A 15 point LinkedIn safety checklist for AI outreach

Run AI outreach on LinkedIn with a hosted login, a daily and weekly invitation cap, a quiet window, no manual invites on the same account, a watch on acceptance, a human who reads replies and a plan for a restriction. No tool can guarantee LinkedIn will never restrict an account, so the checklist lowers risk, it does not remove it. The 15 points below are the ones we apply and the ones we would check in any tool.

LinkedIn's rules are written for people, and automation sits in a grey area that the platform can enforce when it chooses. Our posts on whether LinkedIn automation is safe and what LinkedIn's terms say cover the reading of the rules. This post is the practical side: what to check.

A. How the account connects

1. The account connects through a hosted login, not a stored password. In Sellinger you connect your own account and the platform does not store your LinkedIn password. Be wary of any tool that asks you to hand over credentials.

2. No browser extension drives your account. LinkedIn's prohibited software and extensions page names tools that scrape or automate through the browser. Cloud tools that connect through a login differ in how they run, and you should know which kind you use. See cloud vs browser LinkedIn automation.

3. The network path is stable. A session that jumps between countries looks odd. You can connect through your own proxy if you need a fixed location, which the connect flow accepts. We make no claim about any proxy type being safer.

4. One account, one owner. Do not share a profile between people in ways that make it behave like several.

B. Volume and timing

5. A daily invitation cap. Our default is 20 a day, varied by a few each day, so activity is not identical day to day.

6. A weekly cap. Our default is 150 a week per account. LinkedIn's own pages do not publish one number for invitations, and they disagree on notes, which is covered in LinkedIn invitation limits. Start below what you think the limit is.

7. A quiet window. No sending in the night: default 01:00 to 08:00 account time, or the days, hours and time zone you set. Activity at 3 a.m. local time is not natural for most people.

8. Gradual changes. Raise limits in steps and watch the effect. We do not have an account ramp in the product, so a new or dormant account should start low by your own choice.

C. Who is sending what

9. No manual invites on the same account as the agent. Manual and agent invitations share one weekly limit. Sellinger watches for invitations sent outside the platform on a seat the agent also uses, and warns when it sees three or more in a week. See how to spot invites sent outside your tool.

10. Messages that read like a person. Short, specific, no repeated paste. Variation in wording protects the recipient experience and reduces the chance of reports. See personalisation at scale.

11. A list that fits. Sending to people who do not fit gets ignored invitations, and LinkedIn lists ignored and pending invitations among the reasons for a restriction. A tight list is a safety measure.

D. Signals to watch

12. Acceptance rate. About 4 in 10 invitations were accepted across 42 customer teams in the 90 days to 19 September 2026, and 41% on our own account. A rate far below your own norm is an early warning to pause and look at the list.

13. Pending invitations. A growing pile of pending invites is a signal. Withdraw old ones, and look at why they are ignored.

14. Warnings from the platform. A prompt to verify, an invitation limit message or a restriction notice means stop sending, not retry. A good tool pauses a seat automatically on repeated refusals. Ours pauses engagement for 24 hours after three consecutive resource access restrictions.

E. When something goes wrong

15. A plan for a restriction. Know in advance who checks, what you do (stop all outreach, follow LinkedIn's instructions, appeal if appropriate) and how you tell clients. The steps are in what to do if LinkedIn restricts your account.

How do you use it?

Print the 15 points and score a tool or your own setup: yes, no or not sure. Every "not sure" is a question to ask the vendor. If you have more than three "no" answers, fix those before sending at volume.

What does a scored checklist look like?

A made up example, not a real account. An agency runs outreach for one client on two LinkedIn accounts and scores itself on the 15 points.

• Yes on 1, 5, 6 and 7. Hosted login, a daily cap, a weekly cap and a quiet window are set.
• Not sure on 3 and 4. Nobody has written down which network path each account uses, and the client's assistant has the password to one profile.
• No on 9. The client still sends a few invitations by hand from the phone app.
• Not sure on 13. Nobody has looked at the pending pile since the account was connected.

The agency does not need to fix all four at once. It fixes the "no" first (agree that invitations come from one place), then turns each "not sure" into a yes or a no by checking. That is the use of the list: it turns a vague worry into a short to do list.

Which points matter most if you can only fix three?

We cannot rank the 15 by measured effect, because we do not publish a restriction rate and LinkedIn does not publish how it decides. What we can do is pick the points that follow directly from what LinkedIn says. Its help pages list ignored and pending invitations among the reasons for a restriction, and manual and agent invitations share one limit. So start with:

1. Point 11, a list that fits. It is the only point that lowers the number of ignored invitations at the source.
2. Point 9, one sender per account. It stops your own count from being lower than LinkedIn's.
3. Point 14, stop on a warning. A verify prompt or a limit message means stop, not retry.

Who checks what, and how often?

How does the checklist change with more accounts?

Each account needs its own score. A limit of 150 invitations a week is per account, so a team with five accounts does not have one big pool to spend. Points 4 and 9 matter more as accounts multiply, because a profile connected twice or sent from twice is easy to lose track of. Keep one row per account in a shared sheet, with the owner, the network path and the last date someone checked for outside invites.

Does a lower volume make it safe?

No. Lower volume gives LinkedIn less to look at, and it gives you more room before a limit. It does not remove the risk, and it does not help if the list is poor. A small send to the wrong people produces ignored invitations too. Treat volume as one of five levers, next to the list, the content, the timing and your reaction to warnings.

What the checklist does not do

It does not make an account safe. LinkedIn can restrict an account for reasons you cannot see, and no vendor can promise otherwise; if a tool does, ask for the evidence. We do not publish a restriction rate. What you control is the list, the volume, the content and how fast you react to a warning. Those are the points above.