How to Manage Telegram DMs at Scale

There is a point where a Telegram inbox stops being something you work through and starts being something that works you. The messages arrive faster than you answer them, the follow-up you meant to send three days ago never went, and the conversations that would have closed are sitting unread behind the ones that never will.

The word people reach for at that point is automation. It is the right instinct and the wrong word, because automation on Telegram means three separate things with three separate risk profiles, and confusing them is how accounts get restricted.

This guide separates them, says which parts hold up under volume, and is specific about where the ceilings sit. Platform features come from Telegram's own material. The numbers around sending limits come from operators rather than from Telegram, which publishes none, and they are labelled as such wherever they appear. The campaign figures are our own measurements.

Three things called automation

The first is response automation. Something answers on your behalf when you are not there. A greeting to a first-time contact, an away message outside working hours, a canned reply to a question you get forty times a week. This is the safest kind, because Telegram builds it for you and the recipient asked for the conversation.

The second is outbound automation. You start conversations with people who did not write to you first. This is where the volume is, and where every restriction comes from.

The third is follow-up automation. Someone replied, or half-replied, or went quiet, and a second message goes out later. This is the one almost nobody builds properly, and it is the one with the best return, because the person already knows who you are.

Most people trying to solve a scale problem build the second, ignore the third, and never notice that the third is where their money was.

Response automation: what Telegram gives you

Start here, because it is free, permitted and it removes the largest amount of repetitive typing.

Telegram's business features include a greeting message sent automatically to anyone contacting you for the first time, with a configurable period after which a returning contact receives it again. There are away messages for when you are closed or on holiday. There are quick replies, which are shortcuts that send preset content, and Telegram's own description is worth reading closely: a quick reply "may contain several messages and support text formatting, links, stickers, media, and files."

That last detail is where the leverage is. One shortcut can be a whole sequence: the answer, the image, the file, the next step. You type two characters and four messages go out in order, in your own voice, to a person who wrote to you.

You can also connect a bot that processes and answers messages on your behalf, choosing which chats it can reach. Telegram's developer documentation confirms a connected bot receives new messages, edits and deletions from the business account, so it can keep its own record of the conversation.

None of this puts an account at risk. The recipient initiated the conversation, and everything above is a supported feature rather than a workaround. We covered the full configuration, including the folder and tag limits, in turning Telegram into a customer tool.

Outbound automation: where the ceiling actually is

This is the part everyone means when they say scale, and the honest version of the answer disappoints people.

Telegram does not publish per-account sending limits, and that absence is not an oversight. There is no single number to publish, because the limit is produced by at least five inputs at once, and changing any one of them moves it.

Getting this wrong in either direction is expensive. Treat the limit as a fixed number and you will hit it with a message people hate. Treat it as purely a function of complaints and you will conclude that an unreported message can be sent forever, which is false and is the fastest way to lose a batch of accounts.

The five inputs, and what each one does

Account age and history. A fresh account and a two year old account are not on the same ladder. Published guidance from the tooling side puts a fresh account somewhere around forty to fifty new contacts a day, and an account warmed over two or three months at eighty to a hundred. Those figures come from operators rather than from Telegram, so treat them as a range other people arrived at rather than as a specification. The direction is not in doubt: a new account attempting volume in its first days is the most reliable way to get restricted, and a handful of unsolicited messages inside the first forty eight hours is enough on its own.

Connection health. Accounts sharing an address are grouped, so one flagged account puts the others at risk. Datacentre addresses are treated worse than residential ones because bulk operations run on servers, and an account registered from one can carry limitations from the first day. This is why the proxy question is not a detail, and we went through the types and the trade-offs in the proxy guide.

Rate and rhythm. Velocity matters separately from daily volume. Fifty messages spread across a working day and fifty inside an hour are the same number and a different signal, because the second one does not look like a person. Bursts are what get noticed, not totals.

The message itself. Identical text sent repeatedly is the single clearest pattern there is, more legible than volume, because no human sends the same paragraph to two hundred people. Varying the message is not a trick to evade detection; it is what actually happens when a person writes to people individually.

