers: the specification, the price list with its caveats, the thing you always get wrong when typing at speed.

Say what the automation cannot do

A line stating plainly that this is an automatic reply and that a person reads everything, costs you nothing and prevents the most common complaint about automated support, which is not that it exists but that people cannot tell whether anybody is behind it.

Setting each one up, in order

Short, because none of these is difficult once you know which one you are doing.

The built in replies

They live in the business settings of the account, alongside opening hours and chat tags. Set the away message first, because it is the one that works while you are not there, then the greeting, then quick replies once you have audited your own answers. Use the exclusion controls in the same pass: Telegram lets you keep automated messages out of chosen chats or whole categories, and doing that on day one avoids sending your opening hours to your own team.

Connecting a bot to your account

Settings, then Chat Automation, then the bot, then the scope. Spend your time on the scope rather than the bot: start with new chats only, watch it for a day, and widen from there. The rights are separate for a reason, so grant replying before you grant deleting.

Your own bot

Create it, take the token, and then face the real work, which is the part after the token. Pick long polling or a webhook, not both. Store who wrote to you as they arrive, because there is no method that hands you that list later. And decide up front what happens when your code does not know the answer, because that path is used more than any other.

A shortcut that is not a cheat

If the part after the token is the part you do not want to build, that is precisely what Bot Manager is: the inbox, the keyword answers set from a form, the welcome flow and the broadcast, without writing the server that keeps them alive.

The handover, which is the part that actually fails

Every automatic reply eventually meets a message it should not answer. What happens at that moment decides whether the whole setup reads as helpful or as a wall, and it is the part that gets the least attention because it is not a feature anybody advertises.

Silence is a decision, so make it deliberately

When somebody writes a sentence instead of choosing an option, the worst answer is to repeat the menu at them. The second worst is to guess. Going quiet and letting a person pick it up is usually right, but only if a person is actually watching, which is a staffing fact rather than a software one.

The handover has to be visible from both sides

The customer needs to know a person now has it, and your side needs to know the automation has stopped so it does not talk over the conversation. A setup that only does the first half produces the situation everybody has been on the wrong end of: a human is typing and the machine interrupts them.

Coming back is as important as leaving

A conversation handed to a person should be able to return to the flow, otherwise the customer who took a wrong turn is stuck for good. We learned this the expensive way on our own accounts: the steps that deliberately ended in silence had no route back at all, and anybody who pressed the wrong thing had no way home. One documented way out, offered on every ending message, fixes it.

Do not automate the moment money is involved

An automatic message can carry a price, a wallet address and a set of options safely, because those are facts. Judging whether an unusual request is acceptable, agreeing an exception, or deciding a discount is not something to hand to a keyword. Those are the conversations worth a person's time, and the automation's job is to make sure a person still has time left for them.

Knowing whether it worked

Automatic replies are unusually easy to feel good about and unusually easy to get wrong, because the thing they produce is silence and silence looks the same either way.

Count replies, not sends

The number that matters is how many people wrote again after the automatic message. A greeting that goes to everybody and is answered by nobody is not neutral, it is a message that ended conversations. Sends tell you the machine ran; replies tell you it helped.

Watch the second message, not the first

If your greeting works, the person's next message contains what you asked for. If it does not, their next message is a repeat of the first one, or nothing. That difference is visible without any analytics at all, by reading ten conversations.

Time to a human, measured honestly

The away message promises a person. Measure how long that actually takes, from their message to a human reply, not from your intention. On our own accounts this number was consistently worse than we assumed before we started counting it, which is our measurement rather than a general claim.

The one number that decides the tooling

How many conversations a day arrive. Under a handful, the built in replies and a person are the whole answer and any tool is overhead. Above that, and especially across more than one account, the inbox becomes the problem before the answering does.

Choosing, in one pass

Match the situation rather than the feature list.

People write to your account and get nothing at night

Away message, ten minutes of setup, no code, no bot. Add opening hours to the text so the reader can work out when to expect you.

Everybody asks the same five questions

Quick replies for the answers you send by hand, a greeting message for the one everybody needs, and only then consider a bot. Most of this problem is not an automation problem.

You want real answers, keyed to what they wrote

A bot. Connected to your account if the conversations arrive there, standalone if you can send people to it. This is where a tool earns its price.

You are running several accounts

Then the question stops being auto reply and becomes an inbox question: who sees the message, from which account, and who answers. That is a different shape of problem, and we wrote about it separately in managing multiple Telegram accounts.

What it costs

Four routes, four different kinds of bill.

The built in replies

Included with the subscription that unlocks them. No per message cost, no server, nothing to keep running.

