Key finding
Across 180,155 matured LinkedIn connection requests, acceptance peaked at 32.0% for accounts averaging 10 to 19 invites a day and fell to 26.7% for accounts averaging 20 to 29, the band that carried 89.6% of the 180,155 matured requests measured.
What we measured
LinkedIn limits how many connection invitations an account can send, and the practical ceiling sits at roughly 25 a day. Almost every discussion of outreach volume is really a discussion about how close to that ceiling to run.
This study groups every matured connection request in the dataset by the average daily invite volume of the sending account it came from, then compares acceptance and reply rates across those groups.
It also reports what a real sending day looks like, which turns out to matter more than the rate comparison does.
Methodology
Dataset window. Every figure covers connection requests sent between 8 January 2025 and 6 August 2026, pulled on 6 August 2026.
Maturity. Rates are computed on requests sent at least 30 days before the pull, so recent requests are not counted as failures before they have had a chance to be accepted. That is the same cohort of 180,155 requests the rest of this research series uses, which is what makes these rates directly comparable to it. One robustness check below is computed on the unrestricted all-sent cohort and says so where it appears.
How accounts were bucketed. For each sending account, invites were counted per calendar day, and the account's average across its active days determines its band. Every request that account sent is attributed to that band. This describes an account's habitual operating volume, not what it did on the specific day a given request went out.
Coverage is complete. The three bands contain 2,042, 16,789 and 161,324 requests, totalling 180,155, which is every request in the matured cohort. None are unaccounted for.
That completeness is deliberate. Bands are derived from the request records themselves rather than from a separate daily-usage table, because only part of the account population carries the key needed to join the two. Joining would have silently dropped a large share of accounts and produced a breakdown that looked precise while covering a fraction of the data.
Acceptance rate is accepted requests divided by requests sent, within each band. Reply rate of accepted is connections producing an inbound message divided by accepted connections, within each band.
Day bucketing is UTC.
Active days only. An account's average is computed across days it actually sent something, so weekends and pauses do not drag the average down.
The findings
| Average invites/day | Sent | Accepted | Acceptance | Reply (of accepted) |
|---|---|---|---|---|
| Under 10/day | 2,042 | 465 | 22.77% | 23.66% |
| 10 to 19/day | 16,789 | 5,374 | 32.01% | 27.50% |
| 20 to 29/day | 161,324 | 42,997 | 26.65% | 27.60% |
| 30+/day | none |
There is no 30+ band because no account in the dataset averaged 30 or more invites a day. LinkedIn's cap is the reason.
The shape is the finding: a peak in the middle, not a slope.
The relationship is not a straight line, and lower volume is not uniformly better. Acceptance rises from 22.77% in the lowest band to a peak of 32.01% in the middle, then falls to 26.65% in the highest. The lowest-volume band performed worst of the three. The penalty this study is named for appears between the middle band and the top of the allowed range, a gap of 5.36 points (derived from the table above), not across the whole range.
The 20 to 29 band is the largest of the three and the second-worst of the three. It carries 161,324 of the 180,155 matured requests, 89.55% of them. The peak band is the smallest of the two large ones: 16,789 requests, 9.32% of the cohort. The lowest band is 2,042, or 1.13%. All three shares are derived from the table above, and the 9.32% figure is worth holding onto, because the headline peak rests on it.
Reply rate among accepted connections is essentially flat at 27.50% and 27.60% across the two large bands. Whatever volume affects, it affects who accepts, not what happens afterwards.
What a real sending day looks like
The rate table describes a choice almost nobody makes. Across 9,136 active account-days:
| Metric | Value |
|---|---|
| Mean invites per active day | 22.42 |
| Median | 25 |
| 90th percentile | 25 |
| Maximum | 328 |
| Days at 25 or more invites | 6,899 |
| Share of active days at 25 or more | 75.51% |
Three quarters of all active sending days ran at 25 invites or more. The median day and the 90th percentile day are the same number, which is what a population pressed against a ceiling looks like.
Does it survive removing the biggest customer?
The largest customer workspace accounts for 21.97% of all requests. Excluding it entirely, acceptance by band runs 22.76%, 30.97% and 26.99%. These three figures are computed on the unrestricted all-sent cohort rather than the matured one, so their levels are not directly comparable to the table above; what they establish is the shape. The peak stays in the same band, the lowest band stays worst, and the gap between the middle and top bands is 3.98 points rather than 5.36 (derived from the three figures above).
What this means
There is a cost to running at the ceiling, and it is smaller than the phrase "volume tax" suggests: 5.36 percentage points of acceptance between the peak band and the top band, or 3.98 with the largest customer removed (both derived).
What makes it interesting is not the size but the concentration. Three quarters of sending days sit at or above 25 invites, because the platform's cap acts as a target rather than a limit. A number published as a maximum gets treated as a setting.
The lowest band complicates any simple reading. If less were straightforwardly better, accounts averaging under 10 a day would lead the table, and they finish last by a wide margin. That band is small at 2,042 requests, and the accounts in it are unusual by definition, but it is enough to say the data does not support "send fewer invitations and acceptance goes up" as a general rule. What it supports is a peak in the middle.
The flat reply rate is the cleanest result here. Volume moves acceptance and leaves what happens after acceptance alone, which means the two are worth measuring separately.
In practice
An SDR sending 25 connection invites a day is operating in the largest of the three bands measured here and the second-highest performing on acceptance, 5.36 points below the middle band and 3.88 points above the lowest (both derived from the findings table). They are also doing what most accounts in this dataset do: 75.51% of active sending days ran at 25 or more.
A team lead reading acceptance rate as a measure of message quality is reading a number that varies with sending volume. The bands in this dataset differ by several points, so an acceptance rate carries information about an account's operating volume as well as about its copy. What this study cannot do is tell any individual account which way it would move, because the accounts in each band differ in ways it does not measure.
Anyone treating the platform cap as a target is looking at the specific behaviour this dataset is saturated with. The median active day and the 90th percentile active day are both 25 invites, so there is almost no variation left above the median to observe.
Limitations
The lowest band is small. 2,042 requests, roughly 1% of the cohort. Its 22.77% acceptance rate is the least reliable figure in the table and should not carry much weight.
The range is narrow by construction. LinkedIn's cap means this compares roughly 10 a day against roughly 25 a day. It is not a test of extreme volume, and nothing here describes what happens at 50 or 100 a day, because no account in the dataset did that.
Band assignment uses an account average, not the day the request was sent. An account averaging 22 a day that sent 8 on a particular Tuesday has those 8 requests counted in the 20 to 29 band. The bucketing describes operating habit rather than the conditions of an individual send.
One account-day recorded 328 invites, far above the platform cap and far above the 90th percentile of 25. It is an outlier of unknown cause, and the maximum should not be read as representative.
Volume may be downstream of performance rather than upstream of it. An account whose acceptance is weak may be told to send more to compensate, which would produce this association with the causation running backwards. Nothing in this data distinguishes the two directions.
Volume is entangled with everything else. Accounts that run at the ceiling differ from accounts that do not in targeting, offer, industry and the seniority of the people they contact. This study measures an association between volume and acceptance. It does not isolate volume as the cause, and no randomisation was involved.
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%. The check above addresses the first of those, not the rest.