Recipient reaction. Reports and blocks feed back fastest of the five. Operator guidance puts a handful of reports inside a day as enough to produce a temporary block, which means a small number of annoyed people outweighs a large number of indifferent ones.

Why the arithmetic still does not work

People want volume divided by a number. The reason that fails is not that the limit is mysterious, it is that the five inputs multiply rather than add. A warmed account on a clean residential connection sending a varied message at a human pace has a very high ceiling. The same account sending identical text in bursts has a low one, and no amount of age rescues it.

So the practical order is: fix the message and the pace first, because they are free, then the connection, then age, and only then add accounts. Doing it in the reverse order is what produces the common story of someone buying thirty OLD accounts and losing them in a week.

What a restriction actually looks like

Telegram's own bot for this reports the restriction and, when it lifts, tells you. The important part is that it is usually temporary and usually appealable, and that the appeal has a specific shape. Answering it badly wastes the attempt. We wrote the process out in the spam restriction guide, including what the reply should and should not contain.

Two things are worth knowing before you get there. A restricted account can usually still talk to people it has spoken to before, so an existing conversation continues while new outreach stops. And a fresh authorisation on the account, a new device or a new session, does not clear anything; the restriction sits on the account.

The mistake that costs the most: treating follow-up as more outbound

Here is the asymmetry nobody plans around.

A first message to a stranger is the riskiest message you will ever send, and it has the lowest reply rate of anything in your pipeline. A second message to someone who already answered is almost risk-free, and it converts several times better, because the recipient recognises you and has already chosen not to report you.

Yet almost every setup we have seen automates the first and does the second by hand, which means the second mostly does not happen. The person who said "interesting, send me details" on Tuesday gets nothing on Thursday, because Thursday was busy.

Flip the priority. If you can only automate one thing, automate the follow-up.

Why it gets skipped

Because it feels like the easy part. Outbound is a project: lists, accounts, a tool, a campaign. Follow-up is just remembering, and remembering does not feel like something that needs a system until you are holding sixty conversations and none of them are moving.

The measurable symptom is a pipeline where everything is in one state. Lots of first messages sent, lots of half-conversations, nothing closed. If that is your inbox, your problem is not sending volume.

Designing a follow-up sequence that does not annoy people

Three messages, and a rule for stopping. That is the whole design.

The first follow-up: answer the question they did not ask

Someone showed interest and went quiet. The reason is almost never that they forgot. It is that something in what you sent did not resolve, and they did not want to spend a message asking.

So the first follow-up carries new information rather than a reminder. Not "just checking in," which contains nothing. Something like the specific thing your product does for the situation they described, or the number they would need to decide. If you cannot think of anything new to say, that is a signal about the first message rather than about them.

The second: make the next step smaller

If the first ask was a purchase, the second ask should be smaller than the first. A question they can answer in three words. The volume they are thinking about. Which of two things fits.

A smaller ask is not a retreat, it is a different question. The person who will not commit to an order will often answer "which of these two" because answering costs them nothing.

The third: close the loop and mean it

The last message says you are stopping. Not as a threat and not as a guilt trip, as a fact: here is where we left it, we will leave it here, write any time.

This message gets replies at a rate that surprises people the first time they use it, because it removes the pressure that was keeping them silent. It also does something for you: it ends the conversation cleanly, so it stops occupying space in your head and in your list.

The stopping rule

Three messages and out, unless they reply. Not five, not "until they buy." A fourth unanswered message does not improve your odds and it does change how you are perceived by someone who might have come back in six months.

Timing: the part people get wrong in both directions

Two opposite mistakes, and both are common.

Too fast. The follow-up that arrives four hours later reads as pressure, and worse, it reads as automated, which is the one impression that turns a warm conversation cold. If your second message arrives before the recipient has had a normal day, the sequence is announcing itself.

Too slow. Two weeks later, the context is gone. They have to reconstruct who you are before they can decide anything, and most people will not do that work for a message they did not ask for.

The middle is not a magic number, it is a principle: long enough that a working day has passed, short enough that the previous message is still in recent memory. In practice that means the next working day for the first follow-up, and a few days for the second.

