"Hi Sarah, I came across your profile and thought you might be interested in..."
You have received this message. Probably this week. You knew within four words that nobody came across your profile, that nobody thought anything, and that the only personal thing in the message was a database field. The name did not make it feel personal. It made it feel processed.
That is the uncomfortable starting point for any discussion of personalization at scale: the technique most tools call personalization is the one recipients have learned to spot fastest. And on Telegram, a channel people use for friends, family and communities they chose to join, a message that feels processed is not just ignored. It gets reported.
This guide is about the space between two failures: messages so generic they read as spam, and messages so specific they read as surveillance. That space exists, it is wider than most people think, and you can write for it at scale.
What the research says people actually want
The demand for personalization is real. McKinsey's Next in Personalization 2021 report found that 71% of consumers expect companies to deliver personalized interactions, and 76% get frustrated when that does not happen.
But there is a second half to that story that marketing decks tend to skip. In a 2019 survey of more than 2,500 customers, Gartner found that more than half would unsubscribe from a company's communications, and 38% would stop doing business with it, if they found its personalization "creepy". Gartner's conclusion was that the personalization that works is the kind that helps the customer, not the kind that shows off how much the company knows.
In 2025, Gartner went further. A survey of 1,464 B2B and consumer buyers found that people exposed to heavy personalization were 3.2 times more likely to regret their purchase and 44% less likely to buy again, largely because it made them feel rushed and pressured.
Put together, the data says something quite precise. People do not want to be recognised. They want to be understood. Those are different things, and the difference is the entire craft of writing outreach.
What Telegram users actually report
Telegram is unusually direct about what its users object to. Its official spam FAQ lists what people typically mind receiving from strangers: unsolicited advertisements, links, invite links to groups or channels, random photos and, above all, anything related to commerce or online popularity. When enough recipients press "Report spam", moderators review the messages and the sending account can be limited — first for a few days, then for longer if it happens again.
Read that list again as a copywriting brief and it becomes practical. A first message that contains a link, an invite and a sales pitch hits three items on the list in one go. A first message that asks a relevant question hits none of them. The difference is not the recipient's mood. It is what you chose to put in the first message.
No tool can make unwanted messages wanted. Pacing, quotas and well-configured accounts reduce technical risk. They do nothing about a message that the person on the other end did not want. The only real protection against spam reports is writing to people who have a reason to care, in a way that shows you know that reason.
The three layers of personalization
The useful way to think about personalization at scale is not "how many fields can I insert?" but "at which level is this message true?". There are three levels, and they do very different jobs.
Layer 1: The context — true for the whole segment
This is where most of the value is, and where most outreach is weakest. If you found two hundred people in a Telegram group about DeFi yield strategies, you know something real about all two hundred: the topic they care about, the community they chose, often the questions that group keeps asking. A message that speaks to that context is relevant to every single recipient, and you only have to write it once.
This is why the most effective way to personalize at scale is to make your segments smaller, not your messages more complicated. Two hundred people from one focused group, with a message written for that group, beat two thousand people from ten groups with a message written for nobody.
Layer 2: The individual token — hygiene, not strategy
The first name, the username. These are useful, in the same way that spelling someone's name correctly is useful: getting them wrong hurts, getting them right earns nothing. Use them for politeness, never as the thing that makes the message "personal".
The practical risk with tokens is the empty field. A contact with no first name produces "Hi , I noticed..." and the illusion collapses. In TeleBoost, template variables such as {first_name}, {full_name} and {username} are resolved at send time, and a variable with no value is removed cleanly, extra spaces included, rather than sent as a raw tag or a broken sentence. Still, write every template so it reads naturally without the name.
Layer 3: The human — after the reply
Real, one-to-one personalization is expensive, and that is exactly why it should be spent where it pays: on people who answered. Once someone replies, you have what no database field gives you — their words, their question, their situation. That is where a person should take over, read the profile and history, and write like a human.
It is also where language matters. Telegram is global, and the reply that matters is often in a language you do not read. TeleBoost translates incoming messages inline, and lets you write in your language, preview the translation and send it in theirs. The contact's language is something you set, never something guessed from a two-word message.
