A like is not a signal

everyone's piping linkedin into clay right now. almost nobody filters it first.

So they're enriching garbage at scale and calling it pipeline.

Here's the problem. Raw engagement isn't intent. A like means someone's thumb moved. A comment might mean they disagree with you, or they're a competitor, or they just want reach for their own post. A follow might mean nothing more than your headline landed that day. None of that tells you whether the person can actually buy from you.

That used to be a slow, human mistake. You'd waste a rep's afternoon on a bad lead and move on. Now it's an automated one. Clay will enrich anyone you feed it. Claude will write a personal message to anyone you point it at. Make will fire the workflow no matter who's on the other end. The tools don't ask whether the signal was real. They act, fast, on whatever you give them. Garbage in, garbage at scale.

We've measured this closely. More than 300,000 signals from 150 sample workspaces. That's a big enough sample to stop guessing. And the numbers say most of what teams treat as a signal isn't one.

If you're piping linkedin into a workflow, this is the part that decides whether the workflow is worth running. If you're an agency proving engagement turned into pipeline, it's the part that decides whether you've got a story or just a screenshot.

So what is a signal?

Start with a clean definition, because most of the confusion is just sloppy language.

A linkedin signal is engagement from someone inside your ICP. That's it. Someone who fits the profile of a buyer, doing something that shows attention. Everything else is noise.

A like from a student is noise. A comment from a competitor is noise. A profile view from someone three industries away is noise. They're all "engagement." None of them are signals, because a signal isn't about the action. It's about who took it.

The question is never whether someone engaged. It's whether the right someone did. Engagement is the event. A signal is the event plus the fit. Drop the fit and you're left with a number that feels like progress and predicts nothing.

How much of it is actually noise

Here's the number that should change how you build.

Across all the signals we analyzed, 84% came from people who don't match their workspace's ICP. Only 15.6% did.

Read that again. For every hundred interactions your team celebrates, about sixteen are worth acting on. The other eighty-four are students, job seekers, competitors, peers, and people who will never be in a position to buy. They're not bad people. They're just not your buyers.

Now put that back in the AI workflow. Pipe raw linkedin engagement into Clay without filtering first, and 84% of your enrichment spend, your message volume, and your new CRM records are aimed at people who won't convert. You didn't build a pipeline. You built a very efficient way to burn credits.

And it compounds. Everything downstream inherits the 84%. Your enrichment bill, your send volume, your reporting, the context you hand your AI, all of it scaled on a base that's mostly wrong. The faster the stack runs, the faster it runs in the wrong direction. That's the part that should worry anyone building a GTM motion on linkedin data: the tools got faster, the filtering didn't.

Not all signals are equal

Even inside the 15.6% that match, the signals aren't worth the same.

The strongest one we see is connection acceptance. When someone accepts a connection request from a campaign, 41.1% of those people match ICP. Set that next to a passing like or a follow and it isn't close. Accepting is a small commitment, and commitments predict.

This isn't a rare event you have to wait around for, either. Across campaigns the acceptance rate sits at 38.4%. It's a steady, high-quality signal most teams already generate and never rank.

Now the counterintuitive part. Volume of engagement does not tell you who fits. ICP leads engage 2.4 times on average. Non-ICP leads engage 2.3 times. Basically the same. So you can't separate your good-fit buyers from the noise by counting how active someone is. The loud ones aren't the right ones. Fit is a different question from frequency, and you have to answer it on its own.

Then there's the trap that costs teams the most. 45.6% of ICP leads engage exactly once. Once. They view a profile, or accept a connection, or leave one comment, and then they go quiet.

That's nearly half your good-fit signals giving you a single shot. You don't get a second one, so you can't afford to miss the first. If your process needs a rep to notice it, tag it, and act on it by hand, most of those leads are gone by Thursday.

Named accounts change everything

One lever moves signal quality more than any other. Target accounts.

When teams match incoming engagement against a named account list, 61% of those signals are ICP matches. For teams working organically, without target accounts, it's 13.1%. That's a 4.7x difference in signal quality from a single decision.

It makes sense. If you've already chosen the companies you want, then engagement from those companies is meaningful by definition. You stop asking "is this person interesting?" and start asking "is this one of the accounts I'm trying to win, and are they paying attention right now?"

The lift isn't subtle. 61% against 13.1% means a team on named accounts pulls roughly five qualified signals for every one an organic team finds in the same pile of engagement. Same linkedin activity. Different starting point.

Here's what surprised us. Only 45 of 152 workspaces use target accounts at all. Most teams are sitting on the single biggest lever for signal quality and leaving it untouched. They're filtering the whole internet when they could be watching a short list they already care about.

What to do about it

The fix is an order-of-operations change, not a new tool.

Qualify against your ICP before the signal touches your CRM or your AI workflow. Not after. The whole point of qualification is to decide what's worth acting on before you spend money and automation acting on it. If Clay enriches it and Claude writes to it and then you check fit, you've run the sequence backwards. You already paid for the noise.

So the rule is simple. Fit first, then enrich, then act. Capture the engagement, score it against who you actually sell to, and only let the matches through to the expensive, automated part of your stack. If you're weighing how different tools handle that step, it's worth comparing them directly rather than assuming they all filter the same way. Most don't filter at all.

This is the layer teamfluence sits in. It qualifies linkedin signals against your ICP before they reach the rest of your tools, so what flows into Clay, Claude, your CRM, or your sequencer is signal, not noise.

The takeaway

The teams winning on linkedin aren't the ones with the most engagement. They're the ones who can tell signal from noise before they act on it.

Download the full linkedin signals benchmark report. The complete data behind this post: more than 300,000 signals, 150+ sample workspaces, the full signal hierarchy, and the ICP match rates by engagement type.

Want the next breakdown when it's out? Join the newsletter.