Questions
FAQ
Everything you want to know about how The Humor Index works.
Is this scored by AI?
Yes. Every episode is analyzed by Claude (Anthropic's AI) using a structured two-phase process: first detecting every joke in the transcript, then scoring each one across multiple dimensions. The AI reads the full transcript and identifies comedic moments — dialogue jokes, physical comedy cues, reaction beats, cringe sequences, and more.
Can AI actually understand comedy?
It's a fair question. AI doesn't "laugh" — but it can reliably identify joke structures, setups, punchlines, callbacks, and misdirection. Our validation tests against manually-scored episodes show strong alignment on joke detection and scoring. Where AI struggles most is with purely visual gags or delivery-dependent humor, which is why we work from transcripts and apply format-based adjustments.
Which show currently ranks #1?
30 Rock sits at the top of the Humor Index at 84.4, with Arrested Development (82.0) close behind. Then The Office (79.2), Community (77.9), Parks and Recreation (77.7), Taxi (77.3), Schitt's Creek (77.3), and Seinfeld (77.0) form a tight chasing pack whose 95% confidence intervals overlap heavily — small per-episode differences move the order there. Friends (73.3) sits clearly below the pack. We've fully scored nine shows: 30 Rock, Arrested Development, The Office, Community, Parks and Recreation, Taxi, Schitt's Creek, Seinfeld, and Friends.
Do you penalize multi-camera shows or laugh tracks?
No longer. Earlier versions of the Humor Index applied a format coefficient (a 15–25% reduction on Impact for multi-cam and laugh-track shows). We removed it on 2026-04-16 after concluding it was a thumb on the scale rather than a data-driven adjustment. All shows now use a coefficient of 1.0. Read the full reasoning on our Methodology page and in the "Seinfeld vs. The Office" blog post.
What does "Craft" measure?
Craft is the writing quality of each joke, scored across five sub-dimensions: Originality (how fresh the joke is), Structure (setup-punchline clarity, misdirection), Character Integration (does it fit who's saying it?), Economy (no wasted words), and Earned vs. Cheap (was it set up through story context, or is it a lazy gag?). Craft is weighted 40% of the final Humor Index.
What does "JPM" mean?
Jokes Per Minute — the total number of detected jokes divided by the episode's runtime. It measures comedic density. A high JPM means the show is consistently firing; a low JPM might mean longer dramatic stretches between laughs. JPM alone doesn't determine quality — a show can have low JPM but extremely high Craft (prestige comedy).
What does "Impact" measure?
Impact estimates the audience reaction to each joke — how hard it lands. It factors in quotability (will people repeat this line?), rewatch bonus (is it funnier the second time?), and cultural footprint. Impact is no longer adjusted by show format (see the laugh-track question above).
Why do some shows show 0.0 scores?
Those shows are queued for analysis but haven't been scored yet. We're working through the catalog — each episode takes a few minutes to analyze. Shows will populate with real scores as analysis completes. We don't fake or estimate scores.
How many jokes do you find per episode?
Typically 30-70 per 22-minute episode, depending on the show's style. Dense comedies like Arrested Development and 30 Rock tend toward the higher end. Shows with more dramatic content or longer scenes between jokes come in lower. Every joke is individually scored.
What is the 0-100 score scale?
Raw scores are on a 0-10 scale internally. We convert to a 0-100 display scale where 75 represents an "average comedy episode." This is a fixed calibration — scores won't shift as we add more shows. A score of 90+ is exceptional, 80-89 is great, 70-79 is solid, and below 70 means the episode had weaker-than-average humor.
Do you analyze every episode of every show?
That's the goal. We're starting with the full runs of 11 major sitcoms — every single episode, not a sample. This means 1,900+ episodes when complete. It takes time, but partial analysis would undermine the rankings.
Can I request a show?
Yes! Head to the Request page and vote for the show you want analyzed next. Top-voted shows get prioritized.
Is the data open source?
The scoring methodology is fully transparent on our Methodology page. The raw joke-level data isn't currently public, but we're considering it for research purposes.
Why transcripts instead of watching the episodes?
AI can't watch video (yet). Transcripts capture dialogue, stage directions, and timing cues — which is where the vast majority of sitcom humor lives. We acknowledge this misses some purely visual gags, which is one reason we adjust scores by format. Single-camera shows like The Office embed more visual comedy cues in their scripts than multi-camera shows.