To automate LinkedIn prospecting, automate the list before the sending. Find people from post and profile keyword searches, score each one against your ideal customer profile, then let a sequence view profiles, send connection requests, and message the people who accept, from your own browser under daily limits. Most ranked lists of tools skip the first half, and almost none say how many connection requests a day is a sane number.
The tools are rarely the hard part. LinkedIn publishes none of its real limits, so every automated account is guessing where the line is. This guide gives you a working setup: where the people come from, how they get scored before anyone is contacted, and the exact sequence and limits that do the sending. Each search gets its own walkthrough in the guide to finding leads on LinkedIn.
To follow along on your own market, run a free post search in a new workspace. Counting the matches costs nothing.
What you can automate in LinkedIn prospecting
Six jobs make up LinkedIn prospecting, and they automate to very different degrees. Sourcing and scoring can run unattended. Sending can too, once you trust it. Judging whether a reply is worth a meeting stays with you.
Job | What runs on its own | What you still decide |
|---|---|---|
Find people | Post and profile searches, re-run on a schedule | The phrases, titles, and keywords |
Score them | A fit band and score for each person, with reasons | What your ideal customer looks like |
Connect | Profile view, request, withdraw after a wait | How many a day, and when |
Message | A message written per person after they accept | The prompt, and approval at first |
Follow up | A second touch on its own delay | When to stop |
Reply | Labeled in one inbox | What you say back |
The six steps below set up the first four jobs in order. Follow-ups and replies come once the sequence is running.
Automate LinkedIn prospecting: find and score
Nobody should get a connection request because they matched a job title and nothing else. The first two steps decide who is worth a request at all, and they cost far less than the sending does, because searching and counting are free.
Find people from post and profile keyword searches
Two searches on Find leads cover two different kinds of prospect.
People posting about a topic reads what people wrote in their own LinkedIn posts. Type a phrase such as month-end close, click + And and add spreadsheet, and a post then has to carry both. Posted in defaults to the past month. Leave it there, because posts reach this search about a week after they go up, so the past week finds very few.
Each person lands once, with their newest matching post, its date, and its reactions and comments. The post is the reason to write to them. It is also the strongest signal you will get, since they said it themselves.
Find people with the Keyword filter, under Professional bio, searches the whole profile. Combine it with Job titles, Seniority, and Countries to find the people whose profile says they own the problem, whether or not they post. This search also brings each person's employer and its domain, which the next step needs.
A few details save rework later:
A new people search leaves out anyone already in one of your tables, so a second run never doubles your list.
Phrases match whole and in order. Cost does not match costs, so add each form you mean.
Save either search and put it on a schedule. A scheduled post search reads only the posts that arrived since its last run.
Score the list against your ICP
From the table, click Tools, open the Agent tab, and add ICP scorer from Built-in agents. It judges each person against the ideal customer profile and buyer personas in your business context and returns three things: a band of strong, possible or poor, a 0 to 100 score, and the two or three facts that drove it.
Two behaviors are worth knowing before you trust it. With no business context it returns nothing, on purpose, because a score against a guessed profile is worse than none. And when a person's record is nearly empty, it says what was missing instead of calling them a poor fit.
The scorer needs a company domain, which the profile search returns and the post search does not. For post authors, the post usually is the qualification. Run Person researcher on them instead, at 5 credits a person: it reads their profile and recent activity and writes a hook for the first message.
Save a view of the table filtered to the strong band. That view, not the whole list, is what the sequence will send to. The ICP scoring guide covers writing the profile itself.
Automate LinkedIn prospecting: connect and message
Anyway, on to sending. Everything from here runs from one LinkedIn account, in one browser, so set it up once and every sequence shares it.
Connect LinkedIn in your own browser
Open the sender in Senders & mailboxes and switch to its LinkedIn tab. It shows a connection token until a browser connects.
Click Open the Chrome Web Store and add the extension. It runs in Chrome, Edge, or another Chromium browser (Google's steps for installing an extension).
Sign in to LinkedIn in that same browser. The extension connects to whichever account it is signed in as.
Copy the token, open the Oneprofile side panel, paste it, and click Connect.
Nothing touches LinkedIn from a data center. Every step opens LinkedIn in your browser and presses the same buttons you would. Before building anything, use Try a step on your own profile URL and watch one run end to end.
Build the connect and message sequence
Create a sequence and, on its Audience tab, choose the table and the strong-fit view. Keep in sync starts on for a new audience, so people who join the view later are enrolled as they arrive.
On the Builder tab, this order works for most cold prospects:
View profile. A light first touch that costs the person nothing.
Wait one day.
Send connection request, left on No note.
On Connected: wait one day, then Send message.
On Not accepted in 14 days: Withdraw connection request, which the builder adds for you.
