In short
- 1"Acceptance runs 18.7 percent in Belkins' own funnel of 14,077 contact records and 26 percent across the 11.5M Expandi requests published inside the same 2026 study, against 30 to 37 percent for well-run B2B outbound (LeadRiver, April 2026). Most of that spread is survivorship, not skill."
- 2"Replies run 7.9 percent cold and 12.2 percent to an existing connection (Belkins 2026). Expandi's 2026 benchmarks put platform-wide message reply at 10.4 percent across 6.7 million messages."
- 3"A connection note lowers acceptance (25.3 vs 27.6 percent) and raises reply among people who accept (8.2 vs 5.3 percent), per Belkins 2026. Optimising acceptance alone tells you to delete the note."
- 4"Meeting rate now has published numbers and they disagree: 1.3 percent of connected prospects (Belkins 2026) against 5 to 18 per 1,000 requests (LeadRiver, April 2026)."
- 5"Positive reply rate still has no verified public benchmark. Measure your own baseline and watch the trend."
Open almost any outreach dashboard and the biggest number on the screen is the one that means least: messages sent. It is the easiest number to move and the only one that tells you nothing about whether the work is producing revenue.
Four metrics actually carry information, and they fail in different places, which is what makes them useful. Acceptance rate tells you about targeting. Reply rate tells you about the message. Positive reply rate tells you whether you are talking to buyers. Meeting rate tells you whether any of it reaches a calendar.
But there is a problem underneath all four, and no benchmark article states it. Published LinkedIn reply rates in 2026 range from a measured 7.9 percent to the 50 percent quoted on pages that name no sample at all, depending on whose page you read. That is not a difference in skill. It is a difference in what got counted. Below, each metric with a healthy range, a named source and a date, then the arithmetic that makes those sources comparable, and finally the one metric that improves as your pipeline gets worse.
Volume is an input, not a metric
Volume belongs on the cost side of the ledger, not the results side. One LinkedIn account can send roughly 100 connection invitations a week by industry consensus, so past that point volume stops being a lever. LinkedIn publishes no official figure. With a fixed top of funnel, only conversion improves.
That matters for how you read every other number. If the top of your funnel is fixed, the only way to get more meetings is to improve the conversion at each step, or to add another sender. No tool raises the platform ceiling, whatever a pricing page implies, and any product promising unlimited sends is promising something the platform itself does not allow. The detail on how those limits behave is in our guide to LinkedIn connection request limits.
BeReach paces deliberately below that line. Two of its stops are unconditional: 50 connection invitations a day per account, and profile visits held under 120 an hour. The others are base figures that a workspace multiplier scales, 300 profile visits and 70 messages a day, with messages hard-capped at 100 whatever the multiplier says. All of them are ceilings rather than targets, and the platform's weekly invitation limit binds long before any of them do.
The four metrics worth reporting
Acceptance, reply, positive reply and meeting rate form a chain, and each one only means something against the step above it. Three now have published 2026 benchmarks from datasets in the millions. Positive reply rate does not, and anyone quoting a precise industry figure for it is guessing.
Two things stand out. First, the published acceptance figures run from 18.7 to 37 percent, and they are not four independent readings. The 15.1 million touchpoints in the Belkins study are Expandi partner data, credited as such on the page, so the 18.7 and 26 percent rows are one study measuring two different books, and the 28.5 percent row is the same provider a year later. Only LeadRiver is a genuinely separate sample. That should make you suspicious of any single number presented as the industry average. Second, the positive reply row is empty on purpose. No dataset we could verify publishes a benchmark for it, and inventing one would be worse than leaving the cell blank.
Why published benchmarks disagree by more than five times
Because they measure different things under the same word. Published LinkedIn reply rates in 2026 run from a measured 7.9 percent to the 30 to 50 percent advertised on pages with no dataset behind them, and almost none of that spread is performance. It is denominator choice and survivorship: measured datasets count every contact, vendor pages count their best campaigns.
Line the definitions up and the disagreement mostly evaporates.
The bottom row is the one to watch. Several pages that rank for this query publish reply-rate benchmarks of 15 to 25 percent for cold sequences and 30 to 50 percent overall, with no dataset, no sample size and no methodology attached. Measured against the two studies that do publish their working, and noting that the large set inside the Belkins study is Expandi partner data rather than a second opinion, the real cold figure is 7.9 to 10.4 percent. If you benchmark a healthy campaign against an unsourced 30 percent, you will conclude it is broken and rewrite a message that was fine.
There is a second reason to distrust a fixed target: benchmarks decay. Expandi's twelve-month series shows reply to the connection request itself falling from 3.5 percent in May 2025 to 2.2 percent in April 2026, a 37 percent relative decline in a single year, while message reply held steady at 10.4 percent. A 2024 benchmark is not a 2026 benchmark. Your own trend line over the last quarter is a better yardstick than anyone's published average.
Acceptance measures targeting, and a note trades it away
Acceptance rate is accepted invitations divided by invitations sent, and it answers one question: are you writing to the right people. It is also the easiest of the four to game, because the fastest way to raise it is to remove the note that makes replies more likely.
