Key finding
Between 2025 and 2026 LinkedIn connection acceptance rose slightly from 26.3% to 28.1%, while the share of accepted connections that produced a reply fell from 32.2% to 22.0%. Getting connected became no harder. Getting a conversation became considerably harder.
What we measured
"LinkedIn outreach is getting worse" is a claim made constantly and rarely with a denominator attached. This study tests it against 180,155 connection requests split by the year they were sent, measuring both acceptance and reply rate on the same population.
The result is not a general decline. It is a divergence between two metrics that are usually reported as if they moved together.
Methodology
Dataset window. Every figure covers connection requests sent between 8 January 2025 and 6 August 2026, pulled on 6 August 2026.
Cohorts are defined by send date, not by when an acceptance or reply arrived. A request sent in December 2025 that was accepted in January 2026 belongs to 2025. Attributing outcomes to the year they landed rather than the year the request went out would smear each year into the next.
Both years are matured. Only requests sent at least 30 days before the pull are included. This matters more than it sounds: comparing a fully mature 2025 against a partly mature 2026 would manufacture a decline out of nothing but the calendar.
Acceptance rate is accepted requests divided by requests sent within the cohort. Reply rate of accepted is connections producing at least one inbound message divided by accepted connections within the cohort.
A reply means a reply. No sentiment is recorded, so a decline counts.
The findings
| Year sent | Requests | Acceptance | Reply (of accepted) |
|---|---|---|---|
| 2025 | 101,191 | 26.33% | 32.19% |
| 2026 | 78,964 | 28.11% | 21.98% |
Acceptance rose by 1.78 percentage points. Reply rate among accepted connections fell by 10.21 points, a relative decline of roughly 32% (both derived from the table above).
Acceptance is a share of requests sent; reply is a share of accepted connections. The two bars in each pair are directly comparable to each other, and the pairs are not comparable across the gap.
The 2026 cohort is smaller, and not because volume fell. Counting every request rather than only matured ones, 2025 sent 101,191 and 2026 sent 103,656 in eight months. The 2025 figure is identical to its matured count above because every 2025 request has long since matured. Sending volume rose. The matured 2026 cohort is smaller because 2026 is a partial year and its final 30 days are excluded by the maturity rule.
Does it survive dropping the unreliable months?
The four opening months of 2025 are small and unstable, and this study tells readers below not to trust them. That makes the 2025 baseline a fair target: it contains those months. Recomputing with them removed:
| Cohort | Requests | Acceptance | Reply (of accepted) |
|---|---|---|---|
| 2025, May to December only | 94,010 | 25.95% | 30.07% |
| 2026 | 78,964 | 28.11% | 21.98% |
The reply gap narrows from 10.21 points to 8.09 (derived from the two tables above). The finding survives, and the honest version of it is the smaller number.
The monthly series
Published in full for transparency. The study does not rest on it, and no single month here should be quoted.
| Month sent | Requests | Acceptance | Reply (of accepted) |
|---|---|---|---|
| 2025-01 | 1,182 | 22.08% | 97.70% |
| 2025-02 | 3,256 | 32.22% | 52.53% |
| 2025-03 | 2,065 | 34.38% | 37.46% |
| 2025-04 | 678 | 33.92% | 73.91% |
| 2025-05 | 5,596 | 23.50% | 25.93% |
| 2025-06 | 6,183 | 14.52% | 28.06% |
| 2025-07 | 4,980 | 25.98% | 39.26% |
| 2025-08 | 7,359 | 23.36% | 32.81% |
| 2025-09 | 13,559 | 25.38% | 28.92% |
| 2025-10 | 18,651 | 29.60% | 33.93% |
| 2025-11 | 20,907 | 29.27% | 28.44% |
| 2025-12 | 16,775 | 24.35% | 26.00% |
| 2026-01 | 8,859 | 30.80% | 23.27% |
| 2026-02 | 13,068 | 26.28% | 23.85% |
| 2026-03 | 15,257 | 30.76% | 20.26% |
| 2026-04 | 11,883 | 27.75% | 25.42% |
| 2026-05 | 11,932 | 28.42% | 19.85% |
| 2026-06 | 14,278 | 25.02% | 19.82% |
| 2026-07 | 23,617 | 23.47% | 23.51% |
| 2026-08 | 4,762 | 13.71% | 18.38% |
These are all-sent figures, so they include immature requests and their levels sit below the matured cohorts above.
