Google Maps now answers questions in plain language through Ask Maps. It's fast. In PlaceBench, our test of six AI assistants on 150 everyday questions about real places, Ask Maps answered in a median 8 seconds. The next fastest, Grok Bot, took 29 (from 53 timed answers).
It also scored lowest: 49 out of 100.
| Assistant | Score | Wrong answers, of 150 | Median time |
|---|---|---|---|
| Dots | 92 | 0 | 57 s |
| Muse | 83 | 6 | 81 s |
| Claude Duet | 79 | 6 | 43 s |
| Instinct | 76 | 12 | 64 s |
| Grok Bot | 73 | 12 | 29 s |
| Ask Maps | 49 | 30 | 8 s |
Where Ask Maps goes wrong
A wrong answer in PlaceBench is a confident answer that a source contradicts, such as a place that's closed at the time you asked about, or a schedule that doesn't match the operator's timetable. Wrong answers cost a point. Saying "I can't confirm that" costs nothing.
Ask Maps got 92 of its 150 first answers fully right, so it's often correct. The problem is the 30 wrong ones. Splitting its grades by kind of question (our count from the public grades file):
| Kind of question | Questions | Ask Maps wrong |
|---|---|---|
| Finding a place nearby | 54 | 16 |
| Planning a route or a time | 33 | 7 |
| Place details | 41 | 3 |
| Sharing with friends | 11 | 3 |
| Booking and contact details | 11 | 1 |
"Finding a place nearby" is the kind of question you'd expect a maps app to win: "Will anywhere near Slussen still be serving hot food at 2 am on Monday?" Ask Maps got 16 of those 54 wrong. Dots got none wrong across all 150.
Most PlaceBench questions name a day and a time, like "Tuesday 17 November, 10:30 pm". An answer has to check that the place is open at that hour, not just that it exists and is nearby.
Asked three times
We asked 20 questions three times each. Ask Maps was right on all three tries for 12 of them, the same as Muse and ahead of Instinct (5) and Grok Bot (8). Dots managed 17.
How we graded
Every question has an answer key built from the place's own website, the transit operator's timetable, its Google Maps listing where it has no page of its own, and other public sources. Where sources disagree, the place's own page wins. 27 keys cite Google Maps. Without those 27 questions Ask Maps scores 54, and it's still last.
Claude Opus 5.5 graded each answer against its key. OpenAI's gpt-6.1-sol regraded 60 answers blind as a cross-check and made the same right-or-not call on 48.
This is one run, from 11 to 14 November 2026, and assistants change.
Data
Questions, keys, sources and every grade are on GitHub under CC BY 4.0: github.com/voygr-tech/placebench. Full write-up: voygr.tech/placebench.
Top comments (0)