A restricted LinkedIn account arrives quietly. The profile still loads, the inbox still works, and the only visible evidence is a queue of invitations that stopped moving. We run cold LinkedIn outreach from a set of sending profiles every working day, and between 27 July and 28 August 2026 those profiles sent 1,495 invitations. Over the same period the platform refused 74 of our attempts. This is what those refusals turned out to mean, how the refusal rate went from 6.1% of attempts to 0.4% inside eight days, and the four pacing rules that did it. Every number has its denominator written next to it, and each rule is something you can apply to your own sending by Monday.
A refusal is the warning, and it comes with a date
When LinkedIn declines an invitation, the answer comes back as a temporary refusal with a cooldown attached. It is the cheapest warning you will ever get, and it is easy to throw away, because to whoever is watching the queue it looks like one invitation that went missing.
The 74 refusals cluster into one bad run: 14 on 12 August, 13 on 13 August, 18 on 14 August, then 8, 7, 7 and 5 across 17 to 20 August, then a single one on 27 August and one on 28 August. Split at the point where we changed how we send, the picture is blunt.
- Through 20 August: 1,117 invitations sent, 72 refused, so 6.1% of attempts came back refused.
- From 21 August: 451 sent, 2 refused, 0.4% of attempts.
Same profiles, same audience, roughly the same daily volume. Two changes to how we pace the sending explain the whole gap.
What a refusal should cost you
Our first version of this rule was expensive. A refusal stopped the profile for the rest of the day, and three refused days in a row stopped it for a week. The reasoning sounded careful: if the platform says no, back away and stay away.
Then we read the journal. Across 23 days that contained at least one refusal, invitations from the same profile went out successfully later the same day on 18 of them. The refusal was a short cooldown on that specific attempt, and our rule was converting it into a lost day. Worse, the profile with a lost day looked to our own counters exactly like a healthy profile with an empty queue, so the loss was invisible until someone counted.
The rule we run now costs one attempt. A refusal pauses that profile for 30 to 60 minutes, and the next attempt goes to a different person. A profile stops for the day only after three consecutive refusals with zero successful sends that day, and the old ladder of a quiet day, then a half limit, then a week off starts from there. An individual person who refuses three times gets closed out permanently, because the provider's own documentation says the refusal code covers "an invitation to this person is already pending or was recently pending", and after a withdrawal the platform holds that person for about three weeks.
The transferable part: read the body of the provider's error, and write down what it actually says. "HTTP 422" with the body thrown away cost us a manual investigation. The same string with the body kept answers the question in ten seconds.
The personalized invitation budget is a hard number
The second change is the one that moved the number.
The provider documentation we work against is explicit: a basic LinkedIn account gets roughly five personalized invitations per month, and roughly 150 plain invitations per week. A premium account carries a much larger personalized allowance. So on a basic account, the note attached to a connection request is a budget line, and a small one.
Our own 21-day measurement across all sending profiles:
- basic accounts, invitation carrying a note: 31 went through, 61 refused, so 66% of attempts refused
- basic accounts, invitation with no note: 627 went through, 2 refused, 0.3%
- premium accounts: 309 invitations, zero refusals
A profile that loses two attempts out of three is repeating a request the platform has already declined, several times a day, every working day. That repetition is what a restriction responds to, and it happens whatever the note says.
So the note now goes out only from profiles that have the allowance for it, and on every other profile the sending layer strips it before the request leaves. The personalization moves one step later, into the first message after the connection is accepted, where it has more room and a reader who already said yes.
One mechanical detail is worth stealing, because it kept the problem alive for weeks. Our budget counter for personalized invitations counted successful sends. A profile that was being refused because of the note therefore never spent any budget, stayed comfortably under its cap, and attached the note again on the next attempt. A counter that scores only successes will report a broken profile as healthy for as long as it stays broken. Count attempts.
Bursts are the signature the platform reads
The third rule is about the shape of the day, and it is the one you can audit in your own data this afternoon.
Until 11 August our sending was spaced only by a few seconds of jitter. Jitter randomises the gap between two consecutive sends and leaves the shape of the whole day alone, so the day keeps the only shape a queue has by default: everything up front. The measurement, taken across the busiest hour of each sending day:
- before 11 August: the busiest hour carried 55.6% of that day's invitations, averaged over 9 sending days, at 17.6 invitations per active hour
- from 12 August: the busiest hour carried 23.1%, averaged over 13 sending days, at 5.8 per active hour
The current rule is one line of arithmetic. The interval between two cold touches equals the remaining working window divided by the remaining daily plan, recalculated on every tick, with a hard floor of one cold touch per profile per minute. A profile with a 12 invitation plan and a nine hour window sends roughly one every 45 minutes, and it slows down on its own if the morning ran ahead.
There is one exception, and it matters: a reply inside a live conversation goes out immediately. Spacing belongs to cold volume, where the sequence is ours to choose. A stretched conversation only makes you slow to the person waiting on an answer.
The side benefit was internal. A profile that finished its whole day by 10am is indistinguishable from a stopped profile for the remaining eight hours, so a real outage used to hide inside normal-looking silence.
Count the ceilings yourself
The fourth rule: you should already know your position against every ceiling before the platform tells you.
The weekly one is the one people meet first. LinkedIn holds most accounts to roughly 100 invitations per rolling seven days. Our highest rolling seven-day count on a single profile is 101, so we live right against the ceiling by choice, and we track the distance in our own database. One query over your sending history answers it, and it costs a lot less than learning the same number from a refusal. Daily limits across our active profiles currently range from 1 to 15 invitations per day, set per profile according to age and history, and a new profile starts at the low end.
Two things distort this count if you leave them out.
First, history you did not create. A profile that was used for outreach before you touched it arrives with pending invitations already outstanding and a monthly personalized allowance already partly spent. We have taken over profiles carrying several hundred pending invitations from a previous operator. Withdraw the old ones before you plan volume, and treat the allowance as shared.
Second, the shape of your own week. A rolling seven-day window carries last Tuesday into this Monday, so Monday's plan competes with a send you already forgot about.
The first twenty invitations carry no verdict
The last piece is the reason people push volume into a restriction in the first place. A new profile posts a bad acceptance rate, the operator reads it as a targeting failure, and the response is more volume, faster.
Across our invitations that are at least five days old, sorted by their position in the sending profile's own history:
- invitations 1 to 20 from a profile: 33 accepted out of 274, 12.0%
- invitations 21 to 40: 39 out of 198, 19.7%
- invitations 41 and beyond: 104 out of 643, 16.2%
The first twenty land differently, and the honest reading is narrow. We can measure that they perform worse. Separating "the profile is still warming up" from "the earliest batches went to weaker targets" needs more data than we have. Either way the practical rule holds: a profile is judged from its twenty-first invitation, and until then a low acceptance rate is the price of being new. The response is another week at a low daily limit.
What to run on Monday
Five checks, all of which work on data you already have.
- Count invitations per profile over a rolling seven day window, the way the platform counts them, and keep the peak under 100.
- Take yesterday's sending and work out what share landed in its busiest hour. Above 40% means the day has a machine's shape, and the fix is spreading the same volume across the whole window.
- Attach a note only where the account actually holds an allowance for it, and move the personalization into the first message after acceptance.
- Make a refusal cost one attempt, log the provider's full response body, and close out any individual who refuses three times.
- Judge a new profile from its twenty-first invitation, and hold its daily limit low until then.
All five run on the same three things: a counter that counts attempts, a clock that divides the day by the plan, and the habit of reading the platform's answer as information about pacing. If you would rather have it operated for you, with the counters and the spacing already built, that is the work we do, and the levels are laid out on our levels and pricing page.