"A member just asked 'Is this a bot?' How should I answer?"
The moment automation users dread most. A single reply tilts chatroom trust either way, and the wrong move chains into member churn + report accumulation + Telegram suspension. This post compiles a five-step response from operators' actual detection events and the data that followed.
Bottom line, four response patterns and their costs
| Response | Trust impact | Member churn | Recommended |
|---|---|---|---|
| Immediate denial ("Not a bot") | -30% | 10~30% | X |
| Immediate admission + honest note | -5% | 0~5% | O (when pre-disclosed) |
| Staged admission (delayed) | -10% | 3~10% | O (when not pre-disclosed) |
| Ignore | -20% | 5~15% | X |
| Switch to manual operation | -3% | 0~3% | O (large chatrooms) |
Key: denial / ignoring inflicts the heaviest trust loss. Admission + disclosure has the lowest recovery cost.
Below is a grouped bar chart visualizing the trust impact and member churn cost for each of the five response patterns. Hover any bar for the exact percent (simulated distribution based on operating observations).
Prevention, three layers
The safest approach prevents detection from triggering an event:
1. Disclose in chatroom intro / pinned message
The first thing new members see should mention automation:
"This chatroom is operated by X. Some replies use an AI tool trained on the operator's tone. Important decisions / consultations are handled by the operator personally."
This one line nullifies the detection event. A member asking "is this a bot?" gets "yes, exactly as the intro says" and that's it.
2. Naturalness (persona / latency / typos)
The LLM reply itself is the detection signal. Three naturalness levers:
- Persona mimic operator tone (colloquial / emotion / familiar vocabulary)
- Latency 3~30s reply time (never instant)
- Typos / human slips occasional typos / shorthand / "lol" "lmao"
Bot signals (all bot-flagging): instant + perfect grammar + uniform reply length + zero emotion + same volume at 3 AM as at 3 PM.
3. Prepare a detection-response answer
You can't drive detection probability to zero. Pre-write the answer:
"I use an AI tool trained on my tone to handle some replies, to protect my own time. Important responses I do myself. Let me know if you'd prefer otherwise."
Three beats: admission + reason + member agency (they can opt out).
After detection, five-step response
The flow once detection happens. The Mermaid diagram below shows the timeline, decision branches, and outcome states.
flowchart TD
D[Detection event
'Is this a bot?'] --> S1[Step 1, within 5 min
pause + manual reply]
S1 --> S2[Step 2, within 30 min
room-wide notice + pin update]
S2 --> S3[Step 3, 24~48h
manual operation + observe]
S3 --> Q{Recovery signal?}
Q -->|positive| S4[Step 4, 3~7 days
gradual return via review mode]
Q -->|more complaints| S3R[Extend manual
2~3 more days]
S3R --> Q
S4 --> S5[Step 5, 1~3 months
survey + policy update]
style D fill:#fef2f2,stroke:#b91c1c
style S1 fill:#fff,stroke:#3b59c5
style S4 fill:#ecfdf5,stroke:#0f7b6c
style S5 fill:#eef1fb,stroke:#3b59c5
Step 1, immediate response (within 5 min)
Pause automation → operator replies personally. Format:
"@asker, good question. Yes, I use an AI tool for some replies. [reason in 1~2 sentences]. If something feels off, let me know which part."
Three components: admission + reason + member agency.
Step 2, chatroom-wide notice (within 30 min)
Replying only to the asker leaves other members thinking "was I the only one who didn't know?". Broadcast to the room:
"Operator update, I use an AI tool for some replies, [your case], important responses I handle personally. Feedback welcome anytime."
Update the pinned message + add to the chatroom intro = detection → explicit policy.
Step 3, 24~48h manual operation
Pause automation + operator responds personally to reassure members:
- 1~2 days of manual replies before re-enabling
- Extra responsiveness to the asker (small thank-you, follow-up)
- Watch the room (churn / additional complaints / positive reactions)
This is the core of trust recovery. Resuming automation right away signals "I don't care that I was caught".
Step 4, gradual return (3~7 days)
Once mood stabilizes, ramp automation back:
- Return first in [manual review mode] (operator approves before send)
- After 3~5 days return to [auto countdown mode]
- Announce the return ("re-enabling auto-reply, opinions welcome")
This mirrors the gradual-return pattern in the burnout recovery guide.
Step 5, quarterly feedback
Track member trust post-event:
- 1 month later, anonymous survey (auto-reply satisfaction / tone naturalness / extra comments)
- 3 months later, integrate into regular quarterly survey cycle
- Treat the event as a trigger for policy / disclosure / tone improvement
Five wrong responses
Common mistakes:
1. Denial ("Not a bot") when it is
The most frequent mistake. Denial after detection seeds doubt in other members. Once the truth surfaces, trust collapses fully. The fast lane to chatroom shutdown.
2. Ignore + change subject
Ignoring the asker drives that member out plus the observing members. Earns the "weird chatroom" label.
3. Permanently kill automation
Overreaction. Hurts operator time and chatroom sustainability. Pause + gradual return is correct.
4. Argue with the asker
"I told you it's not a bot" or "why are you suspicious" reads as defensive and amateurish, accelerating churn.
5. Disappear
If the operator goes silent post-detection, members read "operator fled". A pre-shutdown signal.
Frequently asked questions
Q. Can I drive detection probability to zero?
Nearly impossible. The LLM reply itself is a signal, and some members will ask out of curiosity, academic interest, or suspicion. Realistic goal: make the post-detection recovery cost near zero. Pre-disclosure + admission keeps the cost under 5%.
Q. Won't disclosing automation in the intro shrink signups?
Data says no. Across 12 chatrooms (6 disclosed / 6 not), signup growth differs less than 5%. Disclosed rooms actually retain members 12% better (zero detection events + trust-based operation). Pattern lines up with the stable-operation psychology piece.
Q. Some members object to automation itself, no?
A minority. Average ≤ 5%. Apologize politely and respect their choice if they leave. 100% satisfaction is impossible for any chatroom. Sacrificing the response quality + operator time for 95% to please a 5% minority is irrational.
Q. Is shutdown / starting over the answer after detection?
Almost never. Shutdown wipes accumulated capital (members / content / trust). Detection events recover via the five-step response (most cases recovered, 1 voluntarily downsized). Shutdown is the last resort.
Q. What minimizes member churn after detection?
Three things: (1) immediate admission, (2) operator personally responds (no auto-reply), (3) acknowledge member agency (their objection is welcome). With these aligned, average churn is 05%. Denial / ignoring / hiding pushes it to 1030%. Response pattern matters more than detection itself.
Q. Can persona tone match the operator 100%?
Realistically 9095%. 100% is the LLM limit. After 510 hours of persona writing + test-reply evaluation + operator-correction cycles, most members can't tell the difference. Process in the agent prompt writing guide.
Q. Does the detection event kill the value of automation?
Once pre-disclosure + response patterns are in place, the value sticks. The event often becomes the trigger that turns implicit automation into explicit policy, settling both operator and members. Long-term stability rises. Pattern in chatroom stability psychology.
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.