Your own bot

The Bot API is free. What costs money is the thing that stays running: a server, or somebody else's. A bot that sleeps answers nobody, and this is the line item people forget when they compare a one time purchase against a monthly one.

A connected bot

Same as above plus whatever the bot itself charges. The connection is a permission, not a product.

Third party apps

A subscription, and on the free tiers something else. When a service that costs money to run charges nothing, the account you log in with is the payment. That is arithmetic rather than suspicion, and we went through it in detail in our guide on what a public Telegram repo actually costs to run.

Where this breaks in practice, from our own running

We run automatic replies on our own accounts, and the things that went wrong were not the things the documentation warns about. These are our measurements, not platform claims.

The update stream can die while everything looks fine

We measured a client reporting itself connected, the account live, and a message sent straight at it never reaching a handler. There is no flag for this state. The fix is not cleverness, it is a second path: poll on a timer as well as listening, and let whichever sees the message first handle it. Anything that depends only on the live stream will one day go quiet without telling you.

Two paths mean the same message can be answered twice

The moment we added the second path, a message arrived through both and the customer got two replies, one of them a greeting in the middle of an order. Mark each message as handled the first time. It is three lines and it is not optional.

A keyword that is not the whole message will fire inside a sentence

We use typed words to drive our own flow, and the guard that makes them safe is that the word must be the entire message. Match keywords loosely and an ordinary sentence containing the word triggers the automation in front of a customer.

Every automatic conversation needs a way out

We built steps that deliberately ended in silence and forgot to leave a route back, so anyone who took a wrong turn was stuck with no way to reach a person or restart. Automatic replies fail more often by trapping people than by saying the wrong thing.

Instant is not always better, and then it is

We delayed a payment confirmation by several minutes on the theory that an instant reply reads as a reflex while a pause reads as somebody having looked. Then we measured what customers actually do while waiting and removed the delay. Speed won. The lesson is that this is a question to test rather than to reason about.

Frequently Asked Questions

Can a Telegram account send automatic replies, or only a bot?

Both, since the seventh of May 2026. Telegram added a setting called Chat Automation that lets any user connect a bot to their profile and allow it to respond on their behalf, and you configure which chats it can reach. Before that date only bots could answer automatically, which is why most guides still say an account cannot.

Can an auto reply bot message people who never contacted me?

No, and this is the one limit no tool gets around. A bot may write to people who already have a chat with it, and a bot connected to your account may reply in private chats that had an incoming message in the last 24 hours. Neither can open a conversation.

What is the 24 hour window on a connected bot?

The permission that lets a connected bot reply is defined as covering the private chats that received an incoming message within the last day. So the bot can answer anyone who wrote recently, and cannot touch a conversation that has been quiet longer than that.

Does Telegram have a built in keyword auto reply?

No. Telegram delivers incoming messages to your bot by long polling or a webhook, and everything after that is your own code. Keyword matching, default answers, working hours and handover to a person are things you build or rent, not settings you switch on.

What is the difference between a greeting message and an away message?

The greeting goes to somebody contacting you for the first time and can be set to trigger again after a period of silence. The away message goes out while you are closed or away. One introduces, the other sets expectations, and both can be kept away from chosen chats or whole categories of chat.

Are quick replies automatic?

No, and that is the point of them. They are saved messages you send with two taps, they can contain several messages with formatting, links, stickers, media and files, and you reach them by typing a slash in the message bar. They work in private chats only.

Can the person tell my reply was automatic?

The platform marks it. Messages carry a flag for having been sent by an implicit action, and Telegram's documentation names away messages, greeting messages and scheduled messages as examples. Write them as if the reader knows, because in the only checkable sense they can.

Is a third party autoresponder app safe?

It depends entirely on what it needs from you. One that runs as a bot is ordinary. One that asks you to log in with your account holds a live authorisation for that account until you end the session yourself. Since the platform now offers connected bots with a readable permission list, the case for handing over a login is much weaker than it was.

How many messages can an auto reply bot send?

Telegram publishes about one message per second in a single chat, about twenty per minute into a group, and about thirty per second for bulk notifications. For replying these are never the binding constraint; they matter when the same bot also broadcasts.

What does "AI auto reply" actually mean on Telegram?

Three shipped features rather than one product: bots that can be summoned into any chat by mentioning their username, which see only the message they were tagged in; bots that stream a reply as it is generated; and bots that can respond to other bots. None of them answers your customers by itself.

What breaks most often in a real setup?

Not the wording. In our own running, the update stream went quiet while reporting itself healthy, the same message got answered twice once a second delivery path existed, and conversations ended in dead ends with no way back to a person. Those three cost us more than every message we ever rewrote.