That is not a hypothetical. Belkins' 2026 study found requests sent without a note connected at 27.6 percent, against 25.3 percent for requests carrying a personalised note. On acceptance alone, the note loses. But among people who did accept, those who received a note replied at 8.2 percent against 5.3 percent for those who did not, more than half again as likely to start a conversation.
A connection note is a filter, not an amplifier. It costs you about two points of acceptance and buys you three points of reply from a self-selected audience. A team paid on acceptance rate will delete the note, watch the metric rise, and quietly halve the conversations. This is the clearest example on the page of a number improving while the business gets worse.
This also resolves the apparent contradiction between the two big datasets. LeadRiver's April 2026 review of more than 50,000 requests reports 50 to 60 percent acceptance on notes anchored to something the recipient actually posted, against 15 to 25 percent for generic or blank requests, and says first-name personalisation on its own is table stakes that produces no meaningful uplift over generic. Belkins measures notes in aggregate, most of which are the first-name kind. Read together, the rule is narrower than "personalise your note": a note that carries a real trigger wins, and a note that only carries your pitch performs worse than no note at all.
A trigger is a new role, a funding round, an open job posting, a post they wrote last week. That is the whole thesis of signal-based selling: the reason your message is not spam should be visible before it is opened. Because acceptance also varies enormously by industry and seniority, we aggregated five published studies with their sample sizes in what a good acceptance rate actually is, so you can weight the one closest to your market.
Reply rate measures the message, and warmth moves it most
Reply rate is replies divided by messages that reached someone who could answer. Belkins' 2026 study of 15.1 million touchpoints found 7.9 percent on a cold connect-then-message sequence against 12.2 percent to an existing connection. Expandi's 2026 benchmarks put the platform-wide message reply rate at 10.4 percent.
Read those two Belkins numbers together and you get the most actionable fact on this page: the same message sent to a warmed prospect replies at about half again the cold rate, at no extra send cost. Warming does not mean waiting. It means the person has seen you before the ask arrives, through a comment, a profile visit, a reaction to their post. Practical sequencing for that is covered in how to improve your LinkedIn message response rate.
One caution on comparing vendor claims. If a tool advertises a 25 percent reply rate, ask what it counted as a reply and what it counted as a send. Auto-replies, out-of-office bounces and "please remove me" all register as replies in most systems. That is why the next metric exists.
Positive reply rate is the first number that is genuinely yours
Positive reply rate is the share of replies that move a conversation forward: a question, a request for detail, a referral. It excludes rejections, out-of-office and unsubscribes. No public dataset we could verify publishes a benchmark for it, so it is only meaningful against your own recent baseline.
Measure it anyway, because it is the first metric that separates a working funnel from a busy one. A rising reply rate with a falling positive share means you are provoking answers rather than interest, usually because the message is engineered for a response instead of for a fit. A high positive share on a small reply count is a much better position than the reverse: it means the targeting is right and you simply need more of the same people.
Classify manually for the first month. Three buckets, positive, neutral, negative, decided by a human reading the thread. Automating the classification before you know what a good reply looks like in your market just bakes in a guess.
Meeting rate now has two published numbers, and they disagree sevenfold
Two 2026 datasets now publish a meeting benchmark, and they disagree by up to sevenfold. Belkins reports 1.3 percent of connected prospects book a meeting. LeadRiver reports 5 to 18 meetings per 1,000 connection requests. At a consensus pace that is roughly one to eight meetings a month per account.
Here is that arithmetic in full, run twice. Each column stays inside one book of campaigns, the Belkins column on Belkins' own 14,077 records and the LeadRiver column on LeadRiver's own 50,000+ requests, because mixing an acceptance rate from one sample with a meeting rate from another is exactly the error this page is arguing against.
One to eight meetings a month, from one account worked at the platform ceiling. That range is the honest answer, and its width is the point: anyone quoting you a single meeting-rate benchmark is quoting one vendor's campaign mix, not a law of the platform. The Belkins column is what an agency's mixed book of campaigns actually produces. The LeadRiver column is what tightly targeted campaigns produce, measured on campaigns that were already running well.
Either way, wanting twice the pipeline next month means a second sender, not a second setting, which is why cost per meeting is a more useful tooling question than cost per seat. If you want to run this arithmetic with your own numbers rather than flattering defaults, the what your outreach volume is worth uses the published benchmarks above as its floor.
What actually moves the numbers, ranked by effect size
Rank the levers by how many points they actually move and the popular advice inverts. Who you write to swings reply rate by ten points in Expandi's 2026 data. Whether you used an AI prompt or picked the best weekday moves it by less than one point in Belkins' study.
Acceptance behaves the same way. Expandi's industry breakdown runs from 17.5 percent in consumer electronics to 40.1 percent in broadcast media, though Expandi flags its top two industries, broadcast media and civil engineering at 39.9 percent, as too small a sample to be reliable. Stay inside the well-sampled industries and the top is staffing and recruiting at 36.5 percent, still a 19 point spread that no amount of copy editing will close.