There is one more timing rule and it is the one most systems ignore. Mirror their pace. Someone who took two days to reply to you does not want an answer in ninety seconds. Matching their rhythm reads as a business with other work, and answering instantly reads as a business with none.

The message is the throttle

Of the five inputs above, the message is the one people spend the least time on and the one with the best return, which is why it gets its own section. Age, connection and pacing all cost money or time to improve. The message costs an afternoon and it is the input the other four are compensating for.

That is the argument for doing it first, not the argument that it is the only thing that matters. A perfect message from a fresh account on a datacentre address sending in bursts will still get restricted.

Why most first messages are ignored

Read a hundred cold messages and they share a shape. They open with a greeting that carries no information. They describe the sender. They describe the product. They end with a request for a call.

The recipient's decision happens in the first line, before any of that. If the first line does not tell them why this concerns them specifically, the rest is not read, and the length of the rest does not matter.

So the practical rule is brutal and simple: the first line either says something true about their situation, or the message is already finished.

Length

Short, and shorter than instinct suggests. A long first message does not read as thorough, it reads as a template, because nobody writes six paragraphs to one person they do not know. Length is a tell.

There is a second reason, less obvious. A long message answers questions the recipient has not asked yet, which removes their reason to reply. A short message that leaves one obvious question unanswered gets a reply asking it, and now you have a conversation.

The link problem

A first message with a link in it performs worse than the same message without one, for two separate reasons that compound.

Recipients treat unsolicited links as a risk signal, which is a reasonable instinct. And a message carrying a link asks the reader to leave the conversation and go somewhere else before they have decided they care.

Put the link in the second message, after they have replied. The reply is the permission.

One message, one product

The temptation with a list is to mention everything you do, in case one of them lands. It has the opposite effect: a message describing five things describes nothing, and it tells the reader you do not know which one applies to them.

If your catalogue is broad, segment the list and send different messages, or pick the one thing that fits the group you are writing to. We wrote more about the message construction itself in the guide to sending mass DMs.

Targeting beats volume, and it is measurable

The second lever, and the one with the better ratio of effort to result.

A message to a thousand people who have nothing to do with your product produces some reports, no sales, and a restricted account. The same message to two hundred people in exactly the right place produces a conversation rate that makes the whole exercise worth doing. Volume is the expensive way to compensate for bad targeting.

The measurable part is this: reply rate isolates targeting from writing. If reads are high and replies are near zero across two different messages, the problem is who you are writing to, not what you wrote. Changing the copy again will not fix it.

The quality of a source, in numbers

Not all lists are the same, and the differences are larger than people expect. Two figures from our own scraping work, on real groups.

A group displaying 10,339 members returned three reachable accounts, and all three were bots. Membership counts and reachable members are different quantities, and the gap can be almost the entire number.

In a healthier set, two groups produced 372 members with 369 usernames, and 677 with 474. So even in good sources the usable proportion varies by a third, which is enough to change what a campaign is worth before a single message is sent.

The practical consequence: measure a source before you build a campaign on it. The mechanics of doing that are in the group scraping guide, and the point here is only that source quality is a number you can know in advance rather than a thing you discover afterwards.

Sellers are not buyers

One targeting trap worth naming because it is easy to walk into. Many of the most active groups in any niche are full of people selling the same thing you are. They look ideal by every mechanical measure: large, active, on topic. They convert at nothing, because everyone in the room is on your side of the table.

Before committing a list, read twenty messages from the group. If most of them are offers rather than questions, that group is a supplier directory, not an audience.

Account capacity, honestly

The arithmetic everyone wants is volume divided by accounts. Here is why that arithmetic is a trap, and what to use instead.

Adding accounts does not add proportional capacity, because the accounts inherit whatever is wrong with the other four inputs. Ten accounts sending identical text in bursts from one address produce ten restrictions, slightly later than one would have, and often faster than that, since shared addresses group them together.

What accounts genuinely buy you is isolation. When one is restricted, the others keep working, and the campaign does not stop. That is worth paying for, and it is a different benefit from raising the ceiling.

The practical consequence is a sequence rather than a formula. Start with one account well under whatever volume you want. Watch what happens for a few days. Only add the next account once the first one is stable, because a message that restricts one account will restrict five.

