The 2019–2023 LinkedIn outreach playbook was loud and simple: connection requests at scale, copy-paste DMs, scripted three-step follow-up sequences. It worked because LinkedIn's algorithm and recipients hadn't caught up. They have now, and the playbook is broken.
The funny part is what replaced it: AI-personalised cold outreach. For about eighteen months, that genuinely was a winning move. It scaled the appearance of being thoughtful. Now the inbox is so saturated with "I noticed your recent post about..." messages that recipients have learned to spot the pattern in the first sentence, and the conversion rates have followed.
Two playbooks dead in three years. Worth pausing on what's actually left.
What's actually working
Engagement-led outreach. Comment thoughtfully on a prospect's content for thirty days before you DM. Not three comments. Thirty. The DM, when it comes, is to someone you have an actual track record with. Conversion rates against this motion are currently better than anything we ran in the volume era — by a long way.
Quality over volume, badly underrated. Fifty deeply-researched prospects beat five thousand generic ones. The maths is uncomfortable for sales teams optimised on activity metrics, but the revenue numbers don't lie. We've made the switch internally; we recommend it externally.
Voice and audio DMs. Still novel enough to land. The bar for a good audio DM is much lower than for a great written one — most people just send a clear two-minute Loom, and that's enough.
Founder presence. Founders posting in their own voice, not ghostwritten, not LLM-buffed. The signal-to-noise ratio of "this is what the founder actually thinks" is the strongest content signal on LinkedIn right now. It can't be faked at scale, which is why most accounts won't do it.
Where AI helps
Research and preparation. Understanding a prospect's context, recent posts, public commentary, the shape of their company. Drafting prep notes you'll then act on yourself. Briefing yourself before a call. These are useful. None of them produce a message that goes to the prospect.
Where AI hurts
Anywhere it produces a message a human wouldn't have written. Recipients can smell synthetic personalisation now, and the reaction isn't neutral — it's actively negative. A bad AI-generated DM is worse than no DM at all, because it tells the recipient you don't know how to write to them.
How we run it ourselves
We use Prosp.ai for the campaign infrastructure — list management, sequencing, signal capture. Connections, replies, profile views all come back into our system, where an AI agent prioritises who's worth a real conversation. The AI is the research and routing layer. The actual messages are written by us.
If you've been seeing your LinkedIn numbers crater, the answer isn't more volume or smarter AI. It's a different motion entirely. We're happy to walk through what that motion looks like for your business — that conversation usually fits inside the first half of an AI Audit.