Somewhere between a glance at the dashboard and a half-remembered email, a driver now asks the car where to go in the language of their own life rather than the language of street addresses. Google is extending its Personal Intelligence feature into Android Auto, letting Gemini draw on calendars, Search history, and Photos so that a voice request like "take me to the pumpkin patch we loved last fall" becomes a route. The reviewer at ZDNET found it accurate across every test, though the machine never paused to ask whether it had understood correctly. It is a quiet step in a long story about how
ZDNET Tests Gemini's Personal Intelligence in Android Auto, Finding Destinations From Email and Photos
It chose right every time, but it never confirmed the destination.
Mark
So what's actually new here? Gemini already had Personal Intelligence.
Mimi
The setting is the car. The reviewer says it does the same things as before, but now it acts on them: it finds a parking pass or an appointment and starts navigation.
Mark
What did he actually test?
Mimi
A parking pass from email, a daughter's dentist appointment, a coffee shop list carried across conversations, and a pumpkin patch found through Photos. All worked.
Luke
Worth noting that's one reviewer and a handful of tasks. "Perfectly every time" is his experience, not a measured accuracy rate.
Mark
And the complaint?
Mimi
Gemini never confirmed the destination. It just went.
Luke
Right, and that's the real gap. It was right each time here, but we don't know how often it would be wrong, or what it does with ambiguous cases like two similar appointments.
Mark
What about privacy?
Mimi
It's opt-in, he can stop it anytime, and it only covers Google Workspace apps. It can't read text messages.
Luke
He did call the standard version a little unsettling. The article describes controls, but doesn't say what data is retained or how.
Mark
So the takeaway?
Mimi
It's convenient and, in this test, accurate, but a confirmation step would make it safer.
The Pulse
- A single spoken request can now search a driver's inbox, photos, and calendar and turn the results into a route within seconds.
- The reviewer's test drives produced correct destinations every time, yet the assistant never checked with him before starting navigation.
- Access is opt-in and confined to Google Workspace apps, so text messages remain outside the assistant's reach.
- The open question is whether a confirmation step arrives before the first wrong guess sends a driver miles off course.
Somewhere between a glance at the dashboard and a half-remembered email, a driver now asks the car where to go in the language of their own life rather than the language of street addresses. Google is extending its Personal Intelligence feature into Android Auto, letting Gemini draw on calendars, Search history, and Photos so that a voice request like "take me to the pumpkin patch we loved last fall" becomes a route. The reviewer at ZDNET found it accurate across every test, though the machine never paused to ask whether it had understood correctly. It is a quiet step in a long story about how much of our personal record we are willing to hand over in exchange for a little less effort on the road.
The first test took under ten seconds. The ZDNET reviewer, sitting in his car, asked Gemini to find the parking pass for his game that weekend and take him there. The assistant pulled the address from a PDF buried in his email and started navigation. He had not typed a word.
That is the promise of Personal Intelligence in Android Auto. Google first brought the feature to Gemini earlier this year, letting the assistant read a user's Photos, Gmail, and other sources so its answers fit the person rather than the crowd. Now it reaches the dashboard, where the most natural use is navigation: name a destination in terms of your own life and let the assistant find the address.
The reviewer tried more. Asked to take him to his daughter's dentist appointment, Gemini found the confirmation email and drove him there. Asked to head to the closest coffee shop from a list it had built earlier at home, it did. Asked for the pumpkin patch his family visited last fall, it thought for a few seconds, then chose the apple orchard they go to every year. Every request landed correctly.
His one complaint was that Gemini never confirmed the destination before moving. Had it guessed wrong and he followed without checking, the trip could have gone badly.
The limits are clear. The feature works only with Google Workspace apps and cannot read text messages. It is opt-in and can be switched off at any time through Settings at gemini.google.com. What remains to be seen is whether Google adds a confirmation step before navigation starts.
The first test took under ten seconds. The reviewer at ZDNET told Gemini in his car to "find the parking pass for my game this weekend and take me there." The assistant pulled a PDF out of his email, read the address of the specific parking deck he needed, and started Maps navigation. He had tickets to his alma mater's football game, and he had not typed a thing.
That is the pitch for Personal Intelligence in Android Auto. Google introduced the feature for Gemini earlier this year, giving the assistant access to a user's Photos, Gmail and other sources so its answers could be tailored rather than generic. On Tuesday, Google announced it was coming to Android Auto, where Gemini can draw on calendar events, appointments, Search history and Photos. In a car, the obvious use is navigation: say where you need to go in terms of your own life, and let the assistant work out the address.
The reviewer had tried the standard version several months earlier and found it a little unsettling how much it knew. It helped him shop for car tires, identified his vehicle from pictures in Photos and insurance emails, and found clothing in his size from receipts in Gmail. The Android Auto version does the same things, he wrote, but takes them a step further by acting on them behind the wheel.
He ran several more tests. Asked to "take me to my daughter's dentist appointment," Gemini found a confirmation email and began navigation. He then tried continuity between conversations: before leaving the house, he asked for the best coffee shops near the appointment. Afterward, in the car, he asked Gemini to navigate to the closest one from that list, and it did.
The hardest test involved Photos. He said, "Navigate to that pumpkin patch I went to with my family last fall," adding that there should be pictures. After a few seconds of thinking, Gemini launched navigation to the apple orchard and pumpkin patch his family visits every year. Across every find-and-navigate request he tried, the assistant got it right.
His one complaint was that Gemini never confirmed the destination. It chose correctly every time, but he noted that if it had picked the wrong place and he had followed it without checking, it could have ruined his trip.
There are limits. Personal Intelligence works only with Google Workspace apps. It cannot read text messages, even if Google's Messages is your texting app. It can write a text or reply to one you've received, but it cannot pull information from those messages to complete other tasks.
The feature is opt-in, and you can stop it at any time. To turn it on, go to gemini.google.com, open Settings, choose Personal Intelligence, and select which data the assistant may access. The thing to watch is whether Google adds a confirmation step before navigation begins, since the reviewer's one criticism concerned a wrong guess that, in his testing, never happened.
Notable Quotes
Find the parking pass for my game this weekend and take me there.— ZDNET reviewer's prompt to Gemini
If it had picked the wrong place and I blindly followed its suggestion, it would have ruined my trip.— ZDNET reviewer (paraphrased)
Want the full story?
Read the original at ZDNET ↗
Technology