Google quietly rolled out two features that change how flight search works. Not “changed” in the marketing-speak sense. Changed in the actual, practical sense that the tool now does things it couldn’t do six months ago.

Flight Deals (powered by Gemini) lets you describe a trip in plain language and get personalized fare suggestions back. No fixed dates required. No destination required. Just describe what you’re after.

Canvas is a persistent side panel that turns those vague trip ideas into actual itineraries, with live flight prices, hotel comparisons, and Maps reviews stitched together.

Both are live now inside Google Flights in AI Mode. Here’s what happens when you actually use them.

---

What Flight Deals Actually Does

The traditional Google Flights interface asks: where, when, how many passengers. Flight Deals asks: what are you after?

Type something like “I want a warm beach trip sometime in March, flexible on destination, budget around $600 from Boston” and Flight Deals processes that description using Gemini, then surfaces fare options that match. Ranked by personalized fit, not just price.

This is genuinely different from the Explore map, which requires you to pick a departure airport and shows raw prices. Flight Deals incorporates your constraints, past searches if you’re signed into Google, and current fare trends to prioritize results.

What works well:

• Flexible-date searches where you don’t care about exact days
• Open-destination trips where budget is the primary constraint
• Catching deal windows: Google tracks fares over time and surfaces trips when they drop below predicted norms

What doesn’t work:

• Business travel with fixed dates. The AI angle adds zero value when you need JFK-ORD on March 14.
• Routing control. You can’t say “show me one-stop only through European hubs.” The system decides.
• Budget airlines. Southwest still doesn’t appear. Some European low-cost carriers are absent.

If your trip is flexible, Flight Deals is legitimately useful. If you already know where and when you’re going, skip it and use the regular search.

---

Price Tracking With Confidence Estimates

Google Flights already had price alerts. The 2026 version adds one thing: confidence estimates.

When you set a price alert, you now see “High confidence: fares on this route are predicted to rise before your travel dates” or “Low confidence: prices have been volatile.”

Tested this on four routes over six weeks:

• LAX–Tokyo (April): High confidence alert fired. Prices jumped $200 within 10 days. Accurate.
• BOS–Lisbon (May): Medium confidence said prices would hold. They did for two weeks, then dropped $80. Partially accurate.
• JFK–Dublin (June): Low confidence, prices moved unpredictably. Alert was honest about the uncertainty.
• ORD–Rome (July): High confidence predicted a rise. Prices dropped instead. Wrong.

Three out of four isn’t bad for probabilistic predictions. The confidence labels are honest. Low confidence really does mean the system isn’t sure. Treat “High confidence: prices rising” like a 70% signal, not a guarantee.

The proactive alert behavior is new. Rather than waiting for prices to hit a threshold you set, Google now sends alerts when it detects a meaningful change, including upward movement. Useful if you’re watching a route but haven’t booked yet.

---

Canvas: How the Side Panel Works

Canvas appears on the right side of the screen when you’re in AI Mode. Think of it as a scratchpad that persists across your session.

Start a conversation (“plan a 10-day trip to Japan in late April, two adults, interested in food and hiking, budget-flexible but not unlimited”) and Canvas builds out:

• A day-by-day itinerary skeleton
• Flight options with live prices pulled from Google Flights
• Hotel comparisons with Maps star ratings, price ranges, neighborhood notes
• Estimated daily costs

The itinerary is editable. You can move days around, swap hotels, ask it to add a day trip from Tokyo to Nikko, and the panel updates. Live prices update as you make changes, usually within 10-20 seconds.

What makes Canvas useful vs. just using ChatGPT or Claude:

The live data integration. ChatGPT’s travel planning is static. It generates a plausible-sounding itinerary but can’t tell you that flights from Tokyo to Osaka on April 23rd are $200 today, or that the hotel you’re eyeing has dropped two stars in recent reviews. Canvas pulls real data.

What Canvas doesn’t solve:

Hallucinations are reduced but not gone. During testing, Canvas suggested a restaurant in Kyoto that Google Maps shows permanently closed. It called a 52-minute walk a “quick 20 minutes.” Real-data integration helps with prices and reviews, but logistics reasoning is still imperfect.

Offline: zero. Canvas requires connectivity. Don’t expect it to work on the plane.

Privacy: you’re sharing trip plans, travel dates, and preferences with Google. If that bothers you, use it with a fresh incognito session or avoid it entirely.

