Key takeaways
- The only number the radar cares about is a post's views divided by that account's own median views. Raw view counts just find big accounts.
- Median, not average. One runaway post drags an average somewhere useless and hides everything around it.
- Three times the median is the line. Below that it is noise, above it something about the post actually worked.
- Posts under a week old get held back. They have not finished accumulating views and they will lie to you.
- What transfers is the format, not the topic. The pancake reel is not about pancakes.
The old way of finding short form ideas was scrolling until something felt good. That is not research, that is a mood. And it has a specific failure mode: you keep finding posts from accounts with a million followers, you copy the format, it does nothing, and you conclude the format is dead. It was never the format. It was the audience that was already there.
So we built a radar that cannot see follower counts.
The whole method is one division
Scrape a watchlist of accounts. For each account, find the median view count across its recent posts. Then for every post, divide its views by that median. That number is the only thing that gets ranked.
Median matters more than it sounds like it should. If an account has one post at thirty million views and thirty posts at forty thousand, the average is somewhere around a million and literally every normal post looks like a failure against it. The median sits at forty thousand where it belongs, and the thirty million post shows up as the enormous outlier it actually is.
Three times the median is where we draw the line. That threshold is not sacred, it is just where the signal starts being worth a person's attention. Everything under it gets dropped without being read.
The four rules that keep it honest
The age gate comes first. Anything under seven days old comes out of the ranking entirely and goes into a separate list called still climbing. A two day old post at 4x is not a 4x post, it is a post that had a good first two days. Ranking it next to finished posts makes recency look like quality.
Engagement rate breaks ties, and it is just likes plus comments over views. A post at 5x with weak engagement got pushed by the algorithm. A post at 3x with strong engagement got pushed by people. The second one is the better thing to learn from, even though the first one has the bigger number.
Then everything already surfaced gets skipped. There is a file of every post the radar has ever flagged, by post ID, and those are skipped forever. Without it, the same monster reel from eight months ago wins the queue every single week and the radar becomes a machine for reminding you of one video.
The last rule is the one that surprised me: write down each account's median every single run. Medians move. If an account is growing, its median rises, which means the 3x bar quietly gets harder to clear. Without tracking that, you look at a slow week of outliers and conclude the account went cold, when what really happened is the account got bigger and your measuring stick moved with it.
Reading the result
The output is a ranked table, capped at fifteen rows, and for the top three there is one line explaining why it worked. That line is the whole point of the exercise, and it has to describe the mechanism, not the subject.
There is a food creator whose reels we studied closely. The flagship one is about pancakes. The pancakes are irrelevant. What makes it work is word level captions that land in time with the speech, plus a punchline held all the way to the last second so leaving early costs you the joke. That structure works for a plumber, a dentist, or a CRM demo. The pancakes work for pancakes.
This is why "make content like them" is useless advice and why the radar writes the mechanism down. You cannot copy a topic across industries. You can copy a structure anywhere.
Same math, different platform
The YouTube version scores videos against their channel's median instead. Two rules there are worth stealing. Weight the watchlist toward channels near or slightly above your own size, because a huge channel's title formats do not transfer down. And run Shorts and long form as completely separate passes, since their view scales are so different that mixing them makes both medians meaningless.
What a local business should take from this
You do not need a scraper. You need to stop comparing yourself to the wrong baseline, and the fix is the same division done by hand.
Take your own last twenty posts. Find the middle one by views, not the average. Now look at which of your posts beat that middle number by three times or more, and ask what those have in common. Not what they were about. How they opened, how long they were, whether the first frame had a face, whether the payoff was at the front or the end.
That is your format, discovered from your own audience instead of borrowed from somebody with a hundred times your reach. Most businesses already have two or three posts that beat their own median badly and have never noticed, because the raw numbers looked small next to whatever is trending.
Doing this properly across every platform, every week, is exactly the kind of job that should not be a person's Monday morning. As a GHL consultant in the Bay Area, most of what I build is this shape: a boring calculation, run on a schedule, that hands you a short list instead of a feed. If you want that running for your business, come find us at optechsol.llc.