The first four months are tiny and unstable. January 2025 shows a 97.70% reply rate on 1,182 requests, which is a small-sample artefact and not a fact about 2025. August 2026 is a partial and immature month.
Monthly acceptance ranges from 13.71% to 34.38%. Those two extremes are not the same kind of month: the high is March 2025 at 2,065 requests, one of the smallest in the series, while the low is the partial current month.
Read from May 2025 onward, the reply column drops from the high twenties and thirties into the low twenties and stays there through 2026, while the acceptance column moves around inside a band without trending. That is the same story the annual tables tell, which is the reason to trust them rather than any month inside them.
What this means
Acceptance and reply rate are routinely reported together as though they measure the same underlying thing. Over this window they moved in opposite directions, which means they do not.
Acceptance measures a low-cost decision. Someone glances at a request and taps accept. Nothing in this data suggests that decision got harder to win; if anything it got marginally easier.
Reply measures an expensive decision. Someone reads a message and chooses to spend time on it. That is where the change happened, and a roughly one third relative decline in a year is a large change, even taking the more conservative 8.09 point version.
The obvious explanation is volume: more automated outreach arriving in the same inboxes, with recipients still accepting connections freely but answering less of what follows. Sending volume in this dataset did rise year over year, which is consistent with that story without establishing it. What the data does establish is that anyone tracking outreach health on acceptance rate alone would have seen nothing happen over this period.
In practice
A sales leader whose dashboard reports acceptance rate would have seen that number improve slightly across these two years. The metric that moved is one layer down, and a team could have lost a substantial share of its conversations per connection while its headline number went up.
An SDR comparing their current reply rate against a benchmark published in 2025 is comparing against a number this data no longer supports. Reply rate among accepted connections was 32.19% for requests sent in 2025 and 21.98% for requests sent in 2026. Which of those a rep is measured against is a choice someone made rather than a fact about their performance.
Anyone drawing a conclusion from a single month of their own data is working with the kind of sample the monthly table above shows to be unstable. Acceptance across individual months in this dataset ranges from 13.71% to 34.38%, and those two extremes arise for different reasons: the high is one of the smallest months in the series, the low is a month too recent to have matured.
Limitations
This is one customer base measured twice, not a panel. The customers, industries, offers and targeting in the 2026 cohort are not the same mix as in 2025. A shift in who was sending, or who they were sending to, would produce this pattern without anything changing on LinkedIn's side. The data cannot separate a platform-wide effect from a change in composition, and that is the single biggest caveat on this study.
Two years is two data points. The annual comparison is the defensible cut of this data, and it is also a comparison between exactly two numbers. It establishes that a change happened over this window in this dataset. It does not establish a trend.
The years are not seasonally balanced. 2025 contributes twelve months and 2026 contributes seven to the matured cohort, so any seasonal pattern in reply behaviour falls unevenly between them.
Replies are counted, not classified. If the mix of replies shifted between years, for example toward shorter declines, this measure would not show it.
Maturity is matched but not identical. Both cohorts exclude the final 30 days before the pull, so 2025 requests have had far longer to accumulate late replies than 2026 requests have. Some part of the reply gap is late 2026 replies that have not arrived yet. The 30 day rule limits this but does not eliminate it, and the true gap is therefore smaller than the figures above by an amount this data cannot quantify.
The dataset is concentrated. The largest customer workspace accounts for 21.97% of all requests sent, the largest five for 57.66%, and the largest ten for 76.19%. A change in behaviour at one large customer between the two years would move these numbers on its own.