How to find viral Reels worth recreating

The feed is optimised for your attention, not for your research. It shows you what you will watch, which is mostly large accounts and mostly videos you have already seen, and it gives you no way to ask a question of it. There is no filter, no sort, and no way back to the video you scrolled past four days ago.

The usual fix is a spreadsheet. It works for about three weeks. Every row is typed by hand from a video you had already found, so the sheet can only ever hold what you happened to watch, the view counts are stale the day after you paste them, and nothing in it tells you whether a million views was a good result for that account or an ordinary Tuesday.

A video’s outlier score is its views divided by the median views of the creator who posted it. A Reel on 400,000 views from an account that usually does 20,000 scores 20x. The same 400,000 views from an account that usually does 500,000 scores below 1.

That division is the whole point. Raw view counts measure audience size, which you cannot copy. Dividing by the creator’s own baseline measures how far the video outran what that same person normally does, and what is left over is mostly the format. A high scoring video from a small account is the most useful thing in the library, because nothing about its result is explained by reach.

A creator with too few collected videos to have a real baseline gets no score at all rather than a guessed one. Two videos cannot establish a median, and a 47x built on two videos is noise being sold as a finding.

Search runs over the caption, the verbatim hook, and the transcript text, blending keyword matching with semantic similarity. So a search for a phrase somebody says out loud in the first two seconds finds the video, even when the caption is a single emoji and the hashtags are wrong.

Transcripts are held on a growing subset of the library rather than on every row, and rows without one are still searchable on caption and hook. Filters cover niche, format, a view floor, duration, posted date, and outliers only. Any filter set can be saved and reopened.

A search you have to remember to run is a search you stop running. Rituals are standing searches Minnie runs on a schedule and reports on, in the product and by email. There are 6:

• Morning Brief. The five biggest outliers in your niche from the last day.
• Recreatable Watch. Outliers from accounts under 10k, where the format is doing the work.
• Format Alert. A format several creators started beating their own median with at once.
• Second Wave. Something you saved is being copied. Here are the copies.
• Weekly Plan. Seven shapes off the week’s outliers, Sunday night.
• Comment Triage. Your last three posts’ comments, sorted into questions, buying, collab and what they asked for. She reads them, she does not answer them.

Silence is a correct outcome. If nothing in your niche beat its own baseline yesterday, nothing is sent. A ritual that fires every morning regardless is a notification, not a finding.

Minnie is the front door to the library. You ask her a question in ordinary language, she runs the searches, and she answers from what the searches returned, with the videos she used rendered as cards beside the answer. She does not answer from memory and she does not invent a statistic the query did not produce.

She also keeps track of what she has already shown you, so asking the same question twice brings back different videos rather than the same twelve.

People who publish short-form video and need to decide what to make next: creators, founders running their own marketing, social media managers holding several accounts, and editors looking for a reference before a cut.

It is not an editor, a scheduler, a caption generator or a video generator. Minnie finds the video and explains what is working in it. You make the thing.

Max is $49 per quarter, which is $16 a month, or $132 per year, which is $11 a month. There is no monthly option and both periods are prepaid in full. Every subscription starts with 14 days free.