About
A research desk for people who make videos
Scout answers the questions a UGC creator or a brand asks before a shoot: who makes content in my niche, how they make it, what works for them, and how they script it. It reads a niche's channels, measures their videos against their own baseline, and lays every transcript out as a script.
Scout covers 12 creators on YouTube, TikTok and Instagram in three niches: dev and tech (around Theo and ThePrimeagen), AI tools and economy. A creator is a person or a brand with one account or more, matched across platforms with the evidence shown.
Where the data comes from
Everything is public data, fetched through the ScrapeCreators API when the site is built. On YouTube: each channel's profile, its latest 60 long-form videos, its latest shorts, and the captions of its 12 latest long-form videos (of its latest 10 shorts, for a channel that only posts shorts). On TikTok: profiles, the latest 30 videos with plays, likes, comments and shares, and the captions of the latest 10. On Instagram: profiles, the latest 36 reels, and transcripts of the latest 10. Search results for a few niche queries feed Discover.
TikTok captions come with timings. Instagram's transcripts, and TikToks without captions, come from ScrapeCreators' speech-to-text, which has none: Scout spreads their words evenly over the video and marks those times with “≈”. Videos without any transcript can be transcribed on a Mac with Whisper (tools/transcribe.py), which takes YouTube ids and TikTok or Instagram links.
The numbers are a snapshot as of October 16, 2026. A TikTok or Instagram play isn't a YouTube view, so each account is measured against its own median and audiences are summed with that caveat. YouTube thumbnails load from YouTube's servers; TikTok and Instagram sign their image links for a few days only, so Scout keeps small copies. Scout is not affiliated with these platforms or with the creators.
Whether a creator is on camera or faceless isn't in the data, so it is labelled by hand for now. Detecting faces in thumbnails and frames is on the roadmap.
How accounts are matched
For each creator, Scout looks up the same handle on the other platforms and the handles their profile links to, then scores each candidate: the same email in both bios (+50), a cross-link (+35, or +55 both ways), the same handle (+25), the same website (+20), the display name, the share of bio words in common, and weak hints from the profile picture and audience size.
An account is linked at 70 points with an email, a cross-link or three independent signals. Likely and possible matches are shown as suggestions, never merged. A same-handle account with nothing else in common is rejected: anyone can register a handle. Every verdict and its evidence is public.
How each number is computed
- Median views
- The median views of the channel's recent long-form videos (up to 60). The median ignores the odd viral hit, unlike the average.
- Outlier score
- A video's views divided by its channel's median for the same kind (long-form or shorts). ×2 or more is an outlier. Shorts are measured against the median of shorts.
- Outlier rate
- The share of recent videos at ×2 their account's median or more (long-form and short counted apart).
- Cadence
- Uploads per week over the last 60 days (long-form and shorts), measured over the part of that window the fetched videos cover.
- Length buckets
- Shorts come from the Shorts tab; long-form videos fall under 10 minutes, 10–20, 20–40 or over 40. Average and median length cover the long-form videos.
- Content mix
- The share of shorts among the uploads of the last 60 days (or of the last year, for a channel that hasn't posted in 60 days).
- Views per subscriber
- Median long-form views divided by subscribers: how much of the audience a typical video reaches.
- Sponsors
- Long-form videos whose description has a sponsor line: “sponsored by”, “use code”, “% off”, or a promotional link to someone else's site near the top. Links that appear in 40% of a creator's descriptions count as their own (merch, courses, socials), not sponsors.
- Best day and length
- The weekday and the length bucket with the highest median views, among groups of at least 3 long-form videos. Days are in UTC.
- Hook
- The transcript's first 15 to 25 seconds, ending on a sentence when the captions are punctuated.
- Structure
- The transcript cut into segments of about 75 seconds (90 at most), each with its time range and opening line.
- Words a minute
- Words in the captions divided by the time from the first caption to the last.
- Reading time
- Words divided by 238, an average silent reading speed.
- Phrases they repeat
- Two- and three-word phrases found in at least three of the creator's transcripts, with stopwords and fillers removed.
- Title patterns
- For each title feature (a question, a number, the first person, “vs”…), the share of the niche's long-form videos with it that reached ×2 their channel's median, against the share for all videos, and their median outlier score.
- Similar creators
- Overlap of channel keywords and frequent title words, with a nudge for the same niche and the same format.
- Audience
- Followers summed over a creator's linked accounts (subscribers on YouTube). Someone following on two platforms counts twice.
- Engagement
- The median, over the recent short videos, of likes, comments and shares divided by views. It compares accounts on the same platform; across platforms, the counts differ.
- Same video elsewhere
- Posts on two platforms are the same video when their lengths are within 3 seconds, they went up within 4 days, and their captions or transcripts share a quarter of their words. A missing date or length on YouTube is then borrowed from the twin.
- Short-form hook
- For short videos, the first 3 to 7 seconds, and segments of about 12 seconds instead of 75.
What's next
Instagram and TikTokShipped
Profiles, reels and TikToks with transcripts, the same metrics and the same script view, and cross-platform matching.
Confirm suggestionsNext
Accept or reject a likely match in one click, and probe more candidates from a creator's page.
ThreadsLater
Posts as scripts: openings, structure and the phrases that travel.
Automatic face detectionLater
On camera, faceless or both, read from thumbnails and frames instead of labelled by hand.
Saved boardsLater
Hooks, titles and videos saved to shared boards instead of one browser.
AlertsLater
A note when a tracked creator posts an outlier, or a new channel breaks out in your niche.
