LinkedIn outreach limits in 2026 (invites, messages, and the soft ban)

LinkedIn documents some of its outreach limits precisely and withholds the one everybody asks about. The commercial use limit is published, resets at midnight Pacific time on the 1st of each calendar month, and cannot be lifted on request. The weekly invitation cap is real, account-specific and unnumbered — the product tells you that you have hit it and nothing more. Sorting the documented numbers from the folklore is most of the work here, because the throttle that follows a bad week arrives with no notification at all.

Three classes of limit, and only one of them is documented

Before any number, the question that decides whether you can trust it: where did it come from? Every LinkedIn limit in circulation belongs to one of three evidence classes, and treating them as equivalent is why so much outreach advice is confidently wrong.

Class one: published in LinkedIn's own Help documentation. A named article states the rule, and often states the reset behaviour too. These are the numbers you can plan against, because LinkedIn has committed to them in writing and changes them visibly.

Class two: observable in the product but unnumbered. LinkedIn tells you that a limit exists — usually by refusing an action and naming the limit in the refusal — without disclosing its value. The weekly invitation cap is the canonical example. You can prove the limit exists in about ten seconds; you cannot learn its value from any source LinkedIn controls.

Class three: folklore. A number that appears in blog posts, automation-vendor documentation and forum threads, with no primary source anywhere in the chain. Sometimes it is a reasonable inference from many operators' experience. Sometimes it is one person's guess that got copied four hundred times. From the outside these look identical, which is exactly the problem.

The useful discipline is to label every number you rely on with its class before you build a process around it. A class-one number can anchor a plan. A class-three number can inform a default, provided you say out loud that it is a default rather than a rule.

What LinkedIn actually publishes about your limits

Short list, because it genuinely is short.

The commercial use limit. LinkedIn's Help documentation states that free monthly usage "resets at midnight PST on the 1st of each calendar month". It names what counts toward it: searching for profiles on LinkedIn.com and mobile, browsing profiles through the People Also Viewed module, and viewing member profiles on the People tab of LinkedIn Pages. It also states two things people routinely get wrong: "We are not able to display the exact number of searches or views you have left", and "we also cannot lift the limit upon request". So this limit is documented in behaviour and undocumented in magnitude, and support cannot help you.

Event invitations. LinkedIn's Help page on inviting connections to an Event does state a number: "You can invite 1000 connections per week. If you organize multiple events, this limit applies across events and does not increase." That is a genuine class-one figure, and it is worth noticing precisely because it demonstrates that LinkedIn publishes limits when it wants to. The silence around connection invitations is a choice, not an oversight.

The prohibition on automation. LinkedIn's User Agreement, effective 3 November 2025, is explicit at section 8.2. Members agree not to "Use bots or other unauthorized automated methods to access the Services, add or download contacts, send or redirect messages, create, comment on, like, share, or re-share posts, or otherwise drive inauthentic engagement", and separately not to "Develop, support or use software, devices, scripts, robots or any other means or processes (such as crawlers, browser plugins and add-ons or any other technology) to scrape or copy the Services". The same section prohibits creating a false identity or using another person's account.

Notice what the automation clause targets. It is not phrased around volume. It is phrased around the method and around inauthentic engagement, which is why pacing alone never moves you back inside the agreement.

Why the weekly invitation cap has no number

LinkedIn confirms the cap exists by refusing the action and naming it. It does not publish the value, and the reason is structural rather than coy.

An account-specific, dynamic allowance is far harder to game than a fixed one. If the number were published, every automation tool would run at exactly that number minus one, permanently, and the limit would stop doing any work. Keeping it undisclosed and variable means the ceiling is one more signal in the system rather than a target to optimise against.

That has three consequences for how you should plan.

A published figure from a vendor is a report about that vendor's accounts, not about yours. The widely repeated hundred-per-week is the median of a self-selected population of automated senders, which is close to the worst possible sampling frame for estimating what your specific account may do.

Your allowance is not fixed even for you. A new account, an account with a low acceptance rate and an account with ten years of organic history are not subject to the same treatment, and the same account's headroom moves as its signals move.

Hitting the cap is data, not a scheduling inconvenience. Reaching a limit you did not know you were near means your pacing was set by folklore rather than by measurement, and the sane response is to halve the rate and instrument the account rather than to wait for the window to roll.

The deeper point is that invitation volume is a proxy for the thing LinkedIn is actually measuring, which is whether people want to hear from you. Acceptance rate, reply rate, profile completeness and how often recipients click the "I do not know this person" option are the inputs. Volume is just the axis along which those inputs get sampled. The full ban surface and the ramp that keeps an account inside it covers the behavioural side of that in detail.

