How does LinkedIn detect automation?

LinkedIn does not publish how it detects automation. Its Help pages say it is constantly improving its technical measures and that tools may stop working without notice. Its pages do name behaviours that lead to invitation restrictions: many invitations in a short time, and invitations that are ignored, left pending or marked as spam. Anything beyond that is third party reporting, not fact.

Not legal advice. Read on 7 November 2026. Pages change, so read the originals. Evidence levels used below: documented* (LinkedIn says it), reported (a third party says it, cited, not confirmed by LinkedIn), our view (opinion).*

This post separates those three levels, because the topic is full of confident claims about "the algorithm" with no source. For the rules themselves, see what LinkedIn's terms say about automation. This one is about the narrower question: what is known about how the line gets noticed, and what pacing patterns make sense given how little is known.

What does LinkedIn say publicly about detection?

Evidence level: documented. Three things are on the record.

The honest summary is that LinkedIn states the rule, the consequence and that it keeps working on enforcement. It does not describe the method, and a page that claims to know it is guessing.

Which behaviours does LinkedIn itself link to restrictions?

Evidence level: documented, on the invitation restrictions page. It names these reasons for a restriction on sending invitations:

• You have sent many invitations within a short amount of time.
• Many of your invitations have been ignored, left pending or marked as spam.
• If you send an excessive number of invitations and LinkedIn suspects automation, the account may be restricted or suspended.

It also says LinkedIn limits the number of invitations you can send, without giving a number, and that most restrictions are removed automatically within about a week. Its advice is to send invitations to people you know and trust, and to add a personalized message that explains how you know them.

These are signals about outcomes (how people respond to you) and rate (how fast you send). Both apply to people who never use a tool. That is worth holding on to: the documented triggers are not about software at all.

What do third parties report?

Evidence level: reported. Vendors of LinkedIn tools see many accounts and publish guidance. It is not LinkedIn's word and it is written by companies that sell automation, so weigh it accordingly.

• Volume spikes. Expandi's guidance quotes LinkedIn's help centre on sending an excessive number of invitations in a short time, and warns against going from no activity to dozens of invitations a day overnight. It suggests about 100 invitations a week, and says this is "not an official LinkedIn quota" (Expandi).
• Acceptance and pending pile ups. HeyReach lists an acceptance rate below 20% and unanswered pending requests among the factors it links to risk (HeyReach).
• Mixed tools and manual use. The same HeyReach page lists manual LinkedIn actions and other automation tools running on the same account as risk factors.
• Very short sending windows. HeyReach also lists these, because they push a tool to send many requests at once.
• Trying to get round a restriction. Expandi says switching devices, campaign types or sending methods during a restriction will not lift it and may add risk.

Treat each as a hypothesis. Expandi itself says LinkedIn does not explain how it works out each member's allowance, and calls its own "fuse limit" concept "not a named mechanism documented by LinkedIn".

What is speculation?

Evidence level: none, so we label it rather than repeat it. You will read confident claims about browser fingerprinting, IP reputation, mouse movement, typing speed, session length and request patterns. Some may be true. We found no LinkedIn page that confirms any of them, and none from a third party that shows how they were measured. Two common claims deserve particular caution:

• "A dedicated proxy makes the account invisible." No source we found says so. A proxy decides where traffic appears to come from. It does not change how many invitations you send or whether people ignore them.
• "Cloud tools are safer than extensions" or the reverse. LinkedIn's pages treat third party software and extensions together. Our cloud versus browser comparison covers the practical differences, not a difference in safety.

What patterns are safer, given what we know?

Evidence level: our view, built from the documented triggers. None of this is a guarantee.

1. Ramp, do not jump. A quiet account that suddenly sends dozens of invitations a day is the pattern the pages warn about. See the warm up plan.
2. Send slowly and spread it out. The documented trigger is "many in a short amount of time". Spacing and a daily cap address it directly. Sellinger defaults to 20 invitations a day, varied by a few each day, and 150 a week.
3. Keep a quiet window. Humans sleep. Sellinger's default is no sending from 01:00 to 08:00 account time, and you can set days, hours and time zone.
4. Write to people who are likely to say yes. Ignored, pending and spam marked invitations are on LinkedIn's list. A relevant audience and a note that says why you are writing reduce them.
5. Watch the outcomes. If acceptance falls and pending requests pile up, slow down before LinkedIn does it for you. Withdraw stale invitations. The limits tracker helps you see the pile.
6. Run one tool per account. Mixed tools and manual bursts on top of automation are the reported risk factors above.
7. Stop at the first warning. Rate limits and verification prompts are signals. Stop, wait, and read what to do if the account is restricted.

What does Sellinger do?

Evidence level: our setting. You connect your own LinkedIn account through a hosted login, and Sellinger does not store your LinkedIn password. Limits are set per account and the agent cannot exceed them. When LinkedIn returns a rate limit, Sellinger sets a cooldown on that account before trying again. A campaign can run in approval mode, where each message waits for a person. That does not hide anything from LinkedIn. It keeps a person responsible for what goes out.

What can't we promise?

• That any pattern avoids detection. No one outside LinkedIn knows the method.
• That a restriction will not happen. LinkedIn decides, and it says tools may stop working without notice.
• That third party figures will hold. Vendor numbers are their guidance, not rules, and they change.

If you accept the risk, start with the safety checklist and the wider LinkedIn automation guide.