The result
| Length | Reply rate |
|---|---|
| Under 150 | 21.4% |
| 150 to 240 | 29.8% |
| 240 to 420 | 38.6% |
| 420 to 700 | 24.1% |
| Over 700 | 11.2% |
Sample
2024 to 2025, twelve weeks. Reply rate grouped by character count of the first message across mixed personas and company sizes.
Why it works
Both ends of the curve fail for opposite reasons. A very short message cannot contain anything specific, so it cannot prove a human wrote it, and the reader correctly identifies it as a template. A very long one proves a human wrote it and then asks for two minutes of attention that a stranger has not earned.
The middle band is the length at which you can name one specific thing, describe one problem and ask one question, and no more. That is not a coincidence: the winning structure in our copy test came out at almost exactly this length without anyone measuring it.
The practical value of the number is that it is checkable. Character count is the only element of a message you can verify before sending without judgement, which makes it a useful gate on a team where several people write.
How to apply it
Set 240 to 420 as a hard rule and check it on every message before it goes.
Note that LinkedIn caps a connection request note at 300 characters, which sits comfortably inside the band. If your note is at the cap you are in the right place.
If you cannot fit the message into 420 characters, the problem is usually that you are explaining your product. Delete that part rather than negotiating with the limit.
What this test does not tell you
Mixed personas and company sizes, so segment-level variation is hidden. This is grouped observational data rather than a randomised split, which means length correlates with other things: longer messages tend to contain more product detail, and that is separately penalised in the copy test. Both effects push the same way and we have not separated them.
Method
One variable at a time. A test alters a single element; if two things change we learn nothing. Prospects were split at random inside each client account, holding industry, geography, seniority and company size constant on both sides. No test was called below 2,000 prospects per arm, because outbound is noisy at low volume. Reply rate is reported as the leading indicator and meeting rate as the decision, because several variants lifted replies and produced no extra meetings at all, and those were not rolled out.
This test is one of fifteen in a dataset covering 389,890 prospects and 15,018 meetings across 41 client programmes and 17 clients, run between 2018 and 2026. The full set is on the 2026 outbound benchmarks page.
Cite this page as: Prospectio.ai, Test 15: reply rate by first-message character count, B2B Outbound Benchmarks 2026.