Four rates from the engine that runs our own pipeline and our clients' accounts every working day, next to the three public LinkedIn reports of 2026 with the largest samples. Every figure carries the count it came from and the window it was measured over. Recomputed 9 September 2026 from the engine database, across 17 active sending profiles, all workspaces counted together.
| Rate | Re:Vault, last 30 days | Re:Vault, last 90 days | Published |
|---|---|---|---|
| Invitation accept rate | 22.0% 338 of 1,533 invitations |
21.3% 406 of 1,904 invitations |
28.5% Expandi, 13.2 million invitations, May 2025 to April 2026, published 19 May 2026 26% Belkins, 15.1 million touches, updated 29 June 2026 26% SmartReach, 1,800+ campaigns and 500,000+ invitations, January to June 2026, range 19 to 29% |
| Reply rate after the accept | 11.0% 34 of 308 accepted |
11.5% 41 of 355 accepted |
18% SmartReach calls this the healthy level, above 28% strong 7.2% Belkins average, 4.2 to 10.5% by industry |
| Platform refusals per invitation attempt | 4.3% 75 of 1,762 attempts |
3.4% 75 of 2,220 attempts |
The three reports measure accepts and replies. This column fills in when one of them publishes refusals. |
| Time from accept to the first message | median 45 h, 90th percentile 9.5 days 284 people |
median 39 h, 90th percentile 7.7 days 331 people |
Same as above. SmartReach measures the interval between follow-ups (3 to 5 days) and finds the second follow-up worth 4 percentage points of reply; the first message after the accept is measured here. |
A figure with fewer than 100 people behind it reads collecting on this page until the sample is there. The refusal count covers every workspace on the stack, including the profiles where we test mechanics before they reach a real person's account.
A rate is a claim. A rate with its numerator, its denominator and its window is a measurement.
Invitation accept rate. The people we sent an invitation to, dated by the first invitation to that person, inside the window and at least 3 days old, because the 95th percentile of accepts arrives inside 5 days and a younger invitation would pull the rate down for no reason. Accepted means the connection is recorded in our base, from the provider's new-connection webhook or from a poll of the pending invitations, restricted to people we invited ourselves. One definition, the same one as on the benchmarks page.
Reply rate after the accept. The same people, whose accept fell inside the window and is at least 7 days old. Replied means an inbound LinkedIn message from that person after the accept. Replies to the invitation itself are counted here too, since they arrive as messages once the connection exists.
Platform refusals per invitation attempt. Every attempt the sending layer made in the window: an invitation that went through, or one the platform answered with a temporary refusal. The share is refusals over all attempts. A refusal pauses that profile for 30 to 60 minutes and the queue moves to the next person, which is why the same row keeps sending the same day.
Time from accept to the first message. For people who accepted inside the window and have since received a first message, the hours from the accept to that message, median and 90th percentile. A fix to the loop that sends the first message went in on the night of 8 to 9 September 2026, so this row is expected to move at the next recompute.
Two of the three find the same thing about invitation notes: a note costs a few points of accept and adds to reply. Our rule matches it. A note goes out from premium profiles only, and on basic profiles the fact about the person moves into the first message after the accept, where the platform counts nothing.
A script reads the engine database, writes the counts to data.json next to this page and puts them into the text above. It runs with the weekly review, and the date at the top is the date of the last run. The definitions in the script and the definitions on this page are the same text.
Connect your Claude to the same stack. The code arrives by email within a minute.