Gemini Spark Review 2026: Is Google’s 24/7 AI Agent Worth It in India?

What if your AI could keep working on a task long after you closed the chat? 

Google launched Gemini Spark in India on 29 July 2026, alongside 160 other countries. I spent the following week putting it through eight real-world tasks from Nainital, and one of them kept running for eight days without me touching it.

Disclosure: none of the links in this article are affiliate links. All testing was done on a paid Google AI Pro account.

In this article, you’ll learn what Gemini Spark is, how it works, how to set it up, and what you can actually use it for. Let’s dive in.

What is Gemini Spark?

Gemini Spark is a background AI agent from Google that runs multi-step tasks on Google’s servers and writes the results into your Google Drive. It works whether or not your laptop is open, and it can be scheduled to repeat.

Most Gemini features answer you and stop. Spark keeps going, and you can set it to repeat every evening without opening anything. That’s the whole product. Everything else is Gemini with permission to click buttons. 

What do you need to use Gemini Spark in India?

1) Paid Subscription

Gemini Spark requires a paid Google AI Pro or Ultra plan, a personal Gmail account, and is limited to users over 18 years of age. Pro plan is ₹1,950/mo in India as of 11 August 2026, and Ultra plan starts at ₹6,500/mo, both from Google’s India subscriptions page. There is no free tier and no trial. If you are on Jio, you may already have Google AI Pro at no extra cost: here is how to get Gemini Pro free for 18 months in India.

2) A Personal Google Account 

Work and school accounts are blocked. Not throttled, not limited. They can’t open Spark at all, so if your main Gmail is a Workspace address through your employer, you’ll need a separate personal one. This one annoys me. The people most likely to want an agent chasing invoices are the ones on a work account.

3) Keep Activity Switched On

In order to access Gemini Spark, you need to keep your Activity switched ‘On’ within your Gemini app settings. It’s checked at the door.

4) A Supported Region

Gemini Spark is unavailable in the European Economic Area, the UK, Switzerland, and Nigeria, per Google’s Gemini Spark eligibility and availability documentation. India is fine.

Note: Google and Jio ran an offer giving Jio 5G users aged 18 to 25 eighteen months of AI Pro free, worth ₹35,100. Announced in October 2025, and Google said it would widen. Check whether you already have Pro before you pay for it. I haven’t verified the current eligibility rules.

How to Set Up Gemini Spark?

Now that you know what Gemini Spark is, what it does, and what you need before using it, let’s start with the steps to set it up.

Step 1. Open Gemini Spark

Go to Gemini app and click Spark in the left sidebar. It is still labelled Spark (Beta) as of 11 August 2026.

Gemini Spark also runs inside the Gemini mobile app on Android and iOS, and in the Gemini desktop app on Mac. There is no Windows desktop app.

Step 2. Turn on the apps you want it to reach

Connected apps are off by default. Only Google Workspace is enabled at the start, so Gemini Spark can see Gmail, Calendar, Docs, Drive, Sheets, Slides, Keep, and Tasks, and nothing else.

Everything beyond that is a toggle you flip yourself. Canva, Dropbox, Instacart, OpenTable, and Zillow Rentals are the third-party apps available. Enabling one takes a few seconds.

Step 3. Write the prompt as a brief, not a question

This changes your results more than anything else on this page. Give Spark a goal, numbered instructions, and the name of the file you want at the end. 

One-line requests get you one-line thinking. The example below proves this. 

Working Example: A Jaipur Itinerary in Under 3 Minutes

Spark planned a full Nainital to Jaipur trip in 2 minutes and 56 seconds and saved it as a formatted Google Doc without being asked twice.

What came back: 

Road transit to Kathgodam Junction (35 km) and Lalkuan Junction (60 km), a named train with times and class options, hotels in three budget bands, a four-day plan covering Amber Fort, Hawa Mahal, City Palace and Jal Mahal, and a monsoon warning.

The train was real, and the stations were the right ones for Nainital. That sounds like a low bar. It isn’t. Every AI trip planner I’d used before this one invented a station, and Kathgodam is the kind of detail they get wrong.

What Happened Across 8 Gemini Spark Tasks?

I ran eight tasks through Gemini Spark on 3 and 4 August 2026 on a Google AI Pro account. Six worked, one failed outright, and one produced data with a 2-in-12 error rate.

Task	Time	Output	Result
Jaipur trip itinerary	2m 56s	Google Doc	Worked
Daily 7 PM news and script	Scheduled	Google Doc, daily	Worked, 8 days
Gmail bill prioritisation	Not timed	In-chat only	Worked
Unsubscribe from promo email	Not timed	None	Failed
Invoice retrieval to Sheet	6m 40s	Google Sheet	Worked after a retry
iPhone 17 Pro price tracking	6m 28s	Google Sheet	Worked
Add iPhone to cart	1m 49s	Cart	Worked
Amazon and Flipkart gadget lists	Not timed	2 Google Sheets	Worked
Upcoming phone launches	Not timed	Google Sheet	2 wrong rows in 12

Three of these tasks are worth going into. Let’s discuss them below. 

1) Price tracking across five Indian retailers

Gemini Spark compared the iPhone 17 Pro (256GB) across five Indian retailers in 6 minutes and 28 seconds, extracted instant bank discounts, and wrote the result into a Google Sheet.

Retailer	Base	After instant discount
Vijay Sales	₹1,30,900	₹1,26,900
Amazon India	₹1,30,900	₹1,26,900
Reliance Digital	₹1,30,990	₹1,26,990
Flipkart	₹1,32,900	₹1,28,900
Apple Store India	₹1,34,900	₹1,29,900

Prices as checked on 3 August 2026.

Then I asked it to add the phone to the cart at the cheapest option. It opened a browser, went to Vijay Sales, typed in a pin code, and added the iPhone in 1 minute and 49 seconds. I checked the cart myself. The phone was there.

Vijay Sales was the joint cheapest with Amazon at ₹1,26,900, so it picked the right one. Worth admitting: I nearly published this as a failure. Our own notes had the Vijay Sales price transcribