Playbooks · 5 min read
We Cannot Predict a Reel. We Can Prepare for the Spike.
We do not predict Reel performance. There is no scoring model here, no "this will do well" indicator, no analysis of your draft, no forecast of views or reach. We do not read your videos at all. This product replies to comments after a post is live, and it has no opinion about the post.
That would be true even if we wanted to build it, which we do not, because the honest position on prediction is uncomfortable: nobody outside Meta has the inputs. Whatever a prediction tool is scoring, it is scoring proxies — length, aspect ratio, caption sentiment, a hook rubric — against outcomes it cannot observe the causes of. A confident number produced from that is a guess wearing a suit.
So this post is short, and it is about the thing that is genuinely predictable: not whether a Reel will travel, but what happens to your inbox when one does.
The spike is the part you can plan for
A Reel that gets picked up does not produce a gradual increase in comments. It produces a burst, usually within hours, mostly from people who do not follow you, and mostly saying the same three or four things.
That burst is an operational event before it is a marketing one. It is the moment your automation setup is tested at a volume it has not seen, and the three things that bite are always the same.
The rate budget paces sending. Replies do not go out in a single burst even when the comments did, and during a spike you will see events waiting rather than failing. This exists so an account does not behave in a way the platform reads as abnormal, and it is one of the reasons the account survives its best day.
Quiet hours hold replies outside the window you set. A Reel that takes off at eleven at night, with quiet hours running from nine, means several hours of comments queued for the morning. That is fine and usually correct, but it surprises people who have never watched it happen.
Credits are the one with a floor you control. A templated reply is free and unlimited — it costs nothing whether it sends five times or five thousand. A credit is spent only when the AI writes a new reply, one credit per reply. A spike answered by templates costs nothing. A spike answered entirely by generated replies is the fastest way to find the edge of your balance.
The tail is longer than the spike
The second predictable thing is that a Reel does not stop when the burst does.
Comments are answerable for seven days from when each one lands, not from when the Reel was posted. A Reel that gets rediscovered a fortnight later produces live, answerable conversations again — provided the automation is still switched on, which is the part people get wrong after they have mentally filed the post as finished.
The messaging window is the other clock, and it is shorter. Once somebody comments or messages, there is a 24-hour window to send them a message, and only a message from them restarts it. There is no catching up on a backlog three days later. Whatever was going to be said had to be said at the time.
Both of those together mean the useful preparation is before, not during. During a spike there is very little to do except read what is coming in.
What to actually watch while it happens
- Activity, filtered to blocked. This is the honest report. A stack of paced sends and quiet-hours holds is normal. A stack of window closures on a Reel that just posted is not, and usually means the automation was off when the burst started.
- The comments that do not match anything. A travelling Reel brings questions your trigger list has never seen, because strangers ask different things from followers. Write them down; they are next week's keyword additions.
- Your own hands. The moment you reply to someone by hand, the automation stands down for that conversation and stays down. During a spike that is worth knowing, because it means jumping in on a thread does not create a duplicate.
Why we would not build the prediction feature
Not because it is hard. Because a prediction that is wrong is worse than no prediction.
If a tool tells you a Reel will underperform and you shelve it, you have made an irreversible decision on the basis of a number with no provenance. If it tells you a Reel will do well and it does not, you have learned nothing and lost a day of confidence. Neither outcome improves the account.
The alternative is unglamorous and works: post the thing, read the comments, and let the question distribution tell you what the audience wanted. That feedback is slower than a score and it is real. The content audit post lays out how to read it systematically.
The only reliable signal a Reel gives you is what people ask afterwards. That arrives after publishing, which is inconvenient and true.
The short version
There is no prediction here and there is not going to be. What there is: a setup that behaves sensibly when a post outperforms, refuses in ways it records rather than silently, and costs nothing extra when the answer was one you had already written.
If you want the surge-handling side in more depth, the Reels-to-DM post covers the flow itself, and the safety post explains the ten checks and why several of them exist precisely for the day something takes off.
