How to Automate LinkedIn Outreach with AI (Get More Replies Without Being Spammy)

Arise · 2026-03-23 · 7 min read

Why Most LinkedIn Outreach Gets Ignored

The average LinkedIn InMail has a reply rate of 10–25%. Generic copy-paste messages ("Hi [Name], I love your work at [Company]...") land closer to 3%.

The problem isn't that cold outreach is dead. It's that most people don't have time to do it properly.

Real personalization — reading someone's recent posts, referencing their career trajectory, mentioning something specific about their work — takes 15–20 minutes per prospect. At 5 prospects a day, that's over an hour of research just to start a conversation.

AI agents change this math completely. You can research 50 prospects, draft tailored messages for each, and schedule follow-ups in the time it used to take to write 3.

What the Research Agent Does for LinkedIn Outreach

The Research Agent pulls structured prospect intelligence you can use directly in outreach:

• Company context — recent funding, product launches, team growth signals
• Individual activity — recent posts, articles published, career transitions
• Pain point signals — job postings that reveal what the company is struggling with
• Mutual context — shared connections, events, communities, or topics
• Conversation hooks — specific, non-generic icebreakers for each person

This is the difference between "I saw you work in marketing" and "Congrats on the Series B — saw you're scaling the content team. Timing is interesting given the SEO shifts this quarter."

Installation

# Install the AgentPlace CLI
curl -fsSL https://install.agentplace.sh | bash

# Install the Research Agent
agentplace install research-agent

Basic Usage: Research a Single Prospect

Start with one person to see the output quality:

agentplace run research-agent   --topic "LinkedIn prospect research: Sarah Chen, VP Marketing at DataLayer (linkedin.com/in/sarahchen)"   --depth thorough   --output /tmp/prospect-sarah-chen.md

The agent returns a structured profile:

## Sarah Chen — VP Marketing, DataLayer

### Recent Activity
- Published: "Why attribution models are broken for PLG companies" (2 weeks ago, 847 reactions)
- Career move: Joined DataLayer 4 months ago from Segment

### Company Signals
- DataLayer raised $12M Series A in January
- Currently hiring: Content Marketing Manager, SEO Lead (3 open roles on LinkedIn)
- Product: B2B data pipeline for mid-market SaaS

### Conversation Hooks
- Her attribution post matches DataLayer's likely pain: growth tracking with PLG motion
- Recent hire signals: scaling content output, probably needs distribution help
- Segment → DataLayer move suggests interest in data-driven marketing stack

### Suggested Icebreaker
"Your attribution post nailed it — PLG companies measuring what they can attribute vs. what actually drove pipeline is a huge gap. Curious how you're solving this at DataLayer now that you're scaling."

Scaling to a List of Prospects

For batch research, create a prospect list file and loop through it:

# prospects.txt — one name + URL per line
# "Alex Rivera, Head of Growth at Fintero, linkedin.com/in/alexrivera"
# "Maya Patel, Founder at Loopify, linkedin.com/in/mayapatel"

while IFS= read -r prospect; do
  name=$(echo "$prospect" | cut -d',' -f1 | xargs)
  slug=$(echo "$name" | tr ' ' '-' | tr '[:upper:]' '[:lower:]')
  agentplace run research-agent     --topic "LinkedIn prospect research: $prospect"     --depth standard     --output "/tmp/prospects/$slug.md"
  echo "Researched: $name"
  sleep 3
done < prospects.txt

Writing Personalized Outreach Messages

Once you have prospect profiles, use the Research Agent to draft messages based on the intelligence gathered:

agentplace run research-agent   --topic "Write a LinkedIn connection request for Sarah Chen based on this profile: $(cat /tmp/prospect-sarah-chen.md)"   --format "150 words max, no salesy language, genuine curiosity tone, one specific observation, one relevant ask or shared interest"

Example output:

Hi Sarah,

Your post on attribution gaps in PLG motions hit close to home — we see this constantly
with growth-stage teams trying to justify content investment without clean pipeline data.

Your move from Segment to DataLayer is interesting timing. Curious whether the data
infra side changes how you're thinking about marketing attribution at all, or if it's
still the same measurement problem with better tooling.

Would love to connect — I work with a handful of PLG marketing teams on similar challenges.

Follow-Up Sequence Automation

Most replies come on the 2nd or 3rd touchpoint. Use the Research Agent to draft follow-up variants:

# Generate a 3-part follow-up sequence for a prospect who hasn't replied
agentplace run research-agent   --topic "Write a 3-message LinkedIn follow-up sequence for Sarah Chen at DataLayer. Context: sent initial message about attribution 5 days ago, no reply. Each message should be under 80 words, different angles: (1) share relevant resource, (2) check in with new hook, (3) final polite close."   --output /tmp/followup-sarah-chen.md

Tracking and Organizing Outreach

Build a lightweight tracker by having the agent output to a structured CSV:

agentplace run research-agent   --topic "Summarize all prospect files in /tmp/prospects/ into a CSV with columns: Name, Company, Title, Date Researched, Key Hook, Message Sent (blank), Reply Date (blank), Status (blank)"   --output /tmp/outreach-tracker.csv

Open the CSV in any spreadsheet tool. Update Status manually as you track conversations.

Comparison: Manual vs. AI-Assisted LinkedIn Outreach

Tips for Higher Reply Rates

• Lead with curiosity, not value props. The best opener is a genuine observation about their work, not "I help companies like yours..."
• Reference something specific and recent. A post from last month beats "your impressive background" every time.
• Keep it under 150 words. LinkedIn messages over 200 words get skimmed or ignored.
• Follow up 2–3 times, then move on. No reply after 3 messages means move to the next prospect.
• Personalize the first line only. The hook does the heavy lifting — the rest can be templated.
• Send Tuesday–Thursday between 9am–12pm. Avoid Monday morning and Friday afternoon.

Conclusion

LinkedIn outreach works when it's personal. The problem has always been that personal doesn't scale — until now.

The Research Agent handles the intelligence gathering so your messages can be genuinely relevant without you spending 20 minutes per prospect. The reply rates follow naturally when the message sounds like it was written specifically for that person — because it was.

Get the Research Agent on AgentPlace