The best reply opener is no opener: empty starts outperform fillers

The stat. The agent picks a reply opener — a lead-in like "ngl," "tbh," "hmm," or nothing at all (jumping straight into the substance). In the learned weights, the empty opener carried the heaviest default weight (~30%) and canned openers were capped at ~15% combined, because leading with filler tended to underperform leading straight with the point.

How it was measured. Each reply logged which opener it used; openers were scored for engagement and rolled into weights, with a joint feature also tracking opener × hour-of-day. The no-opener "just say the thing" replies held up best on average, while heavy reliance on the same casual opener ("tbh" on everything) degraded — partly a novelty/repetition effect the model penalizes directly.

The honest caveat. This is one voice on one account, n≈983 replies, and openers are a small effect compared to whether the reply has a specific, mirrored detail. The casual openers ("ngl," "tbh") aren't bad in isolation — they read human and break up the feed — they just lose their value when overused, which the 15% cap exists to prevent. So the finding is really "don't lean on a crutch phrase," not "openers are forbidden."

The tactic it implies. Lead with the substance, not the throat-clearing. "Here's the thing," "Great point," "So true" — these openers signal a generic reply before you've said anything, and readers skip them. Open cold with the specific detail or the actual take. Keep your casual lead-ins ("ngl," "tbh") as occasional seasoning, not a default. The first three words of a reply are prime real estate; don't spend them on filler everyone's already learned to scroll past.