Do not index
Every week an agency owner sends me a screenshot of a post with ninety comments and asks the same thing. Why did this perform and produce nothing? The answer is not comforting. Roughly a third of those comments were not written by a human being, and most of the humans who did comment were farming reciprocity rather than evaluating you as a vendor. Comment count is no longer a performance metric. It is a measure of how many automated accounts and pod members found your post, which is a completely different signal from whether a buyer read it and moved.
The number behind this is not soft. An analysis by AI detection startup Pangram Labs, based on 57,000 public LinkedIn posts, found 30% of all comments posted on LinkedIn between April and June 2026 were entirely AI-generated. That is the baseline noise floor of the platform now. It arrives in the same quarter LinkedIn began rebuilding its comment display and ranking to push more replies into the feed, according to Social Media Today. The platform wants more conversation. The market is supplying it synthetically.
What that means practically is that the engagement number on your dashboard has quietly become two numbers stacked on top of each other, and only one of them carries commercial meaning. There is machine volume, which scales with how generic and prompt-friendly your post is. And there is human intent, which does not scale at all and never has. Generic posts attract more of the first kind. A post specific enough to be wrong attracts fewer comments and better ones.
Who should care about comment quality on LinkedIn
This matters if you run an agency somewhere between $200k and $2M in revenue and your pipeline depends on ten to thirty real conversations a quarter. It matters if you are a ghostwriter charging $5k to $30k per month and your client renews or churns based on whether their inbox filled up, not on whether the screenshot looked impressive. It matters if you are a founder writing your own content with maybe four hours a week for it, which means every post has to do actual work rather than decorate a profile.
This is not for you if you sell a low-ticket product where raw distribution genuinely is the mechanism. If you need 50,000 impressions to move forty units of something at $29, volume is your business model and the bots are, in a grim way, part of your reach. Skip this if your goal is to grow a follower count from 5,000 to 50,000 to sell sponsorships. Different game, different scoreboard. And if you are still reporting comment totals as the headline number in your client updates, this article will not change your model. It will only make that model harder to defend the next time someone asks what those comments turned into.
The Real Reply Test
Here is what I would actually do. I use what I call the Real Reply Test, and it takes about ten minutes a week. Take your last ten posts. Ignore the totals entirely. Open the comments and count only the replies that meet three conditions. The comment references something specific from the post that a model could not have generated from the first line alone. The person who left it holds a title that could sign or influence a contract with you. And the comment moves the argument somewhere, whether by disagreeing, adding a case, or asking a question that proves they read past the hook.
Everything that fails all three is noise. Congratulations, agreed, great post, this resonates, thanks for sharing. Those are not weak signals. They are zero signals, and increasingly they are not even human.
What you get from ten minutes of this is a number that is usually brutal and always useful. A typical account I audit shows 40 to 90 comments per post and two or three real replies. Sometimes zero. The ratio does not correlate with the totals, which is the entire point. I have seen a post with 14 comments produce six qualified replies and two calls, sitting directly beside a post from the same account with 130 comments and nothing behind it. Track real replies across 20 posts and you will find that a small handful of topics carry almost all of them, and those topics are almost never the best performers by raw engagement.
Once you have that list of topics, the work becomes obvious. You write more of the three that produce real replies and you stop writing the ones that produce volume. That is the whole optimization. It is also why judging performance from the native dashboard keeps producing accounts that look healthy and generate nothing, a pattern I have worked through in more detail in how to measure LinkedIn success.
The strategic piece is this. Detection will improve and the platform will keep tightening, but the direction of travel is already set, and it is not toward a cleaner comment section. It is toward a feed where public engagement numbers carry almost no information about commercial outcome, and where the operators who understood that early rebuilt their reporting, their content decisions, and their client conversations around a different number entirely. If your business sits downstream of a metric that is 30% synthetic and climbing, you are steering by an instrument that is drifting. The agencies still standing in two years will be the ones who rebuilt the instrument while everyone else was celebrating the readout.
