How the counting works
The Likeometer is a joke we take extremely seriously. The numbers are real, the method is documented, and every single count links to the exact second it happened.
1 · Transcripts
Solo videos: we pull the video’s official YouTube captions, with timestamps — which means the count is, if anything, an undercount: auto-captions occasionally smooth out stutters and drop repeated fillers. Podcasts and interviews: we transcribe the audio word-by-word and run speaker diarization (who-spoke-when), so only the target person’s own words are counted — the hosts’ fillers are never pinned on the guest. Those videos carry a 🎙 badge and their likes/min divides by the person’s own speaking time, which is the honest denominator.
2 · Every “like” counts — and gets labeled
The headline number is the simplest stat on the internet: every single spoken “like” counts. No judgment calls, no AI deciding what’s in — if the word left their mouth, it’s on the board. Then each one is inspected with its surrounding context and labeled into one of four buckets, so you can always see the split between abuse and normal grammar:
“it was, like, the worst day ever” — deletable discourse glue
“and I was like, no way” — the famous was-like quote intro
“it took like five minutes” — replaceable with “about”
“I like ice cream” / “looks like rain” / “hit the like button” — real grammar, still counted, labeled proper
Filler + quotative + hedge make up the filler share shown on every video page (“87% filler”). Proper uses are part of the total but never part of the filler share.
3 · Benefit of the doubt
The total can’t be inflated — it’s a literal word count. The only judgment is the labeling, and genuinely ambiguous cases default to proper: we’d rather understate someone’s filler share than exaggerate it. Every label is public — tap “Proper” on any video page to audit the calls yourself.
4 · The stats
Likes/min = all spoken likes ÷ runtime (the person’s own speaking time on speaker-attributed videos). Total likes = every spoken “like” across everything we’ve indexed for that person. Filler share = filler + quotative + hedge ÷ total. Hottest minute = the densest 60-second window in any video. Rankings only include what we’ve indexed — it’s a leaderboard, not a census.
5 · Known limits
“Um” and “uh” are mostly stripped by YouTube captions, so those leaderboards need real audio transcription — coming later. Speaker attribution on diarized videos is ~90–95% accurate for clean 2–3 speaker audio; cross-talk seconds can land on the wrong voice. Disagree with a call? Every receipt is public — send the timestamp via the suggest page.