Do not index
Why is my engagement up but my pipeline flat? I get some version of that question every month, usually from an operator who just sent a client report full of green arrows and cannot explain why nothing came of it. There is now a straightforward answer.
Your comment count is measuring machines. Roughly a third of it, in fact. Stop reporting it as a proxy for interest, because it stopped being one.
Pangram Labs analyzed 57,000 public LinkedIn posts and found that "30% of all of the comments posted on LinkedIn between April and June this year were entirely AI-generated." That sits directly against LinkedIn's own Q2 numbers, reported by Andrew Hutchinson at Social Media Today, showing an 18 percent year over year increase in time spent in post comments and content consumption up 10 percent year over year. LinkedIn is also re-ranking comments by relevance to each individual reader and surfacing more timely discussion in the feed, a change first spotted on Threads by creator economy analyst Lindsey Gamble rather than announced in a company blog post.
Read those two facts together and the picture is not ambiguous. Comment activity went up. Conversation did not. The metric inflated while the thing it was supposed to represent stayed flat or got worse.
What a third of comments being fake actually costs you
The damage is not that bots are polluting your replies. The damage is that you built a decision system on top of a number that no longer describes reality. If you decide what to write next based on which posts got the most comments, and 30 percent of comments are generated, you are letting a language model vote on your editorial calendar. It votes for the posts that are easiest to reply to generically. Broad opinions, obvious takes, anything a model can affirm in one sentence without understanding it.
So the feedback loop selects for the blandest thing you publish. You post something sharp and specific about pricing structure inside a 3 person agency, it gets 11 comments from people who actually run one. You post a soft platitude about consistency, it gets 60. The dashboard tells you to write more platitudes. You comply, slowly, over about six months, and by the end you cannot work out why an account with better numbers than last year produces fewer conversations that go anywhere.
This is for agency owners between $200k and $2M in revenue who report content performance to clients or to themselves and need those reports to mean something. It is for ghostwriters charging $5k to $30k per month whose renewal conversations hinge on demonstrating value. It is for founders who have spent two years building to 5,000 to 50,000 followers and are now trying to work out what that audience is worth.
This is not for anyone whose content goal is genuinely brand awareness at scale, where volume of surface contact is the point and the composition of that contact matters less. Skip this if you are running content as a top of funnel awareness play for a product with a self-serve checkout. And if you are still selling engagement growth as your service deliverable, this article will not change your model, because the entire argument undermines what you are being paid for.
The Named Reply Test
Here is what I would actually do instead. I use what I call the Named Reply Test, and it takes about four minutes per post.
Open the comments and count only the replies that could not have been written by someone who did not read the post. That means a reply that references a specific claim you made, disagrees with a particular line, adds a detail from the person's own operation, or asks a question that only makes sense in context. Generic affirmation does not count. Neither does a tagged name, a restatement of your own hook, or anything that would work verbatim under a different post.
Then check who those people are. Not follower count. Whether they are in your buying population. A post that produces four named replies from three agency owners and a fractional CMO outperformed a post that produced 60 comments from nobody in particular, and it is not close.
A reasonable benchmark from what I see across accounts doing this seriously is three to eight named replies on a strong post. If you are getting 40 comments and two of them pass the test, your reach is fine and your resonance is broken, and those are different problems with different fixes. Reach problems are format and timing. Resonance problems are that you are not saying anything a specific person needed to hear.
The deeper issue is that most operators never established what they were measuring before they started measuring it, which is the same reason measuring LinkedIn success almost never happens inside the analytics dashboard. The dashboard reports what is countable. Whether a specific person now understands what you do well enough to refer you is not countable, and it is the whole game.
What this means for where you end up is a question of drift. Content operations that keep optimizing against comment volume will keep getting better at producing content that machines find easy to respond to, and the drift is gradual enough that nobody notices the voice flattening. Operations that switched to counting real replies from real buyers will publish less, get worse numbers on paper, and build a smaller audience that actually converts. Two years from now those two accounts will look similar in a screenshot and function nothing alike. The screenshot is not what you are building.
