Most guides to LinkedIn automation explain what it is and how to avoid getting restricted. This one is the practical version: the order to set things up in, what to put in each field, and what to check before you switch anything on. It follows the same path as our video walkthrough, so you can read along or use it as a checklist afterwards.
Short answer: connect one LinkedIn account and set conservative weekly limits and working hours first. Then create a signal that finds people with a reason to talk now (hiring, engaging with a post, viewing your profile), route them through an ICP check into a list, and let that list feed a campaign. Keep the sequence short: an invite, one or two warm-up touches, a first message after they accept and one AI-written follow-up. Turn on an AI reply agent with clear rules and a hand-off, work every reply from one inbox, and review the numbers weekly.
For background, read how to automate LinkedIn outreach safely (the safety playbook) and the complete LinkedIn automation guide (the wider picture). This post assumes you have decided to automate and want it running.
Key Takeaways
| Step | What you set up | Check before moving on |
|---|---|---|
| 1. Account and limits | One connected account, weekly caps, working hours | Caps sit well below LinkedIn's ceiling |
| 2. Signals | One or two lead sources tied to buying intent | The first run returns people you would want to talk to |
| 3. List | An ICP filter and a destination list | The list fills with the right titles, not near-misses |
| 4. Sequence | Invite, warm-ups, messages, AI follow-up | You have tested what a real prospect would receive |
| 5. Reply agent | Voice, offer, rules, files, a hand-off | You have practised against it with awkward replies |
| 6. Inbox | One place to work every reply | Interested replies get a same-day answer |
| 7. Measure | Weekly report by campaign | You know which step or signal to change next |
Step 1: Connect your LinkedIn account and set safe limits
Everything an automation tool sends goes out from a real LinkedIn account, so this comes first. In Updately you open the LinkedIn accounts page, click Add account, pick the region the account normally signs in from, and log in on the secure page that opens, where 2FA and security checks are handled. If sign-in keeps looping, you can connect with your session cookie instead.
Two rules before you go further:
- One tool per account. Running two automation tools on the same account at once is the fastest route to a restriction, because neither can see what the other is sending.
- Keep the region where the account actually lives. An account that suddenly signs in from another country is exactly what LinkedIn flags.
Set the weekly caps
Open Limits & schedule on the account and set three weekly quotas: profile visits, invitations and messages. Quotas reset every Sunday at midnight. You can set them lower than Updately's defaults but not above its safety caps.
- Invitations. LinkedIn does not publish an exact number, but most accounts hit a ceiling at around 100 invitations a week. Treat that as the ceiling, not the target.
- New or quiet accounts. Start at around 35 to 70 invitations a week and raise it gradually once acceptances come in without warnings. Updately also applies lower limits for the first 7 days after an account is connected.
- Messages. Messages only go to people who accepted, so this cap mostly governs follow-ups. Keep it in the same range as invitations to start.
Set working hours and active days
In the same dialog, set the Timezone, the Launch Hour (the account works for about four hours from then, in that timezone) and the Active Days, which default to Monday to Friday. Set the timezone even if it matches the region: without one, the launch hour is read as UTC.
If you have more than one account, a campaign can send from several of them. Leads are spread evenly, each account stays inside its own limits, and anyone mid-sequence stays with the account that invited them. The trade-offs are in how to scale LinkedIn outreach across multiple accounts.
Step 2: Find leads from buying signals
The sequence matters less than who goes into it. A search for "VP Sales, UK, 50 to 500 employees" returns everyone who fits. A signal narrows that to people who fit and have a reason to talk this week, and it gives your first message its opening line.
In Updately, a signal is a standing rule that runs on a schedule and drops new people into a list:
| Signal | Who it finds | Good for |
|---|---|---|
| LinkedIn job posts | Companies hiring a role your buyers hire, then the decision maker there | A function that is growing (why it works) |
| People search | People matching filters, or a pasted LinkedIn or Sales Navigator search URL | A steady baseline from day one |
| Sales Navigator lead list | The leads saved in one of your lists | Teams that already curate lists |
| Post engagement | People who reacted to or commented on a post, yours or a competitor's | Warm, topic-aware prospects (how to use it) |
| Profile viewers | People who viewed your profile | The warmest list you have |
| Keyword posts | People posting about your topic or problem | People asking for help |
| Network | Your first-degree connections | People who already know you |
| G2 and Trustpilot reviewers | People reviewing a category or competitor | Switching campaigns (how to mine reviews) |
| Tech stack | People at companies that use or adopted a technology | Integration or switching angles |
Pick one or two to start. If you are unsure, people search works immediately because it does not wait for anyone to post, hire or engage. Add a time-based signal such as job posts or post engagement once you know what a good lead looks like. The method behind choosing and scoring sources is in how to automate LinkedIn prospecting, so we will not repeat it here.
