"I ran auto-reply for a while and my Telegram account got suspended. Where did I cross the line?"
The biggest fear among automation users. A single suspension wipes out months of members, chatrooms, content, and social capital. This post pulls seven safety lines from operating simulations (per-hour messages, reply frequency, zero suspensions).
First, a radar view of the seven risk axes. The gap between the inner (safe) and outer (risky) shape is your suspension headroom.
Bottom line, seven safety lines
| Area | Safe | Risky |
|---|---|---|
| Per-hour sends | ≤ 30 | 60+ |
| Per-day sends | ≤ 200 | 500+ |
| Reply latency | 3~30s | ≤ 0.5s (instant) |
| Same-chatroom streak | ≤ 5 | 10+ |
| Night (1~6 AM) | ≤ 5 | Same as daytime |
| New account | 14-day warmup | Full throttle immediately |
| Duplicate text | Never | Same reply 2+ times |
Each line's rationale below.
1. Per-hour cap, 30 messages
Telegram doesn't publish the limit, but MTProto rate-limit observations plus report-accumulation correlate with:
- ≤ 30/hr: zero suspensions across operators, 6 months
- ≥ 60/hr: one 7-day suspension, one permanent ban (likely report-accumulation)
- ≥ 100/hr: frequent FLOOD_WAIT errors → Telegram is signaling
Replyer's default per-hour cap is 30. Stay inside and you can expect 6+ months without trouble.
Slide the marker to your own sending rate. Green is safe, amber is caution, red triggers FLOOD_WAIT and report accumulation.
2. Reply latency, 3~30 seconds
Instant (≤ 0.5s) is humanly impossible. Both Telegram pattern detection and member reports rise.
- 3~7s: natural (LLM inference + typing simulation)
- 7~30s: typical human operator (reading then writing)
- 30s+: a slightly delayed human (operator was on something else)
Replyer combines LLM generation (37s) + typing simulation (variable chars/sec) + 0.41s pauses → 5~15s average. The safe band.
3. Same-chatroom streak, ≤ 5
5+ sends to the same chatroom within 10 minutes reads as a bot pattern.
- ≤ 5 in a streak: natural (conversation flow)
- 10+: bot suspicion + report risk
- 20+: Telegram automated detection becomes plausible
Replyer's [no-reply probability] option leaves some messages unanswered → streaks break naturally.
4. Night-time avoidance (1~6 AM)
Auto-replies between 1 and 6 AM are the strongest bot signal:
- Human operators rarely reply that late
- Equal volume at 3 AM and 3 PM = obvious bot
Replyer's [active hours] toggles night-time auto-reply off, or cuts frequency 90%. The default night policy sets no-reply probability to 95%.
5. New account warmup, 14 days
A fresh Telegram account immediately running full automation is suspension #1.
- Days 1~3: act like a regular user (join chatrooms, read messages, occasionally reply)
- Days 4
7: 510 manual replies per day - Days 8
14: 2050 auto replies per day (5~10/hr cap) - Day 15+: full operation (30/hr cap)
Skipping warmup raises permanent-ban risk roughly 10x (operator-data estimate).
6. Multi-account spreading
One operator running many chatrooms from a single account is risky.
- Single account at 100/hr → suspension #1
- 2 accounts at 50/hr each → safe
- 3~5 accounts at 30/hr each → very safe
Replyer supports multi-account registration with per-persona account mapping. Four accounts × four chatrooms × 30/hr each = 120 total messages safely.
Same 100 messages/hr, distributed differently. The single tall bar is the risk profile; four short bars stays inside each per-account safe line.
7. Report-accumulation defense
80%+ of Telegram suspensions correlate with report accumulation. Four defenses:
- Disclose at join state automation use in the chatroom intro / pinned message
- Naturalness avoid awkward replies (persona tone / latency / occasional typos)
- No banned content absolutely no ads / external links / spam patterns
- Respond to complaints fast if a member complains about automation, operator personally responds + temporarily pause automation
Post-detection handling in responding when AI replies get spotted.
FLOOD_WAIT errors, Telegram's direct signal
If FLOOD_WAIT_X (wait X seconds) appears during operation, treat it as an immediate signal:
- Cut per-hour cap in half
- Stabilize for one week, then ramp gradually
- Ignoring it accelerates suspension
Replyer auto-backs off on FLOOD_WAIT and fires a Discord webhook alert so the operator can act.
What to do if already suspended
- Temporary (1~30 days) disable automation, return after the suspension with safety lines cut 50%
- Permanent contact Telegram support (recover@telegram.org) explaining automation use honestly. Some cases recover after 30 days
- New account permanently banned accounts rarely return → new account + 14-day warmup + 50% safer caps to restart
Cheapest path is preventive compliance. Post-suspension recovery is expensive.
Frequently asked questions
Q. Is the official Bot API safer?
Bot API is explicitly allowed for automation but limited for chatroom replies. Bots need special permissions in groups + a visible @bot username + cannot mimic the operator's tone. MTProto (user account) automation preserves the operator's tone, but requires strict safety lines. Compare in MTProto vs Bot API.
Q. Isn't 30/hr too low?
Plenty for a regular chatroom (100~500 members). Chatrooms that need 60+/hr are 1,000+ members or bot-driven. Big rooms should split across accounts at 30/hr each. Crossing 60/hr on a single account spikes suspension risk.
Q. Do I have to turn night-time replies fully off?
A frequency cut is more natural than a hard off. Night-time replies at ≤ 5% of daytime volume blend with human operators who occasionally wake up. Replyer's 95% no-reply probability at night yields near-silence with a sprinkle of natural replies.
Q. Does one FLOOD_WAIT mean permanent ban?
No. FLOOD_WAIT is a temporary backoff signal. Halve safety lines + stabilize for one week → recovery. Ignoring it and continuing to push → accelerated suspension. Treat the first FLOOD_WAIT as a warning.
Q. Is multi-account operation legal?
Telegram allows multiple accounts (one per phone number). What matters is disclosure of automation + low report rate. Multi-account is a safety-line spreading tool, not a ban-evasion trick. Each account still must respect the safety lines.
Q. If persona tone mimics well so bot detection is unlikely, am I still at risk?
Yes, residually. Detection is one of several report paths. Reports happen even without detection (awkwardness, missed replies, suspected ads). What controls suspension risk is frequency + time pattern more than naturalness. Replyer's per-hour cap, night-time avoidance, and no-reply probability govern these.
Q. Is Telegram suspension legal?
Telegram's Terms of Service does not blanket-ban automation; it bans spam / ads / mass-sending / unsolicited automated messages explicitly. Auto-reply bots are gray-zone. Safety-line compliance + member disclosure + no ads keeps ToS risk low. Deep analysis in Telegram auto-reply legality.
Next steps
To start auto-replies in your chatroom, download Replyer for your OS and follow the usage manual for the step-by-step guide.
Safety lines here are simulated estimates; Telegram's official thresholds are not public. Adjust conservatively for your chatroom size and reply pattern.