Find Outlier Videos in Any Niche With Claude

Find the videos that got far more views than their own channel usually gets, and why each one worked. You send Claude one message, and the kit is free.

Watch the video

What goes with this video

Software

Free

Outlier Finder

Type a topic and see which videos got far more views than their own channel usually gets, ranked, with a CSV file. It only reads public numbers. Free, and nothing is locked.

  • A tool that searches a topic on YouTube and ranks every result by how many times its channel's normal views it got (its outlier number), as a table and a CSV file
  • A tool that saves one video's thumbnail and its English transcript with time marks. It never downloads videos or their sound
  • 2 Claude skills: find the outliers for a topic (with why each one worked and nearby niches to study), and get one video's numbers, thumbnail and transcript
  • A one-time guided setup, about a minute, and a health check
  • A guide to Claude in Chrome, so Claude can browse YouTube like a viewer
See what's included

This needs a computer. On a phone? Email yourself the link for later.

Optional. We send the link once. Nothing else unless you tick the box.

  • No email and no card needed
  • Zip, 41 KB
Get it free

or

Software

Outlier Finder Pro

Honestly, you don't need this. The free kit finds the outliers, and Claude can help you build the extras yourself.

Buy it if you're lazy. We already did the building, so you can skip straight to the studying.

Same result either way. The only difference is who spends the afternoon.

  • Everything in the free kit, unchanged
  • Hook workbook: how the top outliers open their first minute next to normal videos from the same search (how fast they talk, how soon they say what the title promises, who they talk to, how long the intro is, the moves they use), then a plan for your own opening, in your own words
  • Watchlist: a saved list of topics and channels, re-run politely, each run kept in a local database, and one page showing how each video's outlier number and each channel's normal moved, plus a "rising now" list
  • Study board: one printable page from any results file, with the top outliers' thumbnails side by side, titles grouped by the promise they make, lengths, numbers and dates
  • 5 Claude skills, the three pro ones included
See what's included
See what it adds

What each one has

QuestionOutlier FinderOutlier Finder Pro
Can I find the outliers in my niche, the way the video shows?YesYes
Can I get a video's thumbnail and transcript?YesYes
Can I follow topics and channels over time and see what is rising?NoYes
Can I get one printable page of the top outliers side by side?NoYes
Can I see how the top videos open their first minute, next to normal ones, and plan my own opening?NoYes
Do I need to give an email?NoYes, checkout needs one for the receipt

Rather not download a zip?

Send these messages to Claude instead, one at a time. Change the topic in quotes to yours. It is the same text as NO-DOWNLOAD.md in the kit.

You don't need this kit's zip. Here are the exact messages to send Claude to get the same thing. Copy them one at a
time. Change the topic in quotes to yours.

--- ROUTE A: NOTHING TO INSTALL (CLAUDE IN CHROME) ---

Needs Chrome and the Claude in Chrome extension (see CHROME.md, or ask Claude how to install it). Open the
Claude panel in a new tab and send:

1. Open YouTube and search "home workout". List the top 20 results in a table: title, channel, views, how long
   ago it was uploaded, and the length. Skip anything under 4 minutes.

2. For the 8 channels in that table with the most views, open each channel's Videos tab. Read the views of its
   last 20 videos and tell me the median (the middle number). Then divide each video from my table by its
   channel's median. That's its outlier number. Rank the videos by it, highest first.

3. For the top 3, tell me in plain words why each one probably beat its channel: the title's promise, the
   thumbnail idea, the length compared with the channel's other videos, and what happens in the first 30
   seconds (open it and read the description and chapters).

4. Open the top video and read the sidebar and end-screen suggestions. Name 3 other topics with a similar
   viewer. Do the same ranking for one of them and tell me which format I could borrow: hook, structure, pace,
   thumbnail idea. Structure only, never the words.

The numbers on YouTube's pages are rounded, so this route is a bit rougher than the tool. Good enough to find the
big outliers.

--- ROUTE B: CLAUDE BUILDS THE TOOL FOR YOU (CLAUDE CODE) ---

Needs Claude Code, or the Code tab of the Claude desktop app, plus Python 3.10+ and Node.js 22+. Open an empty
folder and send:

1. Make a Python virtual environment in this folder called .venv and install yt-dlp[default] into it. Check
   that Node.js 22 or newer is installed, and tell me how to get it if it isn't.

2. Write tools/outliers.py, a command line tool for public numbers only (never downloads videos or audio,
   never uses browser cookies). yt-dlp always runs with "--js-runtimes node" and "--sleep-requests 1.5".
   A channel's normal: first read its public feed
   https://www.youtube.com/feeds/videos.xml?channel_id=<the UC id> (latest 15 uploads, exact views in
   media:statistics), skip links with /shorts/ and videos under 7 days old, and take the median views.
   If fewer than 5 videos are left, read the channel's /videos tab with yt-dlp --flat-playlist instead
   (last 30 uploads, 4 minutes or longer). Fewer than 5 there too: say "too few".
   Save every search, channel and video lookup in results/cache.json with the time, and reuse it for 24 hours.
   Subcommands:
   - score <video>: the video's views, upload date, length and subscribers, its channel's normal,
     outlier = views / normal, views per subscriber.
   - channel <@handle>: the normal, and the videos ranked by views / normal.
   - search "<topic>" [--n 20] [--min-minutes 4] [--top 10] [--max-channels 15] [--within week|month|year]:
     "ytsearchN:<topic>" with --flat-playlist (for --within, YouTube's results page with its upload date
     filter), drop videos under --min-minutes, look up each channel's normal (biggest videos first, at most
     --max-channels, one at a time, a random 1 to 2.5 second wait between), print a ranked table
     (outlier, views, normal, minutes, channel, title) and save all rows to results/<topic>-<date>.csv.
   On the first "429" or "not a bot", stop, save the CSV and the cache, and print: "YouTube says slow down.
   Wait 15 to 30 minutes, then run the same line again: it picks up where it stopped."

3. Write tools/transcript.py <video>: read the video once with yt-dlp (-J --skip-download). Pick only the
   original English captions: a creator-made "en" track, else the automatic "en-orig" track. Never a
   machine-translated one. Then download just that track with --load-info-json, --sub-format vtt and
   --sleep-subtitles 5, and save the thumbnail from its link. Turn the VTT into clean text without the
   repeated lines, with a [mm:ss] mark every 30 seconds. If there are no original English captions, say so
   plainly. If YouTube blocks it, tell me to open the video and click "Show transcript" instead.
   Never download the video or audio.

4. Run: tools/outliers.py search "home workout" --n 20. Show me the table and explain the top 3 in plain words.

5. Now help me find 2 or 3 other niches with a similar viewer, run the search on each, and tell me what format
   to borrow from their outliers. Structure only, never the words, footage or thumbnails.

--- ROUTE C: JUST ASK (ANY CLAUDE CHAT WITH WEB SEARCH) ---

Find YouTube videos about "home workout" that got far more views than their channel usually gets. For each,
give the video's views, the channel's typical views, and why it probably did well. Then suggest 2 other niches
with a similar viewer and the format I could borrow from them.

This is the roughest route: Claude can't count every channel's videos this way, so treat its numbers as guesses
and check the best ones with Route A or B.
Free means freeFree items need no email and no card.
Not working?Email us within 14 days and we fix it or refund you.
Your card stays privateYou pay on Stripe's checkout. We never see your card.
Registered UK companyNEXARLABS LTD, company 16703034, registered in England and Wales. Check it at Companies House