What LinkedIn actually allows, and what that means for your tooling

A lot of outbound tooling sells you a promise that only works if you don’t look too closely at how it’s kept: full automation of connection requests, messaging and profile visits, running on your account, around the clock, with no human involved. It’s a compelling pitch. It’s also, in most cases, built on top of unauthorised scraping or credentials handed to a bot logging in as if it were you — both things LinkedIn’s terms of service explicitly prohibit, and both things LinkedIn actively detects and enforces against.

The gap between the pitch and the terms

The pitch and the terms of service are not describing the same activity. LinkedIn’s platform agreement restricts automated data collection and automated actions taken on a member’s behalf without authorisation. A tool that logs into your account with a browser extension and clicks “connect” two hundred times before lunch is doing exactly the thing the terms forbid, whether or not the vendor’s marketing page mentions that risk. The account attached to that activity is the one that gets restricted, not the vendor’s.

If a vendor won’t tell you plainly whether an action runs through an approved API or a bot pretending to be you, that omission is the answer.

Two honest ways to actually act on LinkedIn

Strip away the marketing language and there are really only two compliant ways for software to cause something to happen on LinkedIn:

• An approved API integration. Sales Navigator and other official integrations exist, but they are limited in scope and require explicit approval — they are not a general-purpose automation surface, and treating them as one is how accounts end up flagged.
• A human performing the action from their own account. A person reviews the prospect, decides to send the connection request or the message, and does it themselves — a real login, a real click, a real decision.

Everything else — scraping profile data at scale without authorisation, or software logging in as a person to perform actions a human didn’t actually take — sits outside both of those, regardless of how the tool is marketed.

What honest tooling looks like

The practical answer isn’t to give up on efficiency, it’s to design the software around the constraint instead of pretending the constraint doesn’t apply. A human-assisted action queue does exactly that: the system decides who to contact and what to say — using scoring, ranking and AI-drafted messages — and hands each action to a real person as a one-click item in their daily queue. The engagement, the connection request, the message send: a human performs every one of them, from their own account, in a normal browser session that looks exactly like normal LinkedIn use, because it is normal LinkedIn use.

If official API access is granted for a specific action later, the connector behind that action can be swapped out without touching anything else in the pipeline — the scoring, the queue, the sequencing logic are all indifferent to which mechanism executes the final step. That’s the point of keeping the connector layer separate from everything upstream of it: compliance shouldn’t be a redesign, it should be a swap.

Questions worth asking any vendor

Before trusting a tool with your team’s LinkedIn accounts, a short list of questions tends to surface the truth quickly:

• Does an action happen through an official, approved API, or through a login that isn’t the API?
• If a human isn’t clicking the button, who or what is?
• What happens to our accounts if LinkedIn changes enforcement, and who bears that risk?
• Can we see, per action, whether it was a person or a script?

The unglamorous conclusion

None of this is a legal opinion, and every team should read the current terms of service themselves rather than take a vendor’s word for it — including this one. But the general shape of the honest answer doesn’t change: real accounts, real people making the final action, official APIs used only where they’re actually approved for the purpose. It’s a less dramatic pitch than “full automation while you sleep.” It’s also the version that’s still working in a year, on accounts that are still in good standing.