Nobody outside LinkedIn knows the rule. What is knowable is which inputs change the answer, and the biggest one by a distance is where the sign-in appears to come from.
Location, first and loudest
An account that has signed in from Warsaw for three years, signing in from a data centre in Virginia, is the clearest possible signal that something changed. It is also the easiest to avoid: the connect form has a Sign in from setting, and setting it to the country the account actually lives in removes the signal entirely.
This is the setting people skip because it looks like a formality. It is the opposite of a formality.
Newness
A connection that has existed for an hour is treated differently from one that has existed for a month. That is why a new account should be ramped up rather than handed a full allowance immediately, and why the first week of any integration should be quieter than the second.
Rhythm
Two hundred actions in four minutes and then silence for a day is not what a person looks like, whatever the daily total says. Spread work across hours, and spread it where it is created rather than only where it is sent.
What does not help
- No proxy makes automation undetectable. Anyone selling that is selling a sentence, not a product.
- Rotating addresses on every request is worse than one consistent address: it turns one account into a different person every morning.
- Nothing prevents a restriction. The honest claim is a lower chance, and anybody who makes a stronger one has not read the terms.
Questions people ask
Do AI agents trigger more verification?
Not by being AI: LinkedIn sees requests, not who wrote them. What changes the answer is the same as ever — where the sign-in comes from, how new the account is, and the rhythm of what it does. An agent left alone with a long list changes the rhythm fastest.
Should AI agents connect accounts at all?
To create a sign-in link, yes. To handle passwords or cookies, no; keep that out of any shared conversation.