Step 3: Put the leads in a list
The last part of creating a signal is routing: which list the leads go to and which ICP profile they are checked against.
Write the ICP like a brief. Every lead gets an AI fit check against a description you write, so describe your buyer the way you would to a new salesperson: the roles that feel the pain and the ones that sign, the kind of company, the size range, and exclusions such as "not agencies" or "not students". Exclusions remove more noise than any other line. Start in Broader mode while you learn what your signals surface, then switch to High precision.
Make one list per audience. "Heads of RevOps, hiring" and "Founders who liked a competitor post" need different first messages, so they belong in different lists and different campaigns. A lead who sits in two lists feeding two campaigns can get messages from both.
Read twenty rows before you connect a campaign. If the titles are wrong, fix the signal or the ICP now. Automation will happily send good messages to the wrong people.
Step 4: Build the campaign sequence
Create a campaign, name it after the audience, pick the sender account or accounts, and start from blank or a template. Then set the Trigger: the lists that feed it. While the campaign is active, anyone new in those lists is enrolled automatically. That is what turns the setup into a running system rather than a one-off blast.
The connection invite
LinkedIn does not let you message people you are not connected to, so almost every sequence starts with an invitation. Free LinkedIn accounts only get a handful of personalised invitation notes a month, so a plain invite is often the better default. If you add a note, keep it to one line about them; see connection request best practices.
Set Withdraw if not accepted on the invitation step (1 to 90 days). Pending invitations count against your limit, so withdrawing stale ones makes room for new people. LinkedIn notes that after you withdraw an invitation you cannot re-invite that person for up to three weeks, so do not set it too short.
Warm-up steps
Warm-ups make your name familiar before the invite lands, and none of them use your invitation or message limits:
- View profile. You appear in their "Who viewed your profile".
- Like posts. Reacts to one to five of their recent posts, with an option to skip old ones.
- Comment. AI writes a short, specific comment on their newest post, with no pitch and no links. It is capped at 15 a day and skipped if they have not posted recently. LinkedIn comment prospecting explains why the comment has to stand on its own.
One or two is enough. A profile view plus a like a day or two before the invite works for most audiences.
Messages after they accept
Message delays count from acceptance, not from when you sent the invite, so someone who accepts three weeks later still gets your first message at the right moment. Acceptances are detected by periodic checks, so expect a gap of a few hours.
Each message step can be written three ways:
- Same message for everyone, with variables like
{FirstName}and{Company}. Predictable and easy to judge. - AI Follow-up, which writes a personal message for each lead from their profile and posts. You give it a prompt about intent, not format. It can go out as text, a voice note or a meme; save the last two for a later touch.
- From the conversation, which reads the chat so far and follows an instruction such as a soft nudge or a breakup.
If you are not yet sure about AI-written messages, turn on Review before sending, and each one is held for you to approve, edit or skip.
A sequence to start with
| Step | Type | Wait |
|---|---|---|
| 1 | View profile | Right away |
| 2 | Like posts (1 to 2 recent) | 1 day |
| 3 | Invitation, no note | 1 day |
| 4 | Message: open with the signal, ask one question | 1 day after they accept |
| 5 | AI Follow-up: a new angle, not "just bumping this" | 3 days |
| 6 | From the conversation: short breakup | 5 days |
Three messages is plenty. People who want to reply tend to reply early, and extra steps mostly produce "please stop". For opening lines tied to each signal, see signal-based outreach templates.
Before activating, click Test Sequence, pick a real contact or paste a profile URL, and read exactly what they would receive. Fix anything that reads like a template.
