How LinkedIn Detects Automation: Signals Explained
LinkedIn does not have one trip-wire. It watches a cluster of behavioral signals in parallel, and accounts get flagged when enough of those signals turn red at the same time.
That framing matters, because most people optimise for the single metric they can see: daily connection request count. The actual detection logic is considerably wider than that.
The Signals LinkedIn Actually Watches
Volume and burstiness. Raw count matters less than the shape of your activity. Sending 30 requests spread across a working day looks different to LinkedIn's systems than sending 30 requests in a 12-minute window. The second pattern does not look human. Bursts are one of the cleaner signals to detect algorithmically, and they are especially common with tools that run scheduled batch jobs on the hour.
Timing regularity. Humans are erratic. We get distracted, jump between tabs, reply to a Slack message. Automation tools, especially simpler ones, execute actions on fixed intervals: one connection request every 90 seconds, every time, without variation. That mechanical regularity is a strong signal. LinkedIn's systems can measure the standard deviation of your inter-action timing, and a near-zero deviation is not a human pattern.
Session and browser fingerprints. This one is specific to browser-based tools. Extensions like Dux-Soup or Linked Helper operate inside a real Chrome session, which sounds safer but creates its own problems. The extension modifies the DOM, injects scripts, and changes how the LinkedIn page behaves in ways that can be detected. User-agent strings, WebGL renderer data, installed plugin lists, and canvas fingerprints can all differ from a normal user session. LinkedIn cross-references these. A session that claims to be Chrome on macOS but presents an unusual plugin fingerprint is worth a second look from their systems.
IP inconsistency. If your account usually logs in from London and then a connection campaign fires from a Frankfurt data-center IP, that mismatch is a flag. This is common with cloud tools that route through shared residential or data-center proxies poorly matched to the account's normal location.
Acceptance rate and reply rate. LinkedIn does not only look at what you send. It watches what happens afterwards. A campaign that generates 200 connection requests and gets 3 acceptances is sending a signal: either the targeting is poor or the outreach looks like spam. Sustained volume with low acceptance rates pushes accounts into a review queue. The same logic applies to message reply rates on InMail.
Why Extensions Are Easier to Catch Than Cloud Tools
This is worth being direct about. A browser extension runs inside the browser, which means it runs inside the same JavaScript environment that LinkedIn controls. LinkedIn can detect that the browser is being scripted rather than driven by a human hand, by looking for event timing anomalies, missing mouse-movement events before a click, or unusual API call sequences that do not match normal navigation.
A cloud tool that calls the LinkedIn API through an approved data layer, like the Unipile API that Ampliflow uses, does not touch the browser at all. There is no DOM to fingerprint. The laptop can be closed entirely. That is not a guarantee of invisibility, because volume and timing signals still apply, but it removes an entire category of detectable artifacts.
To be fair to the cheaper extension-based tools: Dux-Soup at $14.99 a month and Linked Helper at $15 a month are genuinely affordable, and many people use them without problems at low volumes. The risk is higher, but the price is lower. That is an honest trade-off. If you are running 5-10 connections a day on a warm account, an extension tool is probably fine. If you are running a sales team at any scale, the architecture starts to matter more.
Soft Warnings vs Full Restrictions
LinkedIn does not go straight to account restriction in most cases. There is a progression.
The first signal you will usually see is a CAPTCHA mid-session. Nothing dramatic, just a puzzle before you can continue. That is LinkedIn saying: we noticed something. Pay attention to it.
Next comes a phone verification prompt. LinkedIn asks you to confirm your number, sometimes one you did not register. This is a stronger signal and usually means your activity pattern has crossed a threshold in their internal scoring.
After that: a temporary connection limit. You will see a message stating that your account is limited from sending invitations for a period. This is not a restriction in the full sense, your account is still active, but outreach is blocked.
A full restriction comes last. The account is locked pending review, and recovery is not guaranteed. The path back from a full restriction is slow and stressful, and there is no formal appeals process that reliably works.
The practical lesson: treat a CAPTCHA as a stop signal, not a minor nuisance. When we see one in testing, we pause the campaign, drop volume for several days, and re-evaluate the timing settings.
What a Safe Activity Pattern Looks Like
Here is where we can be specific, based on what we do with Ampliflow accounts and what we observe in the broader practitioner community.
| Account Age | Safe Daily Connection Requests | Safe Daily Messages |
|---|---|---|
| New (0-4 weeks) | 5-10 | 10-15 |
| Established (3-12 months active) | 15-20 | 30-40 |
| Aged and warm (1+ year, strong SSI) | 20-25 | 50-60 |
These are conservative ceilings, not targets. The warm-up schedule matters enormously in the first few weeks. Starting at 5 per day and adding 2-3 per week is the approach we apply to every new account running through Ampliflow. We cap our own team's accounts at 20 connection requests per day even on aged profiles. The marginal gain from pushing to 30 is not worth the risk of triggering a review.
Timing jitter matters more than most people realise. Ampliflow adds randomised delays between actions, so the execution pattern looks like someone who gets distracted rather than a scheduler firing on the dot. It is a small thing that has a measurable effect on flag rates in our testing.
Auto-pausing on reply is worth highlighting separately. When a prospect responds, the campaign should stop immediately. Continuing to send follow-ups to someone who already replied is both a conversion mistake and a spam signal. This is built into Ampliflow's workflow logic, and it is one of the automation mistakes that burns accounts fastest when it is missing.
On Claimed Insider Knowledge
A lot of content about how LinkedIn detects automation reads like someone has a source inside LinkedIn's engineering team. We do not. Nobody publishing this stuff publicly does.
What is described here is based on publicly observed behavior: patterns the community has tested and documented, LinkedIn's own published policies, and what we have seen across accounts running through our platform in beta. LinkedIn's actual detection system is a black box. The signals above are real and worth managing. The exact thresholds are not knowable from outside, and anyone telling you they know the precise number is either guessing or making it up.
What we do know: the accounts that stay healthy longest are the ones that treat LinkedIn like a professional network rather than a dial to crank up. That sounds obvious. The number of campaigns we see configured to send 80 requests on day one of a new account suggests it is not obvious enough.
If you want the full practical checklist, the guide on avoiding LinkedIn restrictions goes deeper on day-by-day limits and what to check before scaling.
Written by Nivedita Verma, Design and Product at Ampliflow. Ampliflow is a cloud-based LinkedIn outreach tool built for founders and sales teams, with visual workflow building, real-time account safety scoring, and execution through the Unipile API so your browser stays out of it. Founding member pricing starts at $19 per month, for the first 100 seats only. See pricing details.