LinkedIn outreach response rate benchmarks, and why they all disagree

Published LinkedIn outreach benchmarks range from about 2% to about 20% reply rate, and almost all of them are honest. They disagree because they count different things. A reply rate measured against connection requests sent is a different number from one measured against connections accepted, which is different again from one measured against messages delivered. The largest study with a disclosed method, Expandi's analysis of 13,218,869 connection requests sent through its platform between May 2025 and April 2026, reports a 28.5% acceptance rate, a 3.0% connection-note reply rate, and a 10.4% message reply rate. Those three numbers describe the same campaigns. If you quote one of them without saying which, you have written a misleading sentence.

What is a good LinkedIn outreach response rate?

The question has no answer in the abstract, and the reason is worth understanding before you read another benchmark page.

Outreach on LinkedIn is a funnel with at least five stages, and a “response rate” can be quoted against any of them:

1. Connection requests sent.
2. Connection requests accepted.
3. Messages sent to accepted connections.
4. Replies of any kind.
5. Positive replies, meaning a meeting or a real question.

Replies divided by requests sent, and replies divided by messages sent, differ by roughly the acceptance rate. At a 28.5% acceptance rate that is a factor of about 3.5. So the same campaign can be described as a 10% reply rate or a 3% reply rate, both accurately, and the difference between a page that sounds encouraging and a page that sounds bleak is which line the author divided by.

There is a sixth stage nobody measures: people who saw your message and did nothing but now recognise your name. It has real value and no reporting line, which is one reason outreach-versus-content arguments never resolve.

Why does every published LinkedIn benchmark disagree?

We pulled the most-cited sources and put their methods side by side. The pattern is not that some are wrong. It is that they describe different populations doing different work.

Read the last column rather than the third. A 45% acceptance rate from an agency that researches every prospect and a 28.5% acceptance rate from a platform where software sends the requests are not in tension. They are a measurement of what research is worth, which is roughly 16 percentage points of acceptance and about 10 points of reply rate in this comparison.

The LinkedIn recruiter data is the odd one out and should not be applied to sales at all. Candidates have a reason to reply to recruiters that buyers do not have to reply to sellers. We still cite it because the length and bulk findings are directionally useful, and because it is one of the few large samples with a stated method that is not owned by an automation vendor.

What does the 13.2 million request study actually say?

It is the largest disclosed dataset in the category and we are not going to try to out-scale it. What is worth doing is reading past the headline.

Expandi's breakdowns by industry are wide. Staffing and recruiting sits at 36.5% acceptance, 6.6% connection-note reply, and 18.9% message reply. Computer software sits at 27.5% acceptance, 2.8% note reply, and 8.8% message reply. Venture capital and private equity sits at 34.9% acceptance and 11.0% message reply. If you sell software, the platform average of 10.4% is not your benchmark. Your industry's 8.8% is closer, and that is a 20% difference from the headline.

The trend line is the part most summaries drop: the report says connection-note reply rates fell from 3.5% to 2.2% over the twelve months it covers. That is a decline of roughly a third in a year, in the population that adopts new tactics fastest. Whatever the number is today, the useful information is that it is going down.

The null result is the most useful finding

Buried in the same report: acceptance rates across every sender seniority bucket land within a three point band, from 29.4% for C-level down to 26.3% for junior and entry level. Across every company size bucket, acceptance runs 26 to 29% and message reply runs 9 to 11%.

That is a null result and it contradicts a lot of received wisdom. Being a C-level sender at a large recognisable company buys you about three percentage points of acceptance. It does not buy you a different game. Which means the levers that people spend money on, seniority signalling, company logos, title inflation, are not where the variance lives. The variance lives in the list and the relevance.

What are the denominators, exactly?

Here is the same hypothetical campaign reported five ways. Every number below is arithmetic from one stated set of assumptions, not a measurement, and we are labelling it that way because this is exactly where most benchmark pages quietly stop labelling.

Assumptions, all illustrative: 400 connection requests sent in a month, a 30% acceptance rate, a message sent to every accepted connection, a 12% reply rate on those messages, and 40% of replies being positive.

Same campaign. It has a 12% reply rate and a 3.6% reply rate and a 1.4% conversion rate, all simultaneously. A vendor wanting to look good quotes 12%. A competitor wanting to make the channel look bad quotes 1.4%. Neither is lying.

When you read a benchmark, the first question is which row it is quoting. If the page does not say, you cannot use the number, and most pages do not say.

Which benchmark should you actually use?

Match the population before you match the number.

