
Here's the thing that's been driving me crazy: my travel details are scattered across Gmail, Photos, Calendar, and YouTube. Ask one app, “When did I last go to Tokyo?” and you can still end up checking three others to confirm the answer.
Hanks here. I test AI tools by breaking them, then seeing which ones actually stick in real workflows. When Google introduced Personal Intelligence for Gemini on January 14, 2026, I had one question: does it work with messy real-world data, or only in demos? If the feature is new to you, I’ve also covered what Gemini Personal Intelligence is.
I spent a week testing it with my actual chaos: 47 unread emails, poorly organized photos, and scattered search history. Here’s what happened with 15 prompts that could make cross-app searches useful.
Some surprised me. Others broke exactly where I expected.

Check access first. Google currently lists Personal Intelligence as a beta for eligible users aged 18 or older with personal Google accounts. It works on the web, Android, and iOS, but availability varies by region. Google lists the European Economic Area, Switzerland, the United Kingdom, and Nigeria as excluded. If you don’t see the option, the rollout or an account setting may be the reason.
Each group below needs different connected data:
Google’s setup guide also lists requirements for Keep Activity, Search and YouTube history, and Google Photos. Connecting an app does not guarantee a complete answer. When a result matters, follow up with: “Which email, photo, or event did you use, and what couldn’t you verify?” Gemini may show its sources; if it doesn’t, ask.
With Google Workspace connected, Gemini can find and summarize information from Gmail. Google describes Personal Intelligence as combining details from connected sources, though you still need to check whether it found the right thread.
The Prompt:
Summarize all emails from [person@email.com] in the last month, highlighting key decisions and attachments.
What Actually Happens: Gemini scans your Gmail threads and pulls together bullet points of major discussions, decisions, and attached files. In my test with project emails, it caught the budget approval ($5,000 confirmed) and linked to the contract PDF attachment.
Where It Works Best: Monthly check-ins with clients or team leads where you need a quick recap before meetings.

Where It Breaks: If the person uses multiple email addresses or forwards from different accounts, Gemini might miss threads. Also, it struggles with subtle tone shifts: sarcasm or frustration get flattened into neutral summaries.
The Prompt:
Find my latest flight confirmation from Gmail, including flight number, departure time, and any hotel bookings attached.
What I Got:
Reality Check: This worked smoothly for my recent booking. Gemini pulled the airline confirmation email and caught the hotel reservation in a separate thread. But it missed a train booking sent to the same address. That confirmation was in a PDF attachment. I can say Gemini missed it in this test; I can’t tell from one miss whether PDF attachments generally cause the problem.
Pro Tip: If a booking detail matters, open the source email or attachment before you rely on the summary.
The Prompt:
Extract action items from my emails with [Team/Colleague] this week, prioritized by deadline.
Output Example:
Where It Gets Tricky: Gemini can identify explicit to-dos (“Please submit by Friday”), but it misses implied tasks. If someone writes “Let me know your thoughts,” Gemini doesn’t always flag that as an action item. You’ll need to review nuanced requests yourself.
The Prompt:
Draft a reply to [Person's latest email] based on our email history, keeping a professional tone and referencing our last discussion on [Topic].
Sample Output:
“Dear [Person], thank you for your update. Based on our discussion last month about the Q1 targets, I suggest we align on the revised timeline before proceeding. Looking forward to your feedback.”
Honest Take: This is where Personal Intelligence starts feeling useful. It pulled context from our previous exchange without me digging through threads. But you still need to specify the tone. I tried it without “professional tone” and got a weirdly casual draft that didn’t match the relationship. For everyday decisions and routines rather than inbox work, Macaron’s AI personal assistant takes a different approach to personal context.

Connect Google Photos for these. For #6, connect Gmail through Google Workspace as well.
The Prompt:
Find photos from my Tokyo trip last summer in Google Photos, sorted by date.
What Worked: Gemini returned a chronological list:
What Didn’t: It missed photos from a day trip to Nikko because I hadn’t tagged the location in Photos. I wouldn’t treat that as proof Gemini requires geotags for every search. It does mean an incomplete result can look convincing, so check the album before assuming a day is missing.
The Prompt:
When did I last visit [Tokyo], based on my Google Photos? Include any related emails from Gmail.
Gemini’s Response:
“You last visited Tokyo on October 5, 2025. Your photos show you at a local restaurant, and related emails include your flight confirmation and hotel booking.”
This Is Where It Got Interesting: Gemini connected photo timestamps with email confirmations without me naming the messages. That’s the kind of cross-source reasoning Google describes. I’d still open the cited photos and emails before treating “last visited” as settled: a later untagged photo or a booking you didn’t use could change the answer.

The Prompt:
Create a memory slideshow from my [Event] photos in Google Photos, with captions based on dates and locations.
Status: In my test, Gemini generated an interactive slideshow link with captions, but the interface felt clunky. I wouldn’t promise that result on every account. Google documents slideshow creation in Photos; whether this particular Gemini request works for you is something to test.
These combine different kinds of context, so check the source before trusting a recommendation. For a focused clothing use case, Macaron’s personal AI wardrobe assistant uses the clothes and plans you tell it about to suggest outfits.