The practical reading is uncomfortable for anyone selling message optimisation, including the AI kind. The top two rows are list decisions, made before a single word is written. The bottom two rows are the ones the internet writes about most, and together they are worth less than a single point. An AI that writes your message is worth 0.3 points. An AI that finds the right person to write to is worth ten. Spend your effort accordingly.
The metrics to stop reporting
Some numbers move when you work and stay flat when you succeed. They are not useless, they are just not results, and putting them at the top of a dashboard quietly redefines the job as activity. These six take the most space for the least information.
- Messages sent. An input with a fixed ceiling. Reporting it rewards the one behaviour the platform punishes.
- Profile views received. Correlates with your own visiting activity more than with buyer interest.
- Total connections. A stock, not a flow. It rises even in a month where you book nothing.
- Post impressions. Useful for content strategy, unrelated to whether outreach converts.
- Credits or actions consumed. A billing number wearing a performance costume.
- Social Selling Index. A LinkedIn engagement score, not a pipeline measure. Nothing published ties it to meetings booked.
The test is simple, and it is the one to apply before anything reaches a leadership deck. If the number can go up in a week where you booked zero meetings and closed zero deals, it is a diagnostic at best, and it does not belong on the first screen. By that test, acceptance rate is a diagnostic too, which is why it should never be reported without the reply rate sitting next to it.
Reading the funnel when a number is bad
Each metric fails for a specific reason, which is the point of tracking them separately. Diagnose from the top, because a weak number early makes every number below it meaningless. The table maps the most common symptom to the cause worth checking first, before you rewrite a single message.
Notice that only the last row is solved by doing more. Every other row is solved by changing what you do, which is the argument for measuring conversion at each step instead of watching one blended number.
Every viral post is 100+ warm conversations waiting.
Tell your agent who you want to reach. It finds them, says which ones are worth your time, writes the first line, and follows up.
Frequently asked questions
What is a good LinkedIn connection acceptance rate?
It depends which sample you benchmark against, and they differ a lot. Belkins' 2026 study reports 18.7 percent in its own book of 14,077 contact records, and 26 percent across the 11.5 million Expandi partner requests published in the same study. Expandi's 2026 benchmarks report 28.5 percent platform-wide across 13.2 million requests. LeadRiver's April 2026 review of 50,000+ requests puts well-run B2B outbound at 30 to 37 percent. Clearing 30 percent is healthy by any of them, and below 20 percent points at your list rather than your wording.
What reply rate should I expect on LinkedIn?
Belkins' 2026 study reports 7.9 percent on a cold connect-then-message sequence and 12.2 percent when messaging an existing connection. Expandi's 2026 benchmarks put platform-wide message reply at 10.4 percent across 6.7 million messages. Treat any figure above 20 percent as a definition question first: pages advertising 30 to 50 percent reply rates generally publish no dataset, no sample size and no denominator.
Should I send a note with the connection request?
Only if it carries a real trigger. Belkins' 2026 data shows notes lower acceptance slightly, 25.3 percent against 27.6 percent without, while raising reply among those who accept from 5.3 to 8.2 percent. LeadRiver's April 2026 data shows notes anchored to something the person recently posted accept at 50 to 60 percent, against 15 to 25 percent for generic ones. A note with a trigger wins. A note that only carries your pitch is worse than no note.
How many meetings can one LinkedIn account book in a month?
Roughly one to eight, worked at the consensus ceiling of about 100 invitations a week. Applying Belkins' 2026 rates, 430 invitations produce about 80 connections and about one meeting at their 1.3 percent connected-to-meeting rate. Applying LeadRiver's April 2026 range of 5 to 18 meetings per 1,000 requests gives about two to eight. The width of that range is why a second sender, not a new setting, is what doubles pipeline.
How do I work out cost per meeting?
Divide monthly tooling and sender cost by meetings booked from outreach. BeReach Pro is 99 euros a month billed monthly, or 948 euros a year, about 79 euros a month billed yearly, priced per workspace rather than per seat, with two connected accounts. Sales Navigator Core is 119.99 dollars a month billed monthly, or 1,079.88 dollars a year, per LinkedIn's compare-plans page checked in August 2026. At one to eight meetings a month per account, the tooling line is small next to the sender's time either way.
Can I measure any of these numbers before I start sending?
No, not the numbers themselves. Acceptance, reply and meeting rate are all defined against invitations or messages that actually went out, so none of the three exists until a real campaign has run; they are lagging indicators by definition. What does pay off before that point is the targeting work underneath them: the industry breakdown in the levers table above is worth ten points of reply rate on its own, so time spent narrowing the list before you send is rarely wasted.
Measure the four that matter, with BeReach. BeReach builds the prospect list from public data, finds the people showing a real trigger, checks them against your ideal customer profile and drafts the message. It paces sends well inside the platform's limits, and every message is approved by you before it sends. See how BeReach runs your outreach.
Reading this in an AI assistant? Hand it the page and let it summarize, so you can ask follow-up questions against the whole argument rather than the part you have read so far.