What actually breaks first at volume, in our experience, is not the accounts. It is knowing which account is where: which one is resting, which one is restricted, which one already spoke to this person. That is the problem Account Manager was built for, and it appears earlier than most people expect, usually somewhere around the fifth or sixth account.

What breaks before the accounts do

Three failures that arrive before any restriction, in the order they usually show up.

You message the same person twice

Two accounts, two lists, one overlap. The recipient receives the same pitch from two strangers and now knows exactly what they are looking at. This is more damaging than a restriction, because it converts a possible customer into someone who tells other people.

Deduplicate before you send, not after. When we scraped nine groups for a client the raw total was 2,180 members and the unique total was 2,145, which sounds like a small overlap until you realise those 35 people would each have received the message twice.

Replies arrive somewhere nobody is watching

The campaign is running from three accounts. The replies land in three inboxes. You are watching one. By the time you open the others the warm ones have gone cold, and the follow-up sequence that was supposed to catch them never fired because nobody saw the reply.

This is the single most expensive failure in the whole discipline, because it wastes the messages that worked.

The sequence keeps running after the conversation changed

Someone replies "I already bought from you" and receives follow-up number two anyway. Automated sequences that do not stop when a human answers are worse than no sequence at all, because they prove nobody read the reply.

Any follow-up automation needs one non-negotiable rule: a reply stops the sequence. Everything else is a preference.

The creator case is genuinely different

People running a channel or an audience arrive at this topic from a different direction, and the standard advice does not fit them.

A creator's inbound volume is not from strangers they chose to message. It is from an audience that already follows them, which means every message is warm and the risk profile inverts. Nobody reports a creator for replying to their own subscribers.

So for a creator the problem is never sending limits. It is triage: separating the question that deserves a real answer from the fortieth identical one, and doing it without becoming unreachable.

That points at a different setup. Heavy use of quick replies, because the same questions really do repeat. A greeting message that answers the top question before it gets asked. Folders that separate a paying member from a casual follower. Almost no outbound at all.

The failure mode for a creator is also different. It is not a restriction, it is the moment the audience notices the replies stopped sounding like a person. A creator who automates their voice loses the thing the audience was there for, and that loss does not show up in any metric until it is done.

Measuring whether any of it works

Most people measure sends, because sends are the easy number. Sends tell you almost nothing.

Delivered is not read. A message can be delivered and never opened, and on some platforms it can sit in a request folder the recipient never checks. A delivery figure on its own is a measure of your effort, not of their attention.

Read is not interested. Read rates tell you the first line worked. That is genuinely useful, because it isolates one variable: if reads are high and replies are low, the problem is in the message body rather than in the targeting.

Reply rate is the only number that predicts revenue. Everything before it is diagnostic.

One caution about click tracking, because it produces the most misleading numbers in this discipline. If a message contains several links and only one of them is tracked, the click rate you see is a fraction of the real one. We watched a customer complain about a click rate mid-campaign that turned out, at close, to be above the industry benchmark we had on file, and only one of three links had ever been measured. Know what your number is counting before you make decisions with it.

Choosing a tool, or not

Three honest options, in increasing order of commitment.

Telegram's own features, plus discipline

Quick replies, greeting, away message, folders as stages. Costs nothing, breaks nothing, and handles a genuinely surprising amount. If you are one person with under a hundred live conversations, this is not a stepping stone to a real solution, it is the solution.

What it does not do is outbound, and it does not span accounts.

A managed campaign

Somebody else runs the sending from their accounts and hands you the results. You give the message and the target, they return delivery, reads and the list reached. The advantage is that the account risk is not yours. The disadvantage is that replies land in their inbox rather than yours, which matters enormously if the follow-up is where your money is.

This suits a single push into a new audience. It suits ongoing relationship building much less.

Software you run yourself

Your accounts, your inbox, your pacing. The replies arrive in accounts you own, which means the follow-up sequence is yours to build and the relationship stays with you. That is what our Mass DMs software does, and the reason we point people at it for anything recurring rather than at the managed version.

The trade is that account health becomes your problem. Which is manageable, and is mostly a question of pacing and of not sending a message people dislike.