Step 5: Let an AI reply agent answer within your rules
When someone replies, they leave the sequence immediately and never get the next canned step. Then either the reply waits for a person, or an AI agent answers. If you turn the agent on, set it up on the campaign's AI Agent tab in this order:
- Voice. Paste two or three messages you actually wrote. The agent copies the style, not the words.
- Offer and goal. What a good conversation ends in, such as "book a 20-minute call", plus your booking link.
- Reply rules. Each is an if and a then: "If they ask about price, then attach the pricing sheet." Start small and add rules as real replies show you gaps.
- Files and knowledge. Case studies, pricing, a demo video, each with a note on when to use it, plus the facts the agent may state. It only states facts it has been given.
The hand-off matters most. The agent passes the chat to a person when it lacks the facts to answer, or when the prospect shares an email address, phone number or calendar link. You want a human in the room at that point anyway.
Then use Try the reply AI to play the prospect. Skip the easy "sounds interesting" and try the awkward ones: a price question, "send me something", "we already use a competitor", a rude one-liner. Each should get a sensible answer or a hand-off. Nothing in the practice chat reaches LinkedIn. If you want to check every answer for the first week, turn on approval. More on how these agents behave is in LinkedIn AI agents.
Step 6: Work replies in one inbox
Logging into each LinkedIn account to find replies does not scale past one campaign. The Unified Inbox collects replies from every campaign and every sender account, and labels each one Interested, Needs info or Not interested, whether or not the agent is on.
A daily routine that takes minutes:
- Answer every Interested reply the same day. These are what the whole system exists to produce.
- Work Needs info. If the AI handed one over, the missing answer is usually a fact you can add to its knowledge.
- Clear anything waiting for approval.
- Once a week, scan Not interested for patterns, such as a wrong-title segment or an objection your first message keeps triggering.
Claim a conversation when you take it, so the AI stays out and teammates can see who owns it. Replies can also post to Slack, and contacts who replied can go to HubSpot or other tools through webhooks or Zapier.
Step 7: Measure and iterate
Let a campaign run for a week or two before changing anything; small samples make every number look meaningful. The Reports page compares campaigns side by side (invites, acceptance rate, messages, reply rate, interested replies, conversations waiting on you), each campaign's Insights tab breaks it down by step, and reports can be emailed as a PDF on a schedule.
| What you see | What it usually means | What to change |
|---|---|---|
| Low acceptance rate | Wrong people, or a profile that gives no reason to accept | Tighten the ICP or signal; fix your headline |
| Good acceptance, few replies | The first message | Rewrite step 4 around the signal; cut the pitch |
| Replies, mostly Not interested | Wrong audience or offer | Split the list or change the offer |
| Many conversations waiting on you | The agent is missing facts | Add them to its knowledge and rules |
| Contacts piling up "in progress" | Enrolment outpaces your limits | Narrow the signal, don't raise the caps |
Acceptance on its own is the least useful number; our acceptance rate analysis explains why replies and meetings are what to optimise, and outbound reply rates in 2026 gives a sense of what to expect.
Change one thing at a time. To compare two versions, duplicate the campaign, change one step and feed each copy from a separate list. A duplicate does not carry over contacts, so nobody is messaged twice.
Frequently Asked Questions
How many connection requests can I safely send per week? LinkedIn does not publish an exact figure, but most accounts hit a ceiling at around 100 invitations a week. Start lower, around 35 to 70 a week on a new or quiet account, and increase gradually. The detail is in is LinkedIn automation safe in 2026.
Should I add a note to my connection requests? Usually not on a free LinkedIn account, since you only get a few personalised notes a month. A plain invite after a profile view or a like works well, and your first message after acceptance carries the context.
Will an AI reply agent say something wrong to a prospect? It only states facts you have given it, and it hands the conversation to you when it lacks an answer or the prospect shares contact details. Test it in the practice chat and turn on approval for the first week if you want to check every reply.
Do I need Sales Navigator? No. Job posts, post engagement, profile viewers and keyword posts work without it. It helps if you already build searches or lead lists there, because you can bring them in directly.
What should I change first if nobody replies? Look at acceptance before replies. If people accept but do not answer, rewrite the first message around the signal that found them. If they do not accept, the problem is who you are inviting, not what you say.
Want the whole workflow in one place, from signals to sequences to an AI reply agent? Try Updately.
Related: How to automate LinkedIn prospecting · LinkedIn buying signals