There are no published limits. LinkedIn does not document the per-day caps on connections, views, messages, or InMails. Every number you see on the internet — including the ones below — is a fingerprint from real accounts running real workloads. Treat all of it as probabilistic, not authoritative.

What follows is a working model assembled from operating multiple accounts at multiple volumes through 2025 and 2026. It is not a promise. It is a place to start that we have found does not get accounts restricted in the first 90 days, which is when LinkedIn does most of its cleanup.

The four budgets

LinkedIn meters at least four things separately. Each has its own threshold and its own recovery curve.

1. Connection requests

The most public limit. Common safe range for an established account (12+ months old, >500 connections, complete profile):

15–25 connection requests per day on a normal week. Sustainable indefinitely.
40–50 per day in a burst for a few days, then back off. Past 50 consistently, restrictions become likely within weeks.
100+ per day triggers the “you’ve been sending too many” warning rapidly, sometimes within 24 hours.

The hidden ceiling is the rolling weekly cap. Above ~100 requests in any 7-day window, withdrawal rates climb. Above ~150, restrictions are common.

2. Profile views

Less metered than people think. Public LinkedIn profile views are essentially unlimited for normal browsing patterns. Where you get into trouble is automation that views the same accounts repeatedly, or that views thousands of profiles per day without corresponding connect/message activity.

50–100 views per day is the “active recruiter” range and is uncontroversial.
500+ views per day from a clean account begins to look like scraping if it’s sustained. LinkedIn’s commercial-use limit will silently kick in (search results truncate to ~3 from ~10).

3. Messages (1st-degree connections)

You can message your connections at a much higher rate than you can connect to strangers. Typical safe range:

50–100 outbound messages per day to 1st-degree connections, well-paced, generally fine.
Identical-template messages are the actual risk — duplicate-message detection runs at the conversation level. Vary the openers; the bulk of the body can be templated.

4. InMail (paid)

Capped by your Sales Navigator/Recruiter seat (typically 50/month, refilled by response). The risk here isn’t restriction — it’s wasting an InMail credit on someone who would have answered a free message. InMail conversion rates have steadily declined since 2022; we’d rather see InMail used only when the prospect isn’t open to messages.

The soft throttles

Before the explicit warning page, LinkedIn applies invisible signals you can read:

Connection requests start hanging at “Pending” for days without ever being seen. Acceptance rate quietly drops below 10%.
Search results truncate. A query that used to return ~10 first-page results returns ~3. (This is the commercial-use limit; it lasts ~7 days.)
The “people you may know” carousel goes empty. Surface signal that LinkedIn is restricting the recommendation engine for your account.
Your own profile views per week drops sharply even though you’ve been more active, not less. LinkedIn de-prioritizes you in the algorithm.

When two or more of these fire at once, the right move is to back off all outbound for 7 days, not to push through.

Patterns that draw attention faster than volume

Volume matters, but the pattern matters more. Things that get accounts flagged at lower volumes than the numbers above:

Bursts. 50 connection requests in 30 minutes, even once, looks like a script. The same 50 spread over a workday looks like a busy person.
Late-night sends. If you’re in San Francisco and your account is sending connection requests at 3:00 AM Pacific seven days in a row, the pattern is obvious. Even cheap human-pace jitter can’t cover for a 24/7 schedule.
Outreach to accounts with no network overlap. Targeting profiles in countries or industries you have zero 1st- or 2nd-degree presence in trips the “this is not how this user normally behaves” signal.
Identical messages at scale. See above.

What we set as defaults in Marno

Marno’s safe defaults sit well below the upper edges above:

~20 connection requests/day, with 90-second human-pace jitter between actions.
Sequences pause automatically at the first session-kill challenge.
Per-account dedicated residential proxy, so the egress IP doesn’t shift mid-week.
Conservative working-hours window enforced per timezone.

You can tune any of these up. We’d encourage you to tune them down before you tune them up.