The inflation files · a Worst Eats investigation
There’s no such thing as a bad restaurant anymore
Google star ratings of 1,850 restaurants across 18 US city centers · collected June 2026 · aggregates only
We run a search engine for terrible restaurants. Lately it keeps coming back nearly empty. Polite suburbs, entire downtowns, nothing under four stars for miles. That was either a bug or a finding. We checked. It’s a finding.
In June 2026 we pulled the Google rating of every restaurant our sweeps could surface within 15 km of 18 US city centers: 1,850 places, deduplicated, allegedly open. Exactly 28 of them (1.5%) rate under three stars. The median restaurant is a 4.4★. And 67.6% of everything lives in the 0.6 stars between 4.0★ and 4.6★. The scale runs from one to five. The data runs from 4.0 to 4.6.
Everything from 1.0★ to 2.9★, the half of the scale that means bad, is statistically a rumor.
Where did the bad restaurants go?
Nobody reviews an acceptable lunch. People write reviews when they’re delighted or when they’re betrayed, and delight is far better attended. Plot any restaurant’s reviews and you get the same J-shape: a mountain of fives, a defiant ledge of ones, and a desert where the twos and threes should be. Average a J and nearly everyone lands on the same high plateau. The pileup between 4.0 and 4.6 up there is 1,850 different restaurants pushed through the same scoring ritual.
Bad restaurants don’t accumulate ratings. They close. A place that earns a genuine 2★ doesn’t hang around collecting more of them. It dies, Google delists it, and the distribution quietly sheds its left tail. The 4.4s remain. Every ratings chart is a census of survivors, taken moments before the next funeral.
Run both forces for a decade and the working scale compresses into the top fifth of the nominal one. That is how 4.0★, a number that reads like a solid B+, ends up meaning bottom sixth of the market.
Methodology, honestly
The sweep. For each city we searched a hexagonal grid (seven overlapping cells covering a 15 km radius around the center, plus one popularity-ranked pass over the whole radius) against the Google Places API (New), June 2026. Eight searches per city, results deduplicated by Google place ID, unrated and non-operational listings dropped. 1,850 unique rated restaurants survived.
The cap. Google returns at most 20 results per search, with no way to page deeper, so a city can yield at most ~160 listings, and which 20 you get is Google’s call. Dense cores blow straight past the cap: Chicago netted 75 unique places, San Francisco 56. Manhattan-grade density is undersampled, and the public API offers no cure.
The frame. This is a sample of what Google surfaces, not a census of what exists. Surfacing skews prominent, and prominent skews well-rated. Somewhere past the top-20 cutoffs sit terrible restaurants we never saw, too obscure for any list.
A hostile reading says we undercounted the bad restaurants of America. We agree. You still can’t eat at them, because your search won’t show them to you either: a normal person sees one search’s top 20, which is more prominence-filtered than our eight. And the under-3★ places we did find are not a stable population; they are restaurants in the process of leaving the dataset (see: funerals, above). Every caveat in this methodology points the same direction. The shelf you can actually shop from is even more inflated than this chart.
| City center | rated | median | under 3★ | ≤ 4.0★ |
|---|---|---|---|---|
| New York, NY | 141 | 4.3 | 11 | 41 |
| Los Angeles, CA | 141 | 4.4 | 2 | 23 |
| Chicago, IL | 75 | 4.4 | 0 | 12 |
| Houston, TX | 137 | 4.4 | 2 | 28 |
| Phoenix, AZ | 106 | 4.5 | 0 | 5 |
| Philadelphia, PA | 64 | 4.3 | 1 | 12 |
| San Antonio, TX | 65 | 4.2 | 0 | 10 |
| San Diego, CA | 79 | 4.5 | 0 | 3 |
| Dallas, TX | 80 | 4.3 | 0 | 15 |
| Austin, TX | 136 | 4.5 | 2 | 21 |
| Seattle, WA | 151 | 4.5 | 2 | 12 |
| San Francisco, CA | 56 | 4.4 | 0 | 0 |
| Boston, MA | 68 | 4.3 | 0 | 9 |
| Miami, FL | 129 | 4.5 | 0 | 15 |
| Atlanta, GA | 66 | 4.5 | 0 | 11 |
| Denver, CO | 147 | 4.3 | 5 | 49 |
| Minneapolis, MN | 65 | 4.3 | 0 | 10 |
| Pittsburgh, PA | 144 | 4.4 | 3 | 32 |
| All 18 cities | 1,850 | 4.4 | 28 | 308 |
New York holds 11 of the nation’s 28 sub-3★ restaurants. Denver (49 of its 147 at or under 4.0★) is America’s most honest dining scene. San Francisco’s worst restaurant, per Google, is a 4.1. Allegedly.
What we publish. Totals only. And per Google’s terms we can’t keep Places data past 30 days, so the raw pull gets deleted and these numbers refresh monthly.
How to read a star rating now
- 4.6★the 75th-percentile restaurant. Genuinely above average. Allowed to feel proud.
- 4.5★barely above the middle: 55.6% of restaurants rate below it.
- 4.4★the median. The most statistically average restaurant in America.
- 4.3★sounds great, sits in the bottom third (31.4% rate below it).
- 4.0★the bottom 16.6%. A warning, not a compliment.
- < 3.0★the bottom 1.5%. Functionally extinct. We keep a list.
The deflator
Street math. Drag to a rating, get what it actually trades for. Look up your favorite place. Our condolences.
Street value is our arithmetic, not Google’s: the rating’s percentile in the June 2026 pool (ties split), stretched back over 1–5. Beating half the field is what a 3 was supposed to mean.
The bottom 1.5% is still out there
28 confirmed specimens across 18 cities, and near you, a few holdouts the inflation hasn’t reached yet. We built the tool that finds them. Five-star hunting is for cowards.
See what survives near you →Free. No app. The reviews are all real, which is the problem.
If you like research like this, feel free to drop us a tip. It funds the next harvest.
← Back to the huntWorst restaurants in NYCWorst restaurants in DenverAboutTaste-crime quizGet certifiedContactPrivacyTerms