OUTLIER: THE RULE AND THE PROMPT


THE RULE

A reel counts as a winner only when its views are at least twice its account's median.
The median comes from ten or more of that account's own recent reels, all read in one sitting.
Compare reels with reels. Leave out carousels, photos and long videos.
Leave pinned reels out. They were chosen, and they're usually older.
A reel under 7 days old is still settling. Its number will still move, so don't call it yet.

For the sheet, a reel under half its account's median is a miss. Everything in between is normal.
The miss line is mine, for picking clear contrasts to watch. The winner line is the rule.

The median is the middle number once the views are sorted.
With an even count, it's halfway between the two middle numbers.
One viral reel can't drag the middle up the way it drags an average.


HOW TO USE IT

1. Use an AI agent that can open web pages in a browser. It has to open each account and read the view counts.
2. Fill in the blanks in square brackets in the prompt below.
3. Paste everything from COPY FROM HERE to COPY TO HERE as your first message.
4. Open outlier-sheet.csv in Google Sheets or Excel. Paste the agent's rows under the EXAMPLE rows, starting in column A.
5. Copy the account_median, multiple and band cells from the last EXAMPLE row, and paste them down over your rows. The sheet does the sums, which checks the agent's.
6. Delete the EXAMPLE rows.
7. Watch every winner, and a few misses from the same accounts. Fill in first_line, format, face, what_is_shown and notes as you go.

If the sheet's median and the agent's summary disagree, trust the sheet and check the views.
The median formula uses FILTER, which works in Google Sheets and current versions of Excel.


BEFORE YOU START

Logged-out pages change, and some accounts won't load.
On 30 September 2026, a logged-out Instagram page showed only an account's latest 12 reels.
Scrolling for more asked for a login.
If three or more of those 12 are pinned, fewer than ten are left to count. That account gets no winners until you have more.
If the agent can't read an account, it should say so and skip it. It should never guess a number.
You can always open an account's reels yourself and copy the views into the sheet.


WHAT TO NOTE WHEN YOU WATCH

Watch the winners and the misses from the same account side by side. For each one, note:
- The first three seconds: what's said, what's on screen and any words on screen, word for word.
- What's on screen: a face, a real screen, a product, a whiteboard, b-roll or a diagram.
- Face or not: is a person on camera? From the start, later, or never?
- The ask: what it asks at the end, such as comment a word, follow or save. Or nothing.

Then look for what the winners share and the misses lack, on the same account.
A trait that shows up in two or more winners and is missing from a miss is worth testing.
One winner on its own is an idea to test, not a rule.
Borrow the shape, never the words. Don't repost anyone's footage or lines.


============ COPY FROM HERE ============
You are a research assistant with a web browser. Find which recent reels in my niche did far better than their own account's usual, and which did far worse.

My niche: [your niche, such as "AI tools for small businesses" or "home cleaning"]
Who I make videos for: [your audience]
Accounts I already know: [their handles, or "none"]

1. Find accounts. List 10 to 15 Instagram accounts that post reels in this niche for this audience. Start with the ones I named. Pick accounts of different sizes, not only the biggest.

2. Pull the reels. For each account, open its reels and record its latest 12 to 20 reels: the link, the date posted and the view count. Read all of one account's counts in one sitting.

3. Skip pinned reels. Don't list them and don't count them.

4. Write exact numbers. If a count is rounded, such as 12.3K, write 12300 and put "rounded" in notes.

5. Never guess. If an account won't load, is private or hides its counts, say so and skip it. Don't estimate a number or fill a gap from memory.

6. Work out each account's median. Sort the views of the reels you listed and take the middle number. With an even count, take the average of the two middle numbers. If an account has fewer than 10 reels with views, write "fewer than 10 reels" and call no winners or misses for it.

7. Work out each reel's multiple: its views divided by its account's median, to two decimal places.

8. Sort each reel:
   - winner: a multiple of 2.00 or more
   - miss: a multiple under 0.50
   - normal: everything in between
   - settling: posted less than 7 days ago, whatever its multiple. Don't call it a winner or a miss yet.

9. For every winner and every miss, add four things. Its first line: the first words said or shown, word for word. Its format: talking head, screen recording, split screen, words on video or b-roll. Whether a face is on camera: Y or N. And what is on screen. Write "not seen" for anything you couldn't check.

Give me back three things.

A. A summary per account, one line each: the handle, how many reels you counted, the median, the winners and the misses.

B. The rows for my sheet, as a table with these columns in this order:
account, reel_url, posted, views, account_median, multiple, band, first_line, format, face, what_is_shown, notes
Write posted as year-month-day, such as 2026-09-20. Leave account_median, multiple and band empty: my sheet works them out.

C. Up to three things the winners share that the misses lack, on the same accounts. Name the reels behind each one. If a pattern rests on one reel, say so: it's an idea to test, not a rule.

Don't copy anyone's footage or words. Describe what they did instead.
============ COPY TO HERE ============