- The advice "hold the account 14 days, set an avatar, subscribe to 10 channels" was relevant for SpamBot version 2022. SpamBot is now at version ~5.
- Account age is not a signal. Activity is. The quality of activity matters more than the duration of passive "resting".
- Real 2026 triggers: interval entropy between actions, swipe patterns, typing cadence, fingerprint stability.
- Myths that don't work: avatar, bio, age, number of channel subscriptions, number of dialogs.
- What you actually need for warmup: 3-5 days of behavioral activity, not passive resting.
This article may cause disagreement among those who sell "account warmup services" or run courses where "14-day warmup" is the key piece of advice. I'm ready for pushback. But the facts are: Telegram's SpamBot has gone through at least 3 generations of refactoring since 2022. Advice that worked back then is now outdated ritual.
How SpamBot evolved
| Version | Key mechanic | What it caught |
|---|---|---|
| v1 (pre-2021) | Simple patterns: frequency, obvious spam words | Head-on mass outreach with a template |
| v2 (2021-2022) | + account age, contact count | Fresh SIMs, accounts without subscriptions |
| v3 (2023) | + content hashing, ASN scoring | Identical texts, DC-proxy clusters |
| v4 (2024) | + behavioral signals (entropy) | Bot patterns in intervals, linear scrolling |
| v5 (2026) | + fingerprint correlation, session graph | Farms with identical device DNA |
Important: each new version didn't cancel the old rules, it added new signals. That's why "account age" still triggers sometimes — as an echo of old logic. But age alone doesn't solve anything.
Warmup rituals
A typical "warmup" in 2022 consisted of:
- Buy an account → hold it "empty" for 14 days.
- Set an avatar, fill in bio.
- Subscribe to 10-20 channels.
- Send 2-3 messages to acquaintances.
- After that — "you can start outreach".
What made sense in 2022: SpamBot v2 counted "subscriptions + messages + time" as a proxy for "live user". The higher the numbers in those fields, the higher the trust. Now:
- Avatar — SpamBot v5 doesn't look at it. Image analysis is too expensive for mass scanning.
- Bio — not looked at, for the same reason.
- 14 days of "resting" — doesn't trigger a positive signal if there was no behavioral activity during that time.
- Subscribing to 10 channels — easily visible as a batch operation (subscribed in a row on the same day).
- 2-3 messages to acquaintances — too few to matter.
Behavioral signals in 2026
What SpamBot v5 actually looks at:
1. Interval entropy
Distribution of pauses between actions. For a live user: log-normal distribution — many short pauses + rare long ones (got distracted by real life). For a bot: typically uniform or fixed delays.
np.random.uniform(30, 60) — the favorite pattern of bot scripts, easy to detect. np.random.lognormal(mean=log(90), sigma=0.7) — human-like.
2. Swipe / scroll patterns
A human's finger trajectory while scrolling is not straight. There's micro-jitter (random twitches), bezier curves (smooth bends), accelerations and decelerations. A bot, when modeling, is usually rectilinear — up-down, at one constant speed.
3. Typing cadence
A human types a message with delays between keystrokes of 150-400ms (depending on typing skill), with occasional deletions (backspace), with pauses for thinking. A bot either pastes the full text instantly (setInputValue) or types at a uniform speed.
4. Fingerprint stability
Between sessions a user doesn't change device. If they log in from "iPhone 15" today and "Samsung S24" the day after — that's either a multi-device user (rare in reality) or fingerprint randomization (a farm flag).
5. Session entropy
Session duration — how long a user "sits in the app". Live distribution: log-normal, from 30 seconds to several hours. Bots: often fixed — "log in, run script, log out".
How to actually warm up an account in 2026
3-5 days of behavioral activity that emulates a live user:
- 5-15 app opens/day, with varying session durations
- Scrolling the chat list in each session, with a realistic trajectory (bezier + jitter)
- Reading messages (changing read state) in random chats
- Reactions to 1-5 messages/day — on posts in subscribed channels
- 2-5 messages/day to contacts (not outreach, but real dialogs)
- Different hours of day — not all actions at 14:00
This activity accumulates a behavioral profile that SpamBot recognizes as a live user. After 3-5 days the account can start outreach — already with a "trust rating".
Important: behavioral warmup continues during outreach too. If the account suddenly becomes "send-only", without any incoming activity — that's a new red flag. A human keeps living: reading news, reacting, texting friends.
Behavioral Warmup worker
For us this is a separate module that runs per-account:
After warmup is launched, the account runs in the background — you do nothing. After N days it's ready for outreach.
TG:ON for macOS · Windows · Linux
Desktop app, 160 MB. Runs locally, your keys stay yours. 3-day trial, no credit card.
Download for freeBehavioral Warmup · 3 days.
Set up in 2 minutes.
Forget about 14-day resting. 3 days of behavioral warmup in TG:ON — and the account is ready. Try it on a 3-day trial.
Start trial