The rule in one line: personalize the segment, not the sentence. Spend your writing effort on the context that is true for everyone in the group, and your human effort on the people who reply.
The creepy test
Gartner's "creepy" threshold is not abstract. On Telegram, it has fairly clear markers, because everything you can see about someone is something they can see you saw.
| Reference | How it lands | Why |
|---|---|---|
| The topic of the group you found them in | Relevant | It is the reason you are writing, and they know they joined it |
| A question the group often discusses | Helpful | It shows you understand the space, not the person |
| A specific message they posted last week | Borderline | Fine in a small community, unsettling from a stranger |
| The number of groups they are in | Creepy | It tells them they were tracked across communities |
| Their activity times or "last seen" | Creepy | It is surveillance, however it is phrased |
A simple test before you send: would you be comfortable if the recipient asked "how do you know that?" If the honest answer is "you posted it in a public group about this exact topic", you are fine. If the honest answer involves the word "tracked", rewrite.
Anatomy of a first message that scales
Most good first messages on Telegram share the same structure. It is short, and every part has a job.
- Who you are, in one line. Not a company pitch. The reason a person with your job would write to a person in their community.
- Why them, at segment level. The context layer: the topic, the group, the problem that community is known to have.
- One question. Easy to answer, genuinely useful to you, and possible to answer with "no" without awkwardness.
What is deliberately missing: the link, the deck, the price and the calendar invite. They belong in the second or third message, once the person has shown they want them — which is also when a tracked link starts producing a meaningful signal, because the person clicking it asked for it.
An example, and what makes it work
Hi Marco — I run partnerships for a small analytics tool used by a few trading communities. Your group comes up a lot when people discuss on-chain data for smaller caps. Quick question: is that something your members actually ask for, or mostly noise?
The name is used once and the message survives without it. The "why you" is true for the whole segment of group admins it was sent to. The question is useful to the sender whatever the answer, and it invites a reply rather than a decision. There is no link, no offer and nothing on Telegram's list of what people report.
Scaling it without losing it
Writing one good message is craft. Sending it to a thousand people without it degrading is process. A few rules that hold up:
- One message per segment, not one message for the database. If you have five communities, you need five first messages. That is five pieces of writing, not a thousand.
- Follow up only the silent. In TeleBoost, each follow-up can target only people who have not replied, so nobody who already answered receives "just bumping this".
- Pace like a person. Per-account daily quotas and delays between messages are yours to set. A human does not send two hundred identical messages in ten minutes, and your accounts should not either.
- Measure per template, with honest samples. Each template in the Vault keeps its own reply rate and the results that followed it, with small samples flagged as small. Retire the losers; do not tweak them forever.
- Respect refusals everywhere at once. A contact marked do-not-contact is excluded from every campaign, follow-up and API send, and the pre-launch review shows who is being held back before anything goes out.
What about AI-written messages?
AI can help, and it can also produce the most generic messages ever written, faster than ever. The honest position is that AI is good at variation — rewriting the same idea in a different tone or language — and bad at knowing why a specific community should care. In TeleBoost, AI drafting is used where variation matters most: group campaigns, where posting the same paragraph in thirty groups is the quickest way to get muted. The AI writes drafts; you stay the editor.
A pre-send checklist
- Can you state, in one sentence, why this segment in particular should care?
- Does the message read naturally if the first name is missing?
- Is there a link, an invite or a price in the first message? Remove it.
- Does it end with one question that is easy to answer, including with "no"?
- Would you be comfortable answering "how do you know that?"
- Do follow-ups skip everyone who already replied?
- Is everyone who asked you to stop excluded from this send?
Seven questions, two minutes. They will not guarantee replies. They will make sure the people who do not reply at least had no reason to be annoyed — which, on Telegram, is how accounts and reputations last.
TeleBoost is built around this way of working: focused segments from the groups you source, templates with clean variables and per-template results, follow-ups that skip people who replied, inline translation for the conversations that follow, and per-account pacing you control. See how outreach works in TeleBoost, or read how to test a message properly before you scale it.