The request waits for the person's answer and then sends them down a branch, and the next step's wait counts from the moment they accepted. Someone already connected goes straight down Connected. Has a LinkedIn URL? and Already connected? under Then check… skip a step rather than waste it.
Let the agent write the messages
Open the message step and switch it to Written by the agent. The prompt says what the message is for, in a sentence, and the agent writes the words per person from your business context, the research it can see, and the hook Person researcher wrote. It keeps the voice of your emails.
Open the step's LinkedIn preview tab to read the message against a real contact before anything goes out. A preview sends nothing and costs no credits. When a draft is off, Needs work in Tasks asks what is wrong, writes it again, and proposes a learning the agent follows in the messages it writes after you approve.
A small opinion, and it has nothing to do with Oneprofile: the connection note is overrated. A free account gets 200 characters for one, and most notes are a pitch squeezed until it reads like a telegram. A request with no note, followed by a real message a day after they accept, reads better and dodges LinkedIn's hidden cap on notes besides. If you do write notes, the connection note character limit post has the counts by account type.
Approve the first steps by hand, then go automatic
A new connection starts on Manual run. Every step waits in Tasks until you click Approve, and you can edit the words first, since what you approve is what sends.
After a few days of approving messages you would have written yourself, open the sender's Settings tab and set LinkedIn run mode to Automatic run. Steps then run on their own within the limits and the sequence's sending window. A sequence can still ask for approval on top, under LinkedIn approval on its own Settings tab.
Limits to set before you automate LinkedIn prospecting
Every new sender starts on deliberately low ceilings. LinkedIn publishes none of its real limits, so treat these as a starting point and raise one only knowingly, per account.
Action | Starts at, per day | Per week |
|---|---|---|
View profile | 40 | No weekly limit |
Send connection request | 15 | 80 |
Send message (and InMail) | 40 | No weekly limit |
Follow contact | 30 | No weekly limit |
Like latest post | 30 | No weekly limit |
Withdraw connection request | 20 | No weekly limit |
All of them live under Limits on the sender's LinkedIn tab, with withdrawals in Housekeeping, which stays folded until you open it. Daily ceilings reset at midnight in the sender's timezone, and a weekly one counts today and the six days before it. Connection requests always keep a weekly limit.
Pacing matters as much as volume. Steps run one at a time, at least the Time between steps apart, which starts at 10 minutes and can be set from 1 to 60. Each wait runs a little longer at random. Until 2026-10-01 the default gap was 90 seconds, which kept a queue moving but packed a day's steps into a burst no person browses at, and that is why it changed.
Two sequences sending as the same sender share its gap and its limits. So adding a second campaign does not double what the account sends. It splits the same budget.
"Playing the numbers game in outreach is expensive and wasteful. A better tool lifts your reply rate far more than sending more ever will."
Utku Zihnioglu, CEO of Oneprofile
A worked example: one week of LinkedIn prospecting
Say you sell a billing tool to finance leads at B2B software companies, and a profile search for controllers and heads of finance with SaaS as a keyword gives you 300 people. ICP scorer marks 90 of them strong. Those 90 are the sequence.
With the defaults, a weekday sending window, and the first profile views on a Monday:
Tuesday to Friday. 15 connection requests a day, 60 in all, each a day after that person's profile view. The weekly limit of 80 never bites.
The next Monday and Tuesday. The last 30 go out. Monday's seven-day count holds Tuesday to Friday's 60 plus its own 15, so the weekly limit still has room.
As people accept. Each one gets the agent's message a day after accepting, well inside the 40 a day message ceiling.
Day 14 after each request. Anyone who has not answered is withdrawn, which keeps your pending list short.
Run the same sequence on a seven-day window and the weekly limit does bite: the sixth day of requests sends only 5, and the seventh none, until the first day's requests leave the seven-day count. Ninety strong-fit people take nine days from the first profile view either way. That is slower than a tool that sends 100 a day, and it is meant to be.
Choosing a lead generation tool for LinkedIn
If you are comparing tools rather than setting one up, four questions sort them faster than any feature grid. The LinkedIn prospecting tools comparison groups the options by job; these are the questions to ask of each:
Where does it run? In your own browser on your own connection, or on a server signed in as you. Oneprofile runs in your browser.
Can you set limits per action? A single "actions per day" number treats a like and a connection request as the same risk, and they are not.
Does it branch on acceptance? A message should go only to people who accepted, and a request nobody answered should be withdrawn.
Where do the people come from? A lead generation tool for LinkedIn that only sends still needs a list from somewhere else, and that list is rarely scored.
Probably the most underrated of the four is the last one. A careful sender working from a weak list still gets a weak reply rate. If your outbound also runs on email, LinkedIn outreach automation shows LinkedIn and email steps in one sequence, and Outreach covers mailboxes and every step type.
The quickest test is your own market. Type the phrase your best customers would post about, read who comes back, and score them. Run your first post search free and see who is writing about your problem this month.
