We integrated IMDb episode ratings across every analyzed show on The Humor Index. And the first thing we did was the obvious data science move: how well do audience ratings predict our comedy scores?
Updated September 2026: this post originally ran on 591 episodes across three shows. We now have IMDb ratings matched to 2,654 scored episodes across all 21 shows, so the headline numbers below have been recomputed on the full corpus. The finding got stronger, not weaker.
The answer: they don't. Not even close.
The Numbers
Across 2,654 episodes of all 21 scored shows, centred within each show so that no cross-show scale differences leak in, the Pearson correlation between the Humor Index and IMDb ratings is r = +0.20. That is positive, and with n this large it is not a fluke, but it means audience ratings account for under 4% of the variance in how well an episode's jokes are written. The other 96% is something else.
Per show, here's where it lands:
- Chappelle's Show (27 eps): r = +0.58 — the strongest on the board, on one of the smaller samples
- Arrested Development (84 eps): r = +0.46
- Fleabag (12 eps): r = +0.40
- The Larry Sanders Show (90 eps): r = +0.37
- Broad City (50 eps): r = +0.36
- The Office (183 eps): r = +0.34
- The Simpsons (552 eps): r = +0.31
- Veep (65 eps): r = +0.29
- Seinfeld (168 eps): r = +0.27
- Futurama (137 eps): r = +0.21 · Flight of the Conchords (22 eps): r = +0.20 · Taxi (114 eps): r = +0.19 · The Fresh Prince (148 eps): r = +0.18 · Parks and Recreation (123 eps): r = +0.16 · Community (110 eps): r = +0.12
- 30 Rock (138 eps): r = +0.06 · Curb Your Enthusiasm (120 eps): r = +0.05
- Friends (236 eps): r = −0.02 · Schitt's Creek (80 eps): r = −0.02 · It's Always Sunny (177 eps): r = −0.03 · Freaks and Geeks (18 eps): r = −0.18
Seventeen of twenty-one shows are positive; the median show sits at r = +0.20, and not one show clears r = +0.6. There is a real signal here and it is weak everywhere.
Across all 21, IMDb explains under 4% of the variance in our Humor Index scores, and in the four shows with negative correlations, none at all.
In plain English: knowing an episode's IMDb rating tells you almost nothing about how funny it actually is by our analysis.
Why This Matters
IMDb ratings measure whether audiences enjoyed an episode. That's a cocktail of plot quality, emotional resonance, character development, guest stars, and yes — comedy. When someone gives "Casino Night" a 9.3, they're rating the Jim/Pam moment at the end as much as any joke.
The Humor Index measures something narrower: comedy craft and density. How many jokes land? How well-constructed are they? How hard do they hit?
These are genuinely different questions, and our data proves it.
The Biggest Disagreements
Some episodes where our AI sees comedy gold but audiences shrug:
- "Andy's Ancestry" (S9E03): Humor Index 94.8, IMDb 7.1. Dense with character comedy, and one of the widest gaps in the show. Season 9 fatigue dragged the audience score down regardless of what landed.
- "Angry Andy" (S8E21): Humor Index 92.2, IMDb 6.6. Packed with jokes, but the late-season Andy arc turned audiences off no matter how many of them worked.
- "Here Comes Treble" (S9E05): Humor Index 90.9, IMDb 6.7. Same pattern, same season, same verdict from the crowd.
And episodes audiences adore that don't score as high on pure comedy:
- "Goodbye, Michael" (S7E21): Humor Index 87.6, IMDb 9.8. Tied with the finale for the highest audience rating in the series, and merely a good one by joke count. It is an exit, not a comedy showcase.
- "Livin' the Dream" (S9E21): Humor Index 82.1, IMDb 9.0. Dwight finally gets the manager job. Payoff, not punchlines.
- "The Merger" (S3E08): Humor Index 81.2, IMDb 8.6. The Stamford branch arrives, the plot moves, and the joke rate sits at the bottom of Season 3.
What Predicts IMDb Ratings?
We tested which of our sub-metrics best correlates with audience scores:
Within The Office specifically:
- Impact (r = 0.39) — how hard jokes land is the closest thing to an audience proxy we have
- Craft (r = 0.35) — how well they are built
- Humor Index (r = 0.34) — the composite score
- JPM (r = -0.19) — joke density is the least related thing to an audience rating
The last one is the interesting one. More jokes per minute predicts slightly lower audience ratings. Episodes with the highest joke density often give up plot and character time to get there, and audiences notice.
Season-by-Season Patterns
The correlation varies wildly by season:
- Season 2 has the strongest correlation (r = 0.51), the stretch where the show's best episodes are also its funniest by joke count
- Season 8 is the only negative season (r = -0.24). It is also the Michael-less season, where our scorer keeps finding jokes the audience had stopped showing up for
- Seasons 5 and 1 land in between (r = 0.42 and r = 0.39); Season 4 is nearly flat at r = 0.14
The Dinner Party Correction
Earlier versions of this post led with "Dinner Party" as the cringe paradox: our highest-scoring Office episode, an IMDb 7.6, brilliant discomfort that audiences would not sit through twice. Both halves of that were wrong.
"Dinner Party" is an IMDb 9.4, one of the highest-rated episodes the show ever aired. The 7.6 came from our own join: IMDb splits the hour-long episodes into two entries and we did not, so our ratings slid out of alignment in the seasons that have them. That is fixed as of September 2026 by matching on episode title instead of episode number, across every show it affected: 36 of 183 Office ratings came back different, and so did 28 Seinfeld, 60 Futurama, 12 Friends, 9 Larry Sanders, 7 30 Rock and 2 Parks. On the comedy side the September rescore put the episode at 91.2, 36th of 183, not first.
So the cringe paradox does not hold where we said it did. Audiences and our scorer agree about "Dinner Party": both rate it highly. Where they genuinely part company is Seasons 8 and 9, where our scorer keeps counting well-built jokes in episodes the audience had already written off.
That is still the distinction the Humor Index exists to make. Not "did you enjoy this" but "is this comedy operating at a high level". It just turns out the clearest examples of the two coming apart are late-series Andy episodes, not the show's most famous bottle episode.
The Bottom Line
The Humor Index and IMDb are complementary, not competing metrics. IMDb tells you what audiences love. The Humor Index tells you what's actually funny.
Sometimes those overlap. Often, they don't. And the disagreements are where the most interesting conversations happen.
Explore the data yourself — every episode now displays its IMDb rating alongside the Humor Index score. See where you agree with the crowd and where your taste diverges. [Start with The Office](/shows/the-office).