LinkedIn outreach
How does an AI agent for LinkedIn work, and where does it stop?
By Viktorio Shopov, co-founder of Sellinger. Reviewed by Asen Levov. Published . Last updated . 8 minute read.
An AI agent for LinkedIn researches each prospect, writes the invitation and every follow up, answers replies and books the meeting on your calendar. It stays inside daily limits you set. It cannot make people accept, cannot lift LinkedIn's limits, and a person still hosts the call and closes.
What is an AI agent for LinkedIn?
It is software that runs the outreach conversation on a LinkedIn account you connect. It reads each person's profile and recent activity, writes a message for that person, follows up on a schedule, reads the replies and answers them, and offers a time when the person is interested. A person sets the audience and the offer, can approve what goes out, takes the call and closes.
That is different from a LinkedIn automation tool, which sends a sequence of fixed templates to a list and does not read the replies. We compare the two in LinkedIn automation vs an AI SDR agent. An agent is also one slice of the larger idea in What is an AI SDR?, narrowed to one channel.
What does the agent do, step by step?
| Step | What the agent does | What you control |
|---|---|---|
| 1. Research | Reads the prospect's profile, recent activity and company context | Who is on the list, and what the agent may know about your offer |
| 2. Invitation | Writes a connection request for that person | Whether a note is added, and whether a person approves it first |
| 3. First message | After acceptance, an opener based on one observation about their work | The tone and the rules, for example no pitch in the opener |
| 4. Follow ups | A new angle each time, on a schedule, then a clean exit line | How many, and the gap between them. LinkedIn defaults to 4, then email, with a first gap of 5 days |
| 5. Replies | Reads the reply and answers questions, handles objections, asks a qualifying question | What it may say, and when it must stop and hand over |
| 6. Booking | Offers a link or proposes a time on your calendar | Which calendar and which meeting type |
| 7. Hand over | Stops when it cannot answer, for example a price it has no answer for | Who takes over, and how fast |
The agent covers steps 1 to 6 and flags step 7. In our own account, the median agent reply since May 2026 was 2.3 minutes, and 90% of replies came within an hour (Sellinger case study). That speed is the part software does better than a person: it answers at night and at weekends.
What does a first message look like?
We do not publish a template, because the point is that there isn't one. The agent writes each message from the person's profile and recent activity, so two people in the same audience get different messages. The one pattern we use on our own account is an opener that makes one observation about their work, with no pitch, no link and no vendor name, then follow ups about every five days, up to three, each with a new angle. If the person asks what this is, the reply names Sellinger in one line, ties it to their own words and offers a short showcase.
What matters for you is that the agent writes from what you tell it. If your product description is vague, the messages will be vague. If you give it exact facts, limits and a clear next step, it can answer capability questions precisely. The foundry in our Lamothermic study told its agent what it casts and what it does not, and the agent could then answer engineers precisely (Lamothermic case study).
What will it not do?
Being plain about the limits is the point of this post.
- It will not guarantee acceptance. Acceptance depends on who you invite and why they should care. Across 42 customer teams in the 90 days to 19 September 2026, about 4 in 10 invitations were accepted (39%). On our own account 41% of 68,818 requests were accepted. Those are averages, not promises, and a cold audience with no reason to connect will accept far less.
- It will not bypass LinkedIn's limits. LinkedIn says that limits exist for all members and that restrictions are applied, but its help pages publish no weekly number. A restriction typically lasts one week, and the reasons LinkedIn lists include sending many invitations in a short time and invitations that are ignored, left pending or marked as spam (Types of restrictions). The agent has per account caps and works inside them. It does not try to get around LinkedIn's. See LinkedIn invitation limits.
- It will not make the account risk free. LinkedIn does not permit third party software that automates activity on its site, and says it may restrict or close the account of a member who uses such tools (Prohibited software and extensions). No tool can guarantee that LinkedIn will never restrict an account, including ours. We discuss the trade offs in Is LinkedIn automation safe?.
- It will not set your offer or fix your audience. The same agent on our own account booked 2.6% to 3.7% meetings per person messaged from one audience and under 0.4% from another.
- It will not close. A person hosts the call. The agent books it.
- It will not invent what it does not know. When it cannot answer, it stops and a person decides.
How does it stay inside limits?
You connect your own LinkedIn account through a hosted login, and Sellinger does not store your LinkedIn password. You can connect through your own proxy. After that, the agent works inside per account limits:
- A daily invitation limit, 20 by default, varied by a few each day.
- A weekly invitation cap, 150 by default.
- A quiet window with no sending, 01:00 to 08:00 account time, unless you set your own schedule.
- Invitations spread across the day, not sent in a burst.
An account can carry its own limits. These are our defaults, not LinkedIn rules. Treat them as a starting point and watch your own account: how many invitations are accepted, and how many stay pending.
Can you approve what it sends?
Yes. A campaign can run in approval mode, where every message waits for a person, or send automatically. You can also split it: send connection requests automatically but require approval for follow ups and meeting invitations. Our advice is to start in approval mode, read the first 50 to 100 messages, change the rules, then relax approval for the message types that are consistently right. A person can take over any conversation at any time.
Where do people start, and when does it fit?
An agent fits when you have a defined audience, a clear offer and a person who will take the calls. It fits three situations:
- Cold outreach to a list, where the agent researches each person before writing.
- People who engage with a topic, for example liking or commenting on a post about it, using intent signals.
- Inbound requests, where people send you connection requests or messages and the agent qualifies them and books a time.
It fits less well when the audience is unclear, when your offer changes weekly with nobody updating the agent, or when you have nobody to take the call. In those cases, fix the foundations first. The causes of failure are in Why do AI SDRs fail?.
How do you judge whether it works?
Not by replies. On our account replies held near 40% for a year while meetings per person messaged drifted below 1% from June 2026. Count people messaged, acceptances, replies, positive replies, meetings booked and meetings held, and keep them separate. The weekly plan is in Do AI SDRs actually work?. Start with one audience, 30 days, and one person who owns it.
What is the short version?
- The agent writes, follows up, answers and books. A person decides, hosts and closes.
- It works inside limits you set, and none of them are LinkedIn's promises.
- Acceptance and meetings depend on audience and offer more than on the software.
- No tool can guarantee that LinkedIn will not restrict an account.
- Start in approval mode, with one audience, and count held meetings.
See the LinkedIn agent page for how setup works and what the free trial includes.
Common questions
What is an AI agent for LinkedIn?
Can it guarantee that people accept my invitations?
Can it get around LinkedIn's limits?
Does a person still have to do anything?
Is it safe for my LinkedIn account?
How Sellinger helps
Read how it works on See how the LinkedIn agent works, or see a real campaign in Read the Sellinger case study.