---

Google Flights AI vs. Hopper vs. TripIt

No other tool does exactly what Flight Deals + Canvas does as a combined system. But travelers comparing their options usually land on these three alternatives:

vs. Hopper

Hopper has been doing price prediction and flexible-date recommendations for years. It’s better at pure price forecasting on specific routes. Its prediction model is trained on more historical flight data than Google has published about its own system.

Where Google wins: the integration with search. Flight Deals surfaces deals you discover in the moment. Hopper requires you to already know your route. It predicts prices on that specific trip.

Freeze pricing is Hopper’s real differentiator. You can lock in a price for a small fee while you decide. Google has no equivalent.

Hopper is better for: specific routes you’re monitoring over time. Google Flight Deals is better for: open-ended discovery when you don’t know where you want to go.

vs. TripIt

TripIt is an itinerary organizer, not a flight search tool. You forward confirmation emails and it assembles your trip timeline. It’s excellent at that specific task.

Canvas tries to replace TripIt for the planning phase (building itineraries before you book). TripIt handles the post-booking phase, organizing what you’ve already confirmed.

They solve adjacent problems. If you book through airlines and OTAs and want everything organized automatically, TripIt Pro ($49/year) still has no real competitor. Canvas doesn’t import existing bookings.

The gap nobody fills

No practical guide exists that walks through using Flight Deals and Canvas together, end-to-end, for a real trip. The Google help docs are thin. Most coverage is press release regurgitation.

The actual workflow, for anyone who wants to try it:

1. Open Google Flights, switch to AI Mode
2. Type a natural-language trip description in Flight Deals (vague is fine)
3. Review the fare suggestions, click into routes that look promising
4. Open Canvas and describe the full trip
5. Use Canvas to pressure-test the itinerary: ask it to add specific constraints, check if the logistics actually work
6. Verify every accommodation recommendation in Google Maps directly (don’t trust Canvas’s summary alone)
7. Book flights and hotels through their native sites, not through Canvas links

Step 7 matters. Canvas surfaces links but doesn’t complete bookings. Full agentic booking (where the AI actually completes a reservation) is on the roadmap for later in 2026. It’s not live yet. Don’t expect to hand your credit card to the AI and walk away.

---

Who Gets Real Value From This

Flexible travelers with loose dates: Flight Deals was built for you. If you can move your trip a few days in either direction, the flexible-date fare discovery is genuinely better than manually poking at the Google Flights date grid.

First-time international trip planners: Canvas reduces the paralysis of starting from a blank itinerary. The framework it generates isn’t perfect, but it’s a better starting point than reading fifteen different blog posts and trying to synthesize them.

People overwhelmed by TripIt/manual planning: If you find itinerary tools complicated, Canvas is more approachable. Lower floor, lower ceiling.

Who should skip it:

Business travelers with fixed dates and specific routing needs. The AI adds no value for LAX-JFK on a specific date, first class, aisle seat. Regular search is faster.

Travelers who care about data privacy. You’re sharing destination, travel dates, and trip details with Google. If you already use Google Maps and Gmail, this is the same trade-off you’ve already made. If you avoid Google products deliberately, Canvas doesn’t change the calculus.

Experienced planners who already have a system. If you’ve booked 20+ international trips and have a research workflow that works, there’s nothing here that will save you meaningful time.

---

The Bottom Line

Flight Deals is useful if you’re flexible. Not because the AI magic is remarkable, but because finding deals on unfixed dates has always been tedious with traditional search tools, and the natural-language input genuinely speeds up the exploration phase.

Canvas is a capable first-draft machine. Treat it that way. It builds the 60% structure of an itinerary faster than you would manually. You still need to verify everything specific: hours, closures, walking distances, whether that restaurant actually still exists.

Price prediction with confidence estimates is honest about its own limits. That’s worth something.

Full agentic booking isn’t here yet. When it arrives later in 2026, the equation changes. Completing reservations inside AI Mode would make this a genuinely different category of tool. Right now, it’s a better planning interface bolted onto the same search engine you already use.

Start with Flight Deals if your trip is open-ended. Skip it if you have fixed dates. Use Canvas to build an itinerary skeleton, then verify the details yourself. And don’t stop checking Skyscanner for routes where budget airlines matter.

---

Tested using Google Flights AI Mode in February 2026. Features are rolling out gradually; not all users will see Flight Deals or Canvas immediately. Features and pricing current at time of writing.