A setup that holds at a few hundred conversations

Concretely, in the order you should build it.

Week one, response automation only. Six quick replies for the six answers you actually repeat. A four-line greeting that answers the top question. An away message with your hours in it. Folders as pipeline stages that drain rather than fill. No outbound yet.

Week two, the follow-up sequence. Write the three messages before you need them. First follow-up next working day with new information. Second a few days later with a smaller ask. Third closes the loop. A reply stops everything.

Week three, outbound, small. One account, well under the volume you eventually want, into a target list you deduplicated. Watch reads and replies rather than sends. Change the message before you change the volume.

Week four, one more account, and only then. If the first is stable. The second account is for isolation, not for doubling; expect the same reply rate, not twice the results.

If you are starting outbound from a standing start, the first week of an account's life deserves its own reading, and we set out what actually triggers trouble in avoiding account bans in cold outreach.

The three questions that decide whether to automate at all

Before building any of this, three questions settle whether automation is your actual problem. They are worth answering honestly because the wrong answer costs weeks.

Are conversations dying, or are they not starting?

Different problems, opposite solutions. If you have plenty of first messages and nothing progresses, your gap is follow-up and message quality, and more sending volume makes it worse. If almost nothing arrives at all, your gap is reach, and a follow-up sequence has nothing to work on.

Count the two separately for a week before deciding. Most people who think they need to send more actually need to answer better.

Is the bottleneck typing, or deciding?

If your day disappears into typing the same six answers, automation solves it immediately and completely. If it disappears into deciding what to say to people whose situation is all different, automation solves nothing, and the thing you need is a clearer offer so the decisions stop being hard.

The tell is whether you could write your replies in advance. If you could, automate. If every reply genuinely requires judgement, you have a positioning problem wearing an efficiency costume.

Would you be embarrassed if the recipient knew?

The most useful filter, and the least technical. A greeting message that says what your hours are is fine to be caught doing; everyone understands it. A sequence that pretends to be a person typing in real time is not, and getting caught at it costs more than the sequence ever earned.

Automate the parts that are honest about being automated. Keep the parts that depend on sounding human actually human.

What we run on our own side

Our own inbox uses Telegram's built-in features for everything they cover: quick replies for the repeated answers, including the payment instructions so that an address is never retyped by hand, folders as stages, a greeting that answers the top question.

Outbound and follow-up run through our own software, from our own accounts, which is the only arrangement where the replies land somewhere we control. The follow-up rule is the one written above: three messages, a reply stops the sequence, and the third one closes the loop honestly rather than pretending there will be a fourth.

The pacing is deliberately below what the accounts could take, because the cost of a restriction is not the account, it is the campaign stopping in the middle. And every list gets deduplicated before it is sent, for the reason in the section above.

The measurement side is the part we would keep even if everything else changed. Sends, reads, replies and the list actually reached, in one place, per campaign, so that a change to the message can be judged against the previous one rather than against a feeling. That reporting is the reason our own Mass DMs pages exist at all; the sending is the easy half.

Frequently Asked Questions

How many DMs can I send on Telegram per day?

Telegram publishes no number, because the limit is produced by five things at once: account age and history, connection health, sending rate, the message itself, and how recipients react. Operator guidance puts a fresh account around forty to fifty new contacts a day and a warmed one at eighty to a hundred, which is a range other people arrived at rather than a specification. The safe reading is that no single factor gives you a ceiling: an unreported message still gets restricted if it goes out identically, in bursts, from a fresh account on a datacentre address.

Does the account age actually matter, or is it just marketing?

It matters, and the first days matter most. A fresh account attempting volume is the most reliable way to get restricted, and operator guidance describes even a few unsolicited messages inside the first forty eight hours as enough on its own. What age buys is headroom, not immunity: an old account sending identical text in bursts still gets stopped.

Do proxies and IP addresses affect this?

Yes, and separately from everything else. Accounts sharing an address get grouped, so one flagged account puts the rest at risk, and datacentre addresses are treated worse than residential ones because bulk operations run on servers. An account registered from one can carry limitations from the first day. The types and trade-offs are in our proxy guide.

Is sending rate different from daily volume?

