Automation Safety System
7-Day Plan SDR System Retention →
Internal Operations

LinkedIn Automation
Safety System

Sending limits, anti-ban rules, tool selection, and the daily AI-assisted loop. Follow this exactly. A banned LinkedIn account kills the entire acquisition channel.

⚠️
Account risk is real. LinkedIn actively detects automation patterns. A restriction on a new account can be permanent. Do not skip the warm-up phase. Do not automate messages in months 1–2. The limits here are conservative by design.
Sending Limits
Phase-by-phase volume and automation rules.
Phase Daily Limit Automation Rules & Actions
Weeks 1–2
Manual Only
20
connections / day
None
Zero automation
Profile warm-up phase. LinkedIn's algorithm watches new and reactivated accounts. Consistent manual activity builds a baseline trust score before any volume increase.
  • Send 20 connection requests manually per day — spread across the day, not in batches
  • Engage with 5–10 posts per day (like, comment) before sending requests
  • Fill out full profile — headline, about, featured, experience, banner
  • Connect with people you actually know first (10–20) to establish graph
  • No automation tools of any kind
  • No bulk messaging
Weeks 3–4
Optional Tool
30
connections / day
PhantomBuster
Connections only
You can optionally introduce PhantomBuster for connection requests — but not for messages. All first messages must still be sent manually. This keeps the human signal on the highest-risk action.
  • Set PhantomBuster with 45–90 second random delays between requests
  • Run during business hours only (09:00–18:00) — not overnight
  • All first messages sent manually within 24–48 hours of acceptance
  • Monitor LinkedIn notifications daily for warning messages
  • Do not automate messages
  • Do not exceed 30 requests/day in this phase
Month 2+
Full Loop
40–50
connections / day max
Connections + First Msg
Never follow-ups
Full automation loop active. Connections and first messages can be automated with Claude-generated personalised openers. Follow-up messages must always be manual. Follow-ups require context from the conversation — automation here is a ban risk and a conversion killer.
  • Use Claude API to generate 40–50 personalised connection notes + first messages at 09:00
  • Send via PhantomBuster with 60–90 second random delays, business hours only
  • Review all replies manually at 17:00 — no automated responses ever
  • Absolute daily cap: 100 total actions (connects + messages combined)
  • Never automate follow-up messages — always manual
  • Never run overnight — only 09:00–18:00
★ Recommended approach from Month 2
Anti-Ban Rules
These are not guidelines. Follow all of them.
🛑
100 action hard cap per day
100
Total actions across all LinkedIn activity — connection requests, messages, profile views, endorsements. Never exceed this. LinkedIn monitors action velocity. 100 is the ceiling for an established account. New accounts: 50.
⏱️
Random delays between every action
45–90s
Randomised delay between every single action, not per batch. Never use fixed delays — LinkedIn detects the pattern. PhantomBuster's built-in randomiser works. Manual senders: use a timer app or just count.
📅
Account age minimum before automation
3 months
Never automate a LinkedIn account under 3 months old. New accounts flagged for automation patterns face permanent restrictions. There is no appeal. If starting fresh: weeks 1–8 are manual only, no exceptions.
✏️
Always personalise the first line
Even in automated messages, the first line must be specific to the recipient. Reference their business name, niche, location, or a recent post. Generic openers are detected both by LinkedIn and by the recipient. Use Claude to generate 40–50 unique openers each morning.
🚨
Stop immediately on captcha or restriction
If LinkedIn shows a captcha, a "we've noticed unusual activity" message, or any account restriction warning — stop all automation immediately. Complete the captcha manually. Do not run any tools for at least 7 days. Resume at 50% of previous volume.
🌐
Use residential IP if running tools
LinkedIn cross-references your IP with your account's usual login location. Data centre IPs are flagged. Use a residential VPN (Bright Data, Oxylabs) or ensure you're running from the same network you normally use. Never run PhantomBuster from a new IP without warming it first.
Tool Comparison
Weeks 1–4 manual. Month 2 PhantomBuster.
Tool Cost Risk Level Capabilities Verdict
Manual
Weeks 1–4 (all phases)
£0
always free
Lowest
✓ Connection requests
✓ Messages + follow-ups
✓ Full conversation control
✕ Time intensive at scale
✕ Hard to maintain 40+/day
Use weeks 1–4
Required during warm-up. No risk. Claude handles copy generation.
PhantomBuster
Month 2+ recommended
~£50
per month
Low–Mid
✓ Connection requests (safe)
✓ First message automation
✓ Built-in random delays
✓ LinkedIn Sales Navigator support
✕ Limited to connection + msg 1 safely
★ Recommended
Safest automation tool. Well-established, cloud-based, good delay controls.
TexAu
Month 3+ optional
~£40
per month
Medium
✓ More workflows than PhantomBuster
✓ Multi-platform (LI + Twitter + email)
✓ Better CSV/CRM export
✕ Slightly more aggressive detection
✕ Requires more configuration
Optional month 3
More powerful but requires careful setup. Only use after PhantomBuster is stable.
Daily Automation Loop
The exact daily sequence from Month 2.
09:00
AI Generate
Generate personalised connection messages via Claude API
Pull today's prospect list from CRM export (CSV). Run the Claude prompt below with each prospect's name, business, and niche. Output: 40–50 unique connection notes + first messages, each with a personalised first line.

Takes ~5 minutes with the API. Output goes into a CSV column ready for PhantomBuster import.
AI
09:30
Send
Send connection requests with random delays via PhantomBuster
Import the Claude-generated CSV into PhantomBuster's LinkedIn Connection Request phantom. Set delay: random(45, 90) seconds. Run only during 09:30–12:00 to mimic natural morning activity. Cap at 40 requests. Let the tool run unattended — do not interrupt mid-run.
Tool
13:00
Send
Send first messages to yesterday's acceptors
Pull the accepted connections list from PhantomBuster's export (updated at 12:30). Match against the pre-generated first messages from yesterday's Claude run. Send via PhantomBuster Message Sender with random(60, 90) second delays, max 20 messages. Afternoon send mimics natural engagement patterns.
Tool
17:00
Human Review
Check replies, qualify leads, draft responses
Open LinkedIn. Review all message replies from the last 24 hours. Apply qualification triggers (from outreach-system.html). Flag red/amber leads in CRM. Draft responses manually — do not automate. Reply within the same session where possible. Move warm leads to Call Booked stage.
Human
17:30
Log
Log today's KPIs in the tracker
Open kpi.html. Add today's entry: connections sent, replies received, calls booked, proposals sent. Takes 2 minutes. Do not skip — the weekly review on Day 7 depends on this data being accurate.
Log
Claude API Prompt — Morning Message Generation
System Prompt
You are a LinkedIn outreach specialist for Aldric Group, an AI marketing agency.

Generate personalised LinkedIn outreach messages for the following prospects.
For each prospect, output two messages:
1. Connection request note (under 280 characters, no pitch, just a reason to connect)
2. First message after connection (pain question, no offer, under 300 characters)

Niche: {{niche}}
Prospects:
{{prospect_list}}

# Format: one JSON object per prospect
# { "name": "", "business": "", "connect_note": "", "first_msg": "" }

# Rules:
# - Personalise the first line of each message with the business name or niche
# - Never mention pricing, packages, or Aldric Group by name in the connection note
# - Pain question must be specific to their niche — not generic
# - Connection note: conversational, peer tone, no pitch
# - First message: one question only, ends with "?"