Warm outreach vs cold outreach: what the data says

Warm outreach vs cold outreach is not really a reply-rate question. Both get replies at close to the same rate once a conversation exists, and the difference is the gate in front: cold has to buy that conversation at about five connection requests each.

That is worth sitting with, because most arguments for warm outreach are made on the wrong number. The usual version says warm messages get replies and cold ones do not. The data does not say that. What it says is that cold outreach pays an entry fee before the message is even read, and warm outreach walks in on a connection that was accepted years ago and then never used.

Reach is a LinkedIn tool that runs in your own browser on your own logged-in session. It imports the connections you already have, scores them against who you actually sell to, and drafts a 1:1 message that you approve before anything sends. No password is handed over and no server calls LinkedIn.

What counts as warm outreach, and what counts as cold?

Cold outreach means the person is not connected to you yet, so the first thing you send is a connection request. Warm outreach means they already accepted you at some point, so you can message them directly and the acceptance step does not exist. On LinkedIn that distinction is structural, not a matter of tone. A friendly cold message is still cold, because it still has to get through the invite.

Most people underestimate how much of their own network is sitting in the warm column. This account has 16,069 people imported and 11,168 of them have never been assessed against anything. They are already connected. Nobody has ever looked at them.

Do warm messages actually get more replies than cold ones?

Once a conversation is open, barely. On 17 September 2026 the LinkedIn outreach platform HeyReach published a benchmark built from 96,051 campaigns on its own platform over twelve months. The median campaign there runs a 20.75% acceptance rate, described in the report as about one accepted connection for every five requests sent, and a 22.22% median reply rate, described as about one reply for every five conversations started.

Against that, the warm side of this account: 3,889 people contacted inside an existing network, 950 of them replied. That is 24.4%, with a median of 0.9 hours to the first reply. Those two numbers, 22.22% and 24.4%, are close enough that nobody should build an argument on the gap between them.

One honest caveat on our figure. Reach sent 180 of the 18,818 outbound messages in that history and the rest were sent by hand, so 24.4% describes what this account does to its own network rather than what the software does on its own. That is the point being made here anyway.

So what does cold outreach actually cost?

The cost is in the funnel, not the message. Take the same report's medians and run them forward. One hundred connection requests produce about 21 accepted connections. That report also gives a median reply-to-acceptance rate of 18.10%, described as turning about one in five accepted connections into a reply. So a hundred requests to strangers produce roughly four conversations.

A hundred messages into people who already accepted you produce about twenty-four, on the numbers above, and there is no invite quota standing in the way. LinkedIn caps weekly invitations. It does not cap messages to people you are already connected to.

So that is the real trade. Cold outreach spends an invite budget and a week of waiting to arrive at a reply rate you could have had immediately with someone who already knows your name.

Does warm outreach vs cold outreach change what you write?

It changes it completely, and this is where warm outreach usually gets thrown away. The advantage is that the person recognises you, and a templated message destroys exactly that advantage.

The opening-line breakdown from this account is blunt about it. One corporate opener starting "I hope you are doing well. We at Linkenite specialize in custom AI solutions" went to 839 people and got a 16.3% reply rate. A four-word one, "Congrats on the new role!", went to 108 people and got 60.2%. Same network, same sender, roughly four times the reply rate for the message that was about them instead of about us.

A smaller slice makes the same case. A 37-person campaign into connections who had been qualified first returned 28 replies. That is 75.7%, on 37 sends, which is far too small a sample to publish as a rate and is included here only because it points the same way as the 839.

How do you run warm outreach without turning it into a template?

This is the part Reach was built for, and it is one feature rather than a philosophy. Reach drafts a personalised nurture touch per person and holds it for your approval before anything goes out. The agent reads what it knows about that specific connection, writes a first touch for them, and then stops. Right now there are 26 of those sitting in this account waiting on a human.

The approval step is not decoration. It is the thing that stops warm outreach collapsing back into a mail merge, because a person reads every message before it sends and the bad ones get killed there. A reply also hard-stops any further automated follow-up for that person, so a real conversation is never interrupted by a scheduled message.

If you want the mechanics of scoring your own connections first, we wrote that up in how to find leads in your existing LinkedIn network. The question of whether personalisation survives being scaled at all is in does personalized LinkedIn outreach at scale still work. Reach itself is at reach.linkenite.com.

Common questions

Is warm outreach always better than cold outreach?

No. If nobody in your network is a plausible buyer, cold outreach is the only option you have, and the funnel cost is just the price of that. Warm outreach wins when you already have connections who fit and have never been contacted, which is the common case for anyone who has been on LinkedIn for a few years.

What is a good reply rate for LinkedIn outreach in 2026?

The 96,051-campaign benchmark published on 17 September 2026 puts the median campaign at 22.22% replies per conversation started, and a 20.75% median acceptance rate on the requests before that. Treat anything above those as above median for cold. Warm messages should beat them comfortably, and if they do not, the message is the problem.

Does LinkedIn limit how many people I can message?

Weekly invitation limits apply to connection requests, which is the cold path. Messages to people you are already connected to are not capped the same way. That asymmetry is most of why warm outreach is cheaper to run.

Can I automate warm outreach without getting my account restricted?

Pacing and behaviour matter more than the label. Reach runs inside your own browser session rather than from a server, holds each message for approval, and enforces weekly send caps. We went through the restriction question in detail in LinkedIn automation without getting banned.

How do I find the warm connections worth messaging?

Import your connections, then score them against a written description of who you sell to, rather than reading 16,000 profiles by hand. In this account 11,168 imported people have never been through that step, which is a fairly typical starting position.

---

Sources

Trending anchor (opened and read, 2026-09-23)

HeyReach, "LinkedIn Outreach Benchmarks 2026: 96,051 Campaigns Analyzed", published 17 September 2026. https://www.heyreach.io/blog/linkedin-outreach-benchmark-report-2026 (HTTP 200, opened in full this run)

Methodology as stated in the report: 96,051 campaigns, filtered down from 103,075 original rows, drawn from the platform's own activity over the last twelve months. No narrower date range given.

Figures used, read directly from the report:

| Figure | Value | The report's own wording | |---|---|---| | Median acceptance rate | 20.75% | "about 1 accepted connection for every 5 requests sent" | | Median reply rate | 22.22% | "about 1 reply for every 5 conversations started" | | Median reply-to-acceptance | 18.10% | "turns about 1 in 5 accepted connections into a reply" |

Funnel arithmetic in the blog is ours, run forward from those two medians: 100 requests → 20.75 accepted → 18.10% of those → 3.76, written as "about four".

Note on the citation: HeyReach is a vendor in the same arena. It is cited here only as the publisher of research, per the house rule that citing a competitor's published research is fine. It is not recommended, not compared against, and not described as struggling.

Considered and not used

• Pin, "AI vs Human Recruiting Outreach: 2026 Data From 5M+ Messages" (https://www.pin.com/blog/ai-vs-human-recruiting-outreach-study/). Opened. Real and well documented (5M+ messages, Jan 2024 to Jun 2026, 1,800+ organisations; AI email 4.97%, hand-written email 12.59%, AI LinkedIn 16.9%). Rejected as the anchor: published 10 July 2026, well outside the ~7-day freshness window, and recruiting rather than B2B sales.
• Expandi's 13.2M-connection-request benchmark and several other 2026 benchmark pages surfaced in search. Not opened, not cited.
• The account-level accept rate of 5.3% from get_results. Deliberately excluded. get_results flags it directly: "acceptances only appear after an inbox sync, and an unsynced mirror produces exactly this number." Connections last synced 287h ago on this run.

First-party (Reach MCP, run 2026-09-23)

get_account_state and get_results, account pravin@linkenite.com:

• Pipeline total 16,069 people. 11,168 have never been qualified.
• Contacted 3,889, replied 950 → reply rate 24.4%, median 0.9 hours to first reply.
• Campaign "Reach — browser-first sellers (v2, sendable rows)": 37 contacted, 28 replied (75.7%). Published with n stated, labelled as too small to support a rate.
• Opening-line breakdown: "I hope you are doing well. We at Linkenite specialize in custom AI solutions" — 839 contacted, 16.3% reply rate. "Congrats on the new role!" — 108 contacted, 60.2% reply rate.
• awaitingHuman.touchProposals: 26.
• Caveat carried verbatim into the blog: "Reach sent 180 of 18818 outbound messages in this history — the rest were sent by hand on LinkedIn."

Feature claim (shipped)

docs/STATE-OF-THE-SYSTEM.md, "Working product":

• "personalized nurture proposals with human approval"
• "inbox sync, replies, reply classification, and automatic pause behavior"
• "network import/sync and company-level canonical people"
• "lead qualification against versioned target audiences and products"

Nothing in the post names a feature outside that list.

Internal links (each opened this run)

• https://reach.linkenite.com — HTTP 200
• https://linkenite.com/blogs/2026-09-10-find-leads-in-your-existing-linkedin-network — loads, H1 "How to find leads in your existing LinkedIn network"
• https://linkenite.com/blogs/2026-09-11-personalized-linkedin-outreach-at-scale — loads, H1 "Does personalized LinkedIn outreach at scale still work?"
• https://linkenite.com/blogs/2026-09-07-linkedin-automation-without-getting-banned — prior story in this series, published