PlaceBench: we tested six AI assistants on 150 questions about real places
Ask an AI assistant "is anywhere near Slussen still serving hot food at 2 am on Monday?" and you get a confident answer. Is it right? We built PlaceBench to measure that: 150 everyday questions about real places in 121 cities, each with an answer key checked against the place's own sources. We asked six assistants every question. Scores run from 49 to 92 out of 100. Results (November 2026)…
PlaceBench tested six AI assistants on 150 questions about real places in 121 cities, assigning each a score out of 100. The assistant Dots scored the highest at 92, with no wrong answers, while the least accurate assistant was Ask Maps, scoring 49. Four other assistants, Muse, Claude Duet, Instinct and Grok Bot, were closely matched in their results.
Each assistant was asked the same questions, and scores ranged from 49 to 92 percent. The study found that closed places were still recommended in 12 answers, which were checked against public sources showing the places were closed, moved or renamed. The questions were written in the city's own language and were specific to a day and time in November 2026, with 46 questions in the city's native language.
The sources for each answer were cited, with the place's own website and notices given priority, followed by the operator's timetable, Google Maps, OpenStreetMap, official registers and dated press. If sources disagreed, the place's own page was accepted as the truth. The grades were double-checked by two independent grading models, and the study found that changes in the assistants over time could affect the results.
The data, keys, sources and grades are available on GitHub under CC BY 4.0, and the full write-up with charts is available at voygr.tech/placebench.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.