Why retweets are a bad growth metric: too noisy to trust

The stat. Of the four obvious engagement actions — like, reply, bookmark, retweet — the study gives retweets zero weight. The engagement score is likes + 3×replies + 5×bookmarks; retweets were tested and deliberately left out as too noisy to be a reliable signal of whether a post actually delivered value.

How it was measured. During the study's outcome scraping, retweet counts were collected alongside the others but found to be dominated by tail events — one mid-size account resharing a post could multiply its retweet count overnight in a way that had little to do with the post's broad resonance or its tendency to produce follows. Including retweets made the score lurch on a handful of posts; excluding them made the score track durable outcomes (bookmarks, follows) far more smoothly.

The honest caveat. "Noisy" doesn't mean "worthless" — a retweet from the right large account is genuinely valuable for reach, and this account simply didn't get enough of them to model well (n≈983 posts/replies, one account). For an account that regularly gets reshared by big accounts, retweets might deserve a real weight. The exclusion is a pragmatic modeling call for this data, not a universal law.

The tactic it implies. Don't let a viral retweet count fool you into thinking you've cracked the code. Bookmarks and follows are the durable signals; retweets are lottery noise. Chasing the reshare — writing for the dunk, the screenshot, the "this needs to be seen" — optimizes for variance, not for the steady, compounding value capture that actually grows an account. Write the thing people save, and let the occasional retweet be a bonus rather than the target.