The soft ban is a throttle, and it never announces itself

Soft ban is an operator term. LinkedIn does not use it, does not confirm it, and there is no page in your settings that reports it. What people are describing is a state in which every action still appears to succeed while the reach behind it is quietly reduced.

The reason it is hard to detect is that nothing fails. Invitations send. Messages deliver. The interface is identical. What changes is downstream and invisible from your side: fewer of your invitations reach a person who is inclined to accept, your profile surfaces less often in other people's search results, and the passive inbound that a healthy profile generates dries up.

Three practical consequences follow, and the first is the one that does the damage.

The natural misreading is that your copy got worse. Acceptance falls, so you rewrite the note, and the rewrite does nothing, so you rewrite it again — while continuing to send at the volume that caused the throttle. That escalation is the single most common way a recoverable throttle becomes a restriction.

There is no confirmation available. Support cannot tell you, because from LinkedIn's side there is nothing to tell. You are asking about a ranking decision, not an account state.

Recovery has no published timetable. Anyone quoting a specific number of days is quoting folklore. What is observable is that reducing activity and generating genuine engagement precedes recovery in most operator accounts, and that continuing at volume does not.

Enforcement above this level does announce itself. A temporary restriction produces an interstitial on login and usually clears through identity verification. A permanent restriction ends the account, and the connection graph inside it is not portable to anywhere else. The asymmetry between those outcomes and one extra week of pipeline is the entire argument for pacing conservatively.

How to detect a throttle from your own data in a week

This is the part that is genuinely actionable, because it turns an unfalsifiable feeling into a measurement. The whole method rests on one distinction: a copy problem is local and gradual, a throttle is global and abrupt.

Step 1 — cohort your acceptance rate by send date, not by accept date. Take the invitations you sent in a given week and measure what fraction of those had been accepted fourteen days later. Measuring accepts-this-week against sends-this-week mixes two populations and smears exactly the step change you are trying to see. This period-versus-cohort error is the reason most teams cannot tell throttles from noise.

Step 2 — keep one segment deliberately unchanged. Pick a segment whose targeting and note you will not touch for a month. It is your control. When acceptance falls in the segment you have been rewriting and falls identically in the one you have not, the copy is not the variable.

Step 3 — record profile views weekly. Profile views are the closest available proxy for whether LinkedIn is still surfacing you. A throttle usually drags views down alongside acceptance; a weak note does not touch views at all, because views happen before anyone reads it.

Step 4 — check your public visibility from outside the session. Open your own profile in a logged-out private window and search for your name and headline the way a stranger would. This does not prove anything on its own, but a profile that has stopped appearing publicly is worth knowing about before you spend another fortnight on subject lines.

Step 5 — read the step, not the level. Plot the three series by week. A throttle looks like a cliff on one date across all series and all segments. A targeting or copy problem looks like a slope inside one segment. If you cannot see which shape you have, you do not yet have enough weeks of data to act, and sending harder while you find out is the worst available option.

Two honest caveats about this procedure. It cannot distinguish a throttle from a market change that happened to hit every segment in the same week, and it produces a hypothesis rather than a verdict, because the system you are measuring will not confirm anything. It is still far better than rewriting copy on a hunch.

What resets, and on which clock

Three different clocks run at once, and conflating them produces most of the bad arithmetic in LinkedIn outreach.

A calendar-month clock. The commercial use limit resets at midnight Pacific time on the 1st, per LinkedIn's own documentation. It does not prorate and it does not carry over, so the last week of a month is systematically tighter than the first if you have been searching heavily.

A rolling window. Invitation allowances behave as a window rather than a calendar reset, which is why burning a week's worth on Monday does not buy you a fresh allocation on the following Monday. Autocloz accounts for invites over a trailing seven-day window rather than a Monday-to-Sunday week for exactly this reason, which is a deliberate design choice rather than a reading of LinkedIn's implementation.

A per-message credit cycle. InMail allowances accrue monthly with their own accumulation and expiry behaviour, and a credit returns under LinkedIn's published response rules. That is a genuinely different currency from an invitation, and the credit economics of InMail against a connection request works through what each one actually costs.

One more clock worth planning around: LinkedIn states that withdrawing an invitation blocks you from re-inviting that person for a period measured in weeks. Bulk-withdrawing a backlog therefore removes a slice of your list from reach for most of a month, which is an argument for withdrawing continuously rather than in sweeps.