Yes, and this is the one people miss. Fifty messages across a working day and fifty inside an hour are the same total and a completely different signal, because the second does not resemble a person. Bursts get noticed before totals do, so spreading the same volume is free protection.

Is Telegram DM automation against the rules?

The response side is built into the platform: greeting messages, away messages, quick replies and connected bots are supported features. Outbound automation to people who did not contact you is the part that draws enforcement, and what gets enforced is behaviour rather than tooling. Volume and recipient reports are what produce a restriction.

What is the safest kind of automation?

Follow-ups to people who already replied. The recipient knows who you are and has already chosen not to report you, so the risk is close to zero, and the conversion is several times better than a first message to a stranger. It is also the part almost everyone leaves manual, which is why it mostly does not happen.

How long should I wait before following up?

Next working day for the first one, a few days for the second. Faster than that reads as pressure and announces that a sequence is running. Slower than that means they have to reconstruct who you are, which most people will not do. Mirror their pace as well: someone who took two days to answer does not want a reply in ninety seconds.

How many follow-ups should I send?

Three, then stop unless they reply. The third message should say plainly that you are leaving it there, which gets more replies than people expect because it removes the pressure that was keeping them quiet. A fourth unanswered message does not improve the odds and does affect how you are remembered.

Does buying more accounts let me send more?

Not in the way people hope. Because the ceiling reacts to recipient behaviour, ten accounts sending a disliked message produce ten restricted accounts. What accounts genuinely buy is isolation: when one stops, the campaign continues. Add the second only once the first is stable.

What breaks first when scaling Telegram DMs?

Not the accounts. First you message the same person twice because two lists overlapped. Then replies start arriving in inboxes nobody is watching, which wastes the messages that actually worked. Then a sequence keeps running after somebody answered. All three are visible before any restriction appears.

Should a sequence stop when someone replies?

Always, without exception. An automated message that arrives after a human answered proves nobody read the reply, and that impression is worse than never having followed up. Every other setting in a follow-up system is a preference; this one is not.

Is this different for content creators?

Substantially. A creator's messages come from an audience that already follows them, so sending limits are not the constraint and outbound barely features. The problem is triage: answering the question that deserves an answer without becoming unreachable. The failure mode is also different, and it is the moment the audience notices the replies stopped sounding like a person.

What should I measure?

Reply rate, because it is the only figure that predicts revenue. Reads are diagnostic: high reads with low replies isolates the problem to the message body rather than the targeting. Sends measure your effort and nothing else. And check what your click tracking actually counts, because a message with three links and one tracked link reports a third of the truth.

Managed service or software I run myself?

A managed campaign keeps the account risk off you and suits a single push into a new audience. Its cost is that replies land in someone else's inbox, which matters if follow-up is where your revenue comes from. Running it yourself keeps the replies and the relationship, at the price of owning account health. For anything recurring, own it.

Should I put a link in the first message?

No. Unsolicited links read as a risk signal, and a link asks the reader to leave the conversation before they have decided they care about it. The second message is the place for it, after they have replied, because the reply is the permission. This costs you nothing and removes one of the two most common reasons a first message is ignored.

How do I know if my list is any good?

Measure the source before building a campaign on it, because the numbers vary more than people expect. One group displaying 10,339 members returned three reachable accounts and all three were bots. In a healthier set, two groups gave 372 members with 369 usernames, and 677 with 474, so even good sources differ by a third in usable proportion. Also read twenty messages from the group first: if most are offers rather than questions, that is a supplier directory rather than an audience.

Should I automate or hire someone?

Answer three questions first. Are conversations dying or not starting, because those need opposite fixes. Is the bottleneck typing or deciding, because automation only solves typing. And would you be embarrassed if the recipient knew, which rules out anything pretending to be a person typing in real time. If the answers point at typing and honest automation, build it. If they point at judgement calls on every reply, you have a positioning problem rather than an efficiency one.

Where should I start if I have never done this?

With response automation, which costs nothing and is entirely permitted: six quick replies, a four line greeting, an away message with hours in it, and folders that describe stages. Then write the three follow-up messages before you need them. Outbound comes last, small, from one account, after the message has been tested on people who already know you.