The Prompt:
Plan a weekend based on my interests from YouTube history and past photos, within 3 hours of my location.
What Gemini Suggested:
Reality Check: The hiking suggestion was spot-on; I do have a bunch of trail photos. The museum recommendation felt generic. Gemini saw one art documentary and assumed I’m into museums. Google acknowledges this kind of over-personalization: photos or viewing history don’t always reveal why you were there or what you liked.
Fix: After I corrected it (“I prefer hands-on experiences, not museums”), future suggestions improved.
The Prompt:
Suggest restaurants I'd like near me, based on my Gmail receipts and YouTube food videos.
Output:
Where It Works: If you have a clear dining pattern, receipts and watch history can give Gemini something specific to work with. In my test, the suggestions reflected sushi receipts and ramen videos.
Where It Doesn’t: If your eating habits are all over the place, like mine, recommendations get scattershot. A receipt may also reflect where you took someone else, rather than where you want to eat again.
The Prompt:
Find free time next week for [Activity], checking my Google Calendar and avoiding travel days from emails.
Gemini’s Answer:
“You have free time on Thursday, Jan 29th, 2:00-4:00 PM. No travel conflicts detected in your emails.”
Gemini cross-referenced Calendar and Gmail for travel. Keep the source check in the workflow, especially before you book anything. If Gemini overlooks Calendar, Google says you can add @Google Calendar to the prompt; Google Workspace must be connected.

These prompts use Gmail receipts and Search history for budget-aware recommendations. I wanted to see whether Gemini could make sense of my purchase patterns. The prices below are what it returned in my tests, so check current listings before buying.
The Prompt:
Research tires for my car based on model from my photos and emails, including pricing.
Output:
How It Worked: Gemini found a photo of my car’s license plate and used car-related email context to identify the model before suggesting tires. Google published a similar tire-shop example, though the table above is the output from my test.
Where I Got Stuck: Pricing was accurate for Japan but might be off for other regions. Verify the tire size against your vehicle documentation and check a current local seller before buying.
The Prompt:
Compare phone options from my past Search history, focusing on battery life.
Sample Table:
Limitation: Gemini pulled phones I’d searched for in the past month, but it didn’t know which features mattered most until I specified “battery life.” The figures above are the sample output, not a current, like-for-like product comparison. Specify your priority, then check how each maker measured battery life and what each phone costs now.
The Prompt:
Recommend budget-aware laptops under $800, based on my email discussions.
Result: Gemini scanned my Gmail for budget mentions in work emails and suggested laptops within that range. But it recommended models I’d already dismissed in previous emails. It didn’t reliably track the difference between “I considered X” and “I decided against X.”
Pro Tip: If you’ve ruled out options, say so explicitly: “Exclude models I’ve mentioned rejecting.”

The Prompt:
Recommend YouTube channels based on my watch history and cooking receipts from Gmail.
Output:
What Surprised Me: Gemini connected my grocery receipts, which included lots of Japanese ingredients, with ramen tutorials in my YouTube history. Both influenced the recommendations. That’s more useful than a generic list, provided those receipts and videos actually represent what you enjoy.
The Prompt:
Suggest books or shows based on my Search history and past event photos.
Output:
“Based on your search history for historical novels and photos from museum events, I recommend ‘The Name of the Rose’ by Umberto Eco and the documentary series ‘Civilizations.’”
Reality Check: This felt half-right. The historical novel recommendation was solid; I do search for those. The museum photos were misleading. As I mentioned earlier, I’m not actually into museums. The suggestions improved after I corrected Gemini’s assumption.

Add “last month,” “since 2025,” or “this week” when the date matters. “Recent emails” leaves Gemini to decide what recent means. Check the dates it returns, too: prompt #10 shows why.
Instead of “What’s in my Gmail?”, ask: “Based on Gmail and Photos, when did I last travel to [City]?” Name the sources you want checked, make sure they’re connected, and ask Gemini to show what it used. For everyday planning prompts that don’t depend on searching a Google account, see Macaron’s AI personal assistant prompts.
Personal Intelligence is optional. You choose which eligible apps to connect and can disconnect them in Settings → Personal Intelligence → Connected Apps. If you’re unsure, connect only the app you need for the prompt you’re trying.
A photo shows where you were, not necessarily why you went. A video you watched once doesn’t define your interests. If Gemini draws the wrong conclusion, correct it directly: “I don’t actually like golf.”
Google says it does not train directly on your entire Gmail inbox or Google Photos imagery. That doesn’t mean nothing related to those apps can be used to improve AI. Its current help documentation says relevant summaries, excerpts, inferences, prompts, and responses may be used for model improvement, and some connected-app interactions may be reviewed by people. Disconnecting an app also does not delete information already saved in Gemini Apps Activity; that has a separate control. Read those settings before connecting anything sensitive.
After a week of testing, I’d keep the prompts that retrieve something I can check: an email thread, a booking, a photo, a Calendar event. The email summaries helped me prepare for meetings, and the photo-email connection surfaced travel details I would otherwise have hunted down myself. Shopping and entertainment took more correction.
At Macaron, the question I’m interested in is what happens after an AI finds a useful detail. Can that context help someone make a better everyday decision or create a small tool they’ll use again? That’s a different job from finding a flight email, and it’s why these tests are useful to us.
If you’re trying Personal Intelligence, start with one connected app and a real task. Ask what source Gemini used. Check the answer. Then try a second app if the first result earns your trust. The feature is still in beta, and eligibility and connected apps can change, so check Google’s current setup page before following an older walkthrough.
You can paste them into a Gem, but Personal Intelligence currently isn’t available there. If a prompt needs your connected Gmail, Photos, or personal history, use a regular Gemini chat.
Yes. Open Tools in that chat and turn Personal Intelligence off. If you switch it off halfway through, Gemini can still use earlier messages in the same conversation. Use a temporary chat if you also don’t want the conversation saved to Gemini Apps Activity.
Disconnecting those apps from Personal Intelligence doesn’t stop Gemini from finding public videos or web pages. A channel recommendation alone doesn’t prove it used your watch history. Ask which source it used before judging the result.
Previous posts: