Last updated July 2026
Data report · February to July 2026
We analyzed 117,021 LinkedIn messages across 32,770 conversations on the Postelix platform. This report covers what actually moved reply rates: how long the message was, what hour it went out, how many times we followed up, and whether an invite carried a note. Every finding states how it was measured, and every number carries its sample size.
31.9%
reply rate within 7 days for messages under 150 characters, against 13.3% over 400. Short messages get 2.4x the replies of long ones.n=5,925 and n=7,589
16.1%
reply rate at 12:00 UTC, the hour we send the most and the hour that performs worst.n=2,805
24.4%
of the conversations we opened got a reply within 14 days.n=8,867 conversations
Under 150 characters. Messages that short got a 31.9% reply rate within 7 days, which is 2.4 times the 13.3% we saw on messages over 400 characters.
| Message length | Reply rate within 7 days | Messages measured |
|---|---|---|
| 0 to 150 characters | 31.9% | n=5,925 |
| 151 to 400 characters | 26.9% | n=7,415 |
| 401+ characters | 13.3% | n=7,589 |
Method: outbound messages from 1 February to 21 July 2026, bucketed by character count. The window ends 21 July so every message gets a full 7 days before the 28 July cutoff. Each reply is credited to the most recent outbound before it, so no reply is counted twice.
How much to trust this: this is correlation, not causation. Shortening a bad message will not produce a 31.9% reply rate. Long messages in this dataset skew pitch-heavy, so the length bucket is partly measuring what kind of message it is rather than how many characters it has. The honest reading is that long messages tend to be the ones carrying a pitch, and those get ignored.
12:00 UTC. It is both the hour we send the most messages and the hour with the lowest reply rate at 16.1%, while 09:00 UTC is the strongest hour at 30.6%.
The pattern is the interesting part. The busiest sending hour is the weakest performing one, which is what you would expect if everyone else is sending then too. Reply rate dips through 11:00 and 12:00 and recovers across the afternoon.
| Hour (UTC) | Reply rate | Messages measured |
|---|---|---|
| 06:00 UTC | 20.9% | n=206 |
| 07:00 UTC | 22.8% | n=746 |
| 08:00 UTC | 28.5% | n=1,157 |
| 09:00 UTC | 30.6% | n=1,212 |
| 10:00 UTC | 24.9% | n=1,184 |
| 11:00 UTC | 20.5% | n=1,378 |
| 12:00 UTC | 16.1% | n=2,805 |
| 13:00 UTC | 24.3% | n=1,652 |
| 14:00 UTC | 23.6% | n=1,793 |
| 15:00 UTC | 24.2% | n=1,511 |
| 16:00 UTC | 24.1% | n=1,314 |
| 17:00 UTC | 26.2% | n=1,287 |
| 18:00 UTC | 24.0% | n=1,106 |
| 19:00 UTC | 25.5% | n=1,070 |
| 20:00 UTC | 23.2% | n=1,005 |
| 21:00 UTC | 26.5% | n=529 |
| 22:00 UTC | 22.8% | n=359 |
Method: the same messages and the same reply attribution, grouped by hour of send in UTC. Only hours with at least 200 messages are shown.
How much to trust this: our sending is mostly from European timezones and recipients sit in mixed ones, so an hour in UTC is not the same local moment for everyone receiving a message. Read this as a pattern in our sending, not as a universal clock.
24.4% of the conversations we initiated received a reply within 14 days, across 8,867 conversations.
That number is well above the benchmarks usually quoted for cold outreach, and we want to be precise about why, because the reason is not the writing.
These are not cold messages in the spray sense. The motion behind this dataset is signal-based: the engine surfaces people who are showing buying behaviour, we connect first, and the opener goes out after the connection is accepted. So the recipient has already seen the name, and the timing is chosen rather than arbitrary. A 24.4% reply rate is what happens when the selection is good, not when the script is clever.
If you take one thing from this section, take the ordering. Who you message and when you message them sets the ceiling. The message decides how much of that ceiling you reach.
Method: conversations opened from 1 February to 14 July 2026, counted as replied if any inbound arrived within 14 days. The window ends 14 July so every conversation gets a full 14 days before the 28 July cutoff. n=8,867.
Two. The second message converts nearly as well as the first for the conversations that get it, 15.3% against 20.3%, and by the fourth touch the yield has fallen to 4.5%.
The table below reads down the funnel. At each touch, eligible is the number of conversations that received that message and had not replied yet, so each row is a fair denominator for the message actually sent rather than a share of the total.
| Touch | Conversations eligible | Replied | Marginal yield |
|---|---|---|---|
| Touch 1 | 8,867 | 1,801 | 20.3% |
| Touch 2 | 1,956 | 299 | 15.3% |
| Touch 3 | 923 | 58 | 6.3% |
| Touch 4 | 133 | 6 | 4.5% |
Method: eligible at touch k is the conversations that received a k-th message with no reply yet. Replies are capped at 14 days from the conversation opening and credited to the touch they followed.
What we take from it: send the second message. A 15.3% yield on conversations that had gone quiet is close enough to the 20.3% first-touch rate that skipping it leaves real replies on the table. The third touch is defensible at 6.3%. The fourth, at 4.5% on 133 conversations, is where we would stop.
This sits awkwardly with the common advice to follow up five or seven times before giving up. We are not going to tell you that advice is wrong everywhere, because it comes from email motions with much larger and more replaceable lists. On this channel, in this dataset, the fourth touch produced 6 replies. Decide accordingly.
We cannot compare directly, because 100% of the invites we tracked were sent without a note, and they were accepted 37.1% of the time, 865 of 2,331.
That is a deliberate choice in the motion rather than an experiment, so treat it as evidence that a note is not required rather than proof that notes hurt. A 37.1% acceptance rate on noteless invites is high enough that the burden of proof sits with the note.
The mechanism we would suggest is simple. A note turns a connection request into a pitch, and the recipient has to decide about your offer instead of about you. Without one, the only question is whether the profile is worth connecting to.
Method: tracked invitations sent inside the window and their acceptance status. 865 accepted of 2,331 sent, every one without a note.
How much to trust this: these targets were signal-selected, so they are more likely than a random list to recognise why you might be connecting. A 37.1% acceptance rate on a cold, unselected list is not what we are claiming.
These eight phrases are the strongest markers of an automated pitch in the cold messages our users receive, and we treat them as things never to open with.
Method: n-gram mining across cold pitches our users receive, compared against a background of human messages. Greeting and name patterns were removed so the list reflects pitch language rather than ordinary openings.
Read the list closely and the pattern is that every one of them is about the sender's calendar. They are the language of asking for time before giving a reason, which is why they cluster in automated pitches. The phrases are a symptom. The thing to fix is opening with a request.
Whether messages that read as human-written actually get more replies than messages that read as machine-written.
Every draft in our system passes a judge that decides whether the message reads as machine-written or human-written, and it has produced 12,168 verdicts so far. What we have not had until now is the link between that verdict and what happened next, so we are instrumenting verdict-to-outcome linkage. We are not going to guess at the answer in the meantime, and we are not publishing a number we cannot stand behind. The next edition of this report will answer it with a real sample size attached.
Every figure on this page is aggregated, anonymized platform data. No message content is read by humans or published here, and no individual account or person is identifiable in any number above.
In our data, 24.4% of conversations we initiated got a reply within 14 days, across 8,867 conversations. That is a useful reference point only if your motion looks like ours: signal-based targeting, a connection request first, and an opener after acceptance. The number reflects who gets messaged more than what the message says. For individual messages the range is wide, from 31.9% on messages under 150 characters down to 13.3% on messages over 400.
Under 150 characters, based on our data. Messages in that bucket got a 31.9% reply rate within 7 days across 5,925 messages, against 26.9% for 151 to 400 characters and 13.3% for 401 characters and up. Short messages got 2.4 times the replies of long ones. This is correlation rather than causation: long messages in our dataset skew pitch-heavy, so length is partly standing in for what kind of message it is. Shortening a pitch does not make it welcome.
09:00 UTC was the strongest hour in our data, at 30.6% across 1,212 messages. The clearer finding is what to avoid: 12:00 UTC is both our highest-volume sending hour and our lowest performing at 16.1% across 2,805 messages. Our sending is mostly from European timezones and recipients sit in mixed ones, so treat this as a pattern in our sending rather than a universal clock.
We cannot answer that with a direct comparison, because 100% of the invites we tracked were sent without a note. What we can say is that noteless invites were accepted 37.1% of the time, 865 of 2,331. That is high enough to suggest a note is not required for this motion. Our targets are signal-selected, so a cold unselected list would not be expected to match it.
Two, on this evidence. Measuring each touch against the conversations that actually received it, the first message replied at 20.3% of 8,867 conversations, the second at 15.3% of 1,956, the third at 6.3% of 923, and the fourth at 4.5% of 133. The second message converts nearly as well as the first, so it is worth sending. By the fourth the yield is marginal and you are spending goodwill in a market you may need to return to. Advice to follow up five or seven times generally comes from email motions with much larger and more replaceable lists.
Postelix is the motion behind this dataset: 100 to 400 tracked accounts, surfaced when they show buying behaviour, warmed up before any DM, written in your voice, with Auto, Review or Off on every channel.
Start now, €149/moLast updated July 2026. Data cutoff 28 July 2026. Message-level window 1 February to 21 July 2026, conversation-level window 1 February to 14 July 2026. Aggregates only, no message content published. 14-day money-back guarantee. Cancel anytime. Postelix is a brand and service of Unfair Advantage Ltd, Meleti 2C, 8570 Pegeyia, Cyprus.