• If software sends your requests: the Expandi platform figures are the closest available comparison, and you should use your industry row rather than the headline.
• If a human researches each prospect: platform averages will understate you badly. Agency-reported figures are the nearer comparison, with the caveat that agencies publish their wins.
• If you are sending InMail:LinkedIn's own 10-25% band is the only first-party figure, and it comes from the party selling credits. Treat the top of the band as marketing.
• If you are recruiting: the LinkedIn Talent Blog data is genuinely about your situation, with the caveat that the window closed in 2022.
• If your outreach follows a warm-up: nothing published describes you. This is an unmeasured category and you will have to build your own baseline. The method is in our piece on outreach that does not feel cold.

What does LinkedIn itself publish about response rates?

Very little, and what exists is worth reading with the source in mind.

LinkedIn's Sales Solutions guide states that InMails have a “10-25% hit rate when it comes to soliciting a response from potential clients, 300% higher than emails with the exact same content.” There is no sample size, no date, and no method attached, and the page exists to sell Sales Navigator. A 15 point band is also not a benchmark, it is a range wide enough to contain almost any result.

The recruiter analysis on LinkedIn's Talent Blog is much better documented: tens of millions of InMails from corporate recruiters, May 2021 to April 2022, with responses counted within 30 days. Its findings are relative rather than absolute, which is actually more useful. Shortest InMails, under 400 characters, responded to 22% above average. Longest, over 1,200 characters, 11% below. Individually sent InMails ran roughly 15% above bulk sends.

Relative findings survive population differences better than absolute ones. “Shorter does better” is more likely to transfer to your situation than any specific percentage.

What can post engagement data tell you about outreach?

Nothing directly, and we want to be explicit about that because we have a data set and it would be easy to imply otherwise.

Our corpus is 12,988 English LinkedIn posts from 65 creators with 1,000+ followers, scored by engagement rate. It has reactions, comments, follower counts, and the visible hook text. It has no invitations, no messages, no acceptances, and no replies. We cannot produce a DM benchmark and we are not going to estimate one.

What it can benchmark is the other side of the channel. Across the cohort the median post earns 0.40 engagement points per 1,000 followers on our formula, the 75th percentile is 1.27, and the 90th percentile is 5.95. In absolute terms the median post gets 122 reactions and 12 comments, while the top decile median is 235 reactions and 72 comments. The comment gap is the interesting one: 72 against 5 in the bottom half, a far wider spread than reactions.

Comments are named people starting conversations with you, publicly and unprompted. They are the closest thing in our data to a reply, and they arrive without a credit, a template, or a connection request. That comparison is not rigorous and we are not presenting it as one, but it is the reason we think content and outreach should be budgeted against each other rather than treated as separate departments. Our engagement benchmarks by follower count break down what good looks like at your size, and the engagement rate calculator does the arithmetic.

Why does InMail make the arithmetic worse?

InMail has its own accounting quirk that quietly distorts comparisons, and almost no benchmark page mentions it.

LinkedIn returns your InMail credit when a recipient accepts the InMail. So the cost of an InMail campaign is not credits sent multiplied by price per credit; it is credits consumed net of returns. A well-targeted campaign is partially self-funding and a badly targeted one is not, which means the effective cost per message varies with the very thing you are trying to measure.

Credits also expire after 90 days and cannot be transferred between Premium subscriptions, per LinkedIn's help page. Allowances are 5 a month on Premium Career, 15 on Premium Business, and 50 on Sales Navigator Core. That caps InMail volume at a level far below what automated connection outreach reaches, which is why comparing an InMail response rate against a connection campaign reply rate is comparing 50 messages a month against several hundred.

The practical consequence: an InMail benchmark is drawn from a much smaller, much more deliberately chosen population of messages. You would expect it to look better than a connection-request benchmark even if the messages were identical, purely because scarcity forces selection. We go through whether the credits justify the subscription in our decision tree on Sales Navigator.

Why do outreach benchmarks go stale so fast?

Faster than almost any other marketing benchmark, for three reasons that compound.

• Tactic adoption. Outreach effectiveness falls as a tactic spreads. The Expandi data shows connection-note reply rates falling from 3.5% to 2.2% inside a single twelve month window. A benchmark from 2023 is describing a different equilibrium, not a slightly older version of the same one.
• Platform changes. Invitation limits, note allowances, and inbox surfaces all change without announcement, and each change moves the denominators. The free-account note allowance is currently documented inconsistently by LinkedIn itself, at three per month on one page and five on another.
• Recipient fatigue. The population being measured is learning. People who received four templated messages a week in 2022 now receive many more, and their threshold for replying has moved accordingly.