The exposure is contractual, and the hiQ litigation shows what that means

No United States statute prohibits automating LinkedIn. That fact gets misread constantly, so it is worth walking the case that settled it.

hiQ Labs scraped public LinkedIn profiles. LinkedIn sent a cease-and-desist and blocked it; hiQ sued. The district court granted hiQ a preliminary injunction in 2017 at 273 F. Supp. 3d 1099. The Ninth Circuit affirmed on 9 September 2019 at 938 F.3d 985, holding that scraping publicly available data likely does not constitute access "without authorization" under the Computer Fraud and Abuse Act. The Supreme Court vacated and remanded on 14 June 2021 for reconsideration in light of Van Buren v. United States, and on 18 April 2022 the Ninth Circuit reached the same conclusion again at 31 F.4th 1180.

Then the part everyone forgets. In November 2022 the district court ruled that hiQ had breached LinkedIn's User Agreement, and the parties settled the following month. hiQ won the criminal-statute question and lost the contract question.

The lesson transfers directly to outreach automation. "It is not illegal" and "it is permitted" are different claims, and only the first one has case law behind it. The remedy for breaching a user agreement is not a prosecution — it is the loss of the account, the network inside it, the message history and the inbound that flows from it, with no notice period and no appeal you can rely on. Anyone selling you LinkedIn automation who describes that risk as zero is not describing the same platform. That asymmetry is also why LinkedIn should be one lane rather than the whole road, and the four adjudicators a multichannel programme has to satisfy sets out how the other channels differ.

Product defaults are not platform policy

This distinction matters enough to state on its own, because software that paces your LinkedIn activity will show you numbers, and it is easy to read those numbers as if they came from LinkedIn.

They did not. Every ceiling in every LinkedIn automation tool is that vendor's guess at a safe operating point inside an undocumented system. Autocloz ships a set of them: a default weekly invitation ceiling of 80, a per-day invite default of 20, day-one dispatch floored at 5 actions growing on a smoothstep curve toward the daily target over 21 days, weekend allowance dampened to 30% of the weekday figure, a refusal to dispatch when five actions have already landed inside ten minutes, and 60 to 300 seconds of random jitter added on top of the configured minimum gap so the inter-send histogram is not periodic. Each refusal names itself in the deferral reason, so linkedin_weekly_invite_cap and linkedin_pending_invite_cap appear in the log rather than a generic skip.

None of those is a LinkedIn-published limit, because LinkedIn publishes none. They are conservative defaults chosen to sit well inside an unknown ceiling, and the weekly 80 in particular exists to leave headroom for the same human's organic activity on the same account. The pacing and safety layer under the LinkedIn channel exposes a per-sender health read — pending invitations against the hard cap, seven-day and thirty-day acceptance rates against the floor, today's action count against today's warmup ceiling, and whether the sender has been auto-throttled and why — which is the alarm the platform will not build for you.

One detail worth internalising: LinkedIn automation ships switched off in the default automation ruleset, alongside automated calling. Email, SMS and WhatsApp ship on. That ordering is a statement about which channels punish a mistake hardest.

Autocloz's free plan covers 5 users and 10 mailboxes, and adding a LinkedIn sender does not add a seat charge — start free and instrument the account before you scale it.

What no pacing plan can do, and what Autocloz does not do here

Several limits, stated plainly, because a post about limits that hides its own is not worth much.

Pacing reduces the probability of enforcement. It does not remove it, and it cannot, because the behaviour that pacing produces is still automated access under section 8.2. Any tool that sends invitations or messages for you operates against the User Agreement regardless of how carefully it spaces the actions.

None of the numbers here is a LinkedIn commitment. The published items — the commercial use limit's reset, the thousand-per-week event invitation cap, the User Agreement's text — are quotable and dated. Everything about connection invitations is inference, and it can change on a Tuesday with no announcement.

The detection procedure produces a hypothesis rather than a diagnosis. LinkedIn will not confirm a throttle, so the best available evidence is a step change in your own series, and a step change has other possible causes.

Autocloz specifically. It does not sell LinkedIn data, and profile access runs through a third-party session provider rather than an official LinkedIn API for outreach actions, which is set out on the session provider integration page — the dependency is real and worth understanding before you build on it. Invitation withdrawal is currently an operator-triggered endpoint that updates the local ledger, and the code's own note records that the provider-side rescind is a forthcoming worker rather than a shipped daily job, so treat automatic withdrawal as a policy you still enforce by hand. There is no read of your actual LinkedIn allowance, because LinkedIn exposes none; the accounting is of what this system sent, which diverges from LinkedIn's own count whenever you also act from the app. And the queue that collects accepted connections and their replies alongside every other channel does not make the account any safer — it only makes the consequences visible sooner.