Everyone’s a 4.8

When I booked an Airbnb for my dad’s 70th, every place I looked at was rated 4.8 or higher. The one we picked was lovely, but the stars hadn’t told me that. We rate food and hotels honestly and go easy on people, and now that AI writes the resumes and reads them too, hiring leans on the ratings we’re worst at giving.

By Brian Grover @briantgrover Methods and sources

Booking an Airbnb? Guess the typical rating.

Think of a place with at least 10 reviews, rated out of 5, and drag the star to your guess.

The typical Airbnb listing in San Francisco and New York with at least 10 reviews is rated 4.84 out of 5. Half of these places are rated 4.84 or higher, and fewer than 1 in 100 falls below 4.

Places with 10+ reviews, San Francisco and New York, June 2026 Inside Airbnb Our code Food: 2025 store reviews

We go easy on people

Grades, sellers, hosts: when a person is on the other end, the scores crowd the top. When it’s a thing, like a hotel room or a jar of salsa, people say what they think, and 1 in 5 Yelp ratings in 2023 was one star.

Each grid is 100 stars. Gold is the share at the top: A’s, positive feedback, or 4.5 stars and up. All sources

Harvard grades that are A’s: 24 percent in 2005, 60.2 percent in 2025.
eBay feedback that is positive, for the average seller: 99.3 percent.
LinkedIn endorsements and GitHub stars can only say yes: 100 percent.
Airbnb listings rated 4.5 or higher in 2015: 94 percent. TripAdvisor hotels: 26 percent.

Likes miss half the best people. Real work and trusted references find them.

The same 80 pretend coworkers, rated two ways, and we know who is truly best. On top, they rate each other with one-click likes. Below, every rating has to point at real work, and counts more from people whose past picks turned out right.

Which rules find the best people?

Shortlist the top 10 of these 80 people. How many of the 10 truly best make it? Each bar is the average across 12 simulated companies. Tap one to run it.

This chart needs JavaScript. In our simulation, of the 10 most skilled people, the top 10 by score finds about 8 with real work and trusted references, about 6 when praise is tied to real work, about 5 with written reviews, hidden reviews or saying how you know someone, about 5 with one-click likes, and about 2 with LinkedIn-style endorsements. Guessing finds about 1. These are simulated, not measured.

How the lab works

It’s a simulation written for this page: 80 coworkers, each with a hidden skill level, rate each other for 30 rounds. Half the ratings go to people they worked with and half to whoever a feed shows, and the feed favors whoever has the most. Each one-tap rating is judged against the current average, plus a little, so with nothing fixed to judge against, the scores climb.

A rating that points at real work can be checked against the work, so it says more about skill than about who is well known. Hiring research finds the same thing: a work sample, where a candidate does a piece of the real job, predicts job performance better than years of experience do (Sackett and colleagues, 2022). In the lab, how firmly real work anchors a rating is our assumption, and “Try other rules” lets you change it.

Every rule and number: methods the code

How this works

80 pretend coworkers. Each one is secretly good or not so good at their job.
Every round, they rate each other. Some ratings are honest, some are just being nice.
We check: do the highest-rated people turn out to be the best ones? Black dots are the best. Gold stars are the highest rated.

This lab needs JavaScript. In the simulation, with one-click ratings that everyone can see, the scores drift up toward 5, and the top 10 by score holds only about half of the 10 most skilled people. When every rating points at real work, the scores stay spread out, and the top 10 holds most of them.

What this means for hiring

Shortlist those 80 people by likes and about half the best make the cut. With real work and trusted references, about 8 in 10 do. Real data points the same way: on one freelance site, 28% of employers who privately said they’d never rehire a worker still gave 4 stars or more in public.

AI now writes the resume and reads it. So ask for a piece of real work, ask references about specific work they saw, and trust people whose past recommendations held up.

I made this after that Airbnb search. How are you judging people when everyone looks like a 4.8?

By Brian Grover @briantgrover Methods and sources

You read to the end. This page is now a 5.0, like everything else.