A useful rule: an outreach benchmark older than about eighteen months tells you about the structure of the funnel, not about the rates. The structure is stable. The rates are not.

How should you read a benchmark page?

Six questions. If a page cannot answer four of them, the number on it is decoration.

1. What is the denominator? Replies divided by what. This alone explains most of the disagreement in the category.
2. How large is the sample, and of what?Messages, accounts, or campaigns. “Thousands of campaigns” is not a sample size.
3. Over what window? A rate averaged across three years hides a trend that might be the most important thing in the dataset.
4. Who sent the messages? Software, an agency, or the people being described. These are three different populations with different economics.
5. Who is publishing, and what do they sell? Not disqualifying. Most useful data in this category comes from vendors because only vendors have volume. But it changes which direction the errors point.
6. What counts as a reply? An automatic out-of-office, a one word no, and a meeting booked are all replies under most definitions.

The two studies we lean on hardest here answer five of the six between them, which is why they appear in the table above rather than the dozen other pages that quote a number with no method attached.

What about LinkedIn versus email comparisons?

These are the least reliable numbers in the category and they get quoted constantly.

LinkedIn's own Sales Solutions page claims InMail response is “300% higher than emails with the exact same content.” No sample, no method, and the publisher sells InMail credits. Agency pages routinely pair a LinkedIn reply rate from their own campaigns against an industry-average cold email reply rate from a third party, which compares a hand-worked channel against a mass-sent one and attributes the gap to the platform.

A fair comparison would hold constant the list, the offer, the level of personalisation, and the definition of a reply. We have not seen one published. Until someone does, the honest position is that LinkedIn and email are different surfaces with different costs and different norms, and that the channel comparison is far less important than whether the message was worth sending. That is the argument we make in why templated messages underperform.

How do you build your own benchmark?

Two weeks and a spreadsheet. This is more valuable than any number in this article because it holds your list, your offer, and your reputation constant.

1. Fix one approach and send 100 requests. Same target profile, same message structure, same time of day. Do not improve anything mid-run.
2. Record four counts, not one. Sent, accepted, messaged, replied. Then split replied into positive, neutral, and negative.
3. Wait 14 days before calculating. Acceptances trickle in for a surprisingly long time and an early read makes everything look worse than it is.
4. Write down the four rates with their denominators named.Literally write “replies divided by messages sent” next to the number, so future you does not make the mistake this whole article is about.
5. Change one variable and run 100 more. The list, or the message, or the warm-up. Not all three.
6. Track hours. Replies per hour spent is the only fair comparison between researched outreach and automated outreach, and it is the number that decides which model fits your business.

Sample size, honestly

100 requests at a 30% acceptance rate gives you 30 messages and perhaps 3 replies. Three replies is not a measurement of anything. You can read acceptance rate off 100 requests with reasonable confidence; you cannot read reply rate off 30 messages. Either run more, or accept that your reply rate number is a direction rather than a rate. Most published benchmark comparisons in this niche fail on exactly this and never say so.

What should you actually expect?

Combining everything above, and labelling clearly what is measured and what is not:

• Acceptance rate. Measured. Roughly 26 to 30% for automated outreach across industries and company sizes, higher for staffing and recruiting, and materially higher for hand-researched lists.
• Reply rate on messages. Measured. Around 9 to 11% for automated outreach, with software sitting below the average and recruiting well above it.
• Reply rate on connection notes. Measured and falling. 3.0% in the latest window, down from 3.5% to 2.2% across the year.
• Effect of a warm-up sequence. Not measured by anyone we could find. Our expectation is that it helps, and that expectation is not evidence.
• Effect of publishing on later reply rates.Not measured. We have post-level data and it cannot answer this. Nor can anyone else's.

Two of those five are honest gaps. We would rather leave them open than fill them with a number, since filling gaps with plausible numbers is how this category got to a 2% to 20% spread with no way to tell which figure applies to you.

The short version on LinkedIn outreach benchmarks

Every published LinkedIn outreach response rate is defensible and none of them is yours. The spread comes from denominators, populations, and message types, not from some campaigns being ten times better than others. Use the Expandi industry rows if software sends your messages, use agency figures if a human researches each prospect, discount LinkedIn's own InMail band because of who publishes it, and build a baseline from 100 of your own requests with the four counts recorded separately. Then compare yourself to yourself.

For the message side of this, see why LinkedIn DM templates stop working and what to use instead. For the platform limits that cap the whole channel, see what LinkedIn actually documents about connection request limits. And for the content side, where the numbers we do own live, start with our analysis of 34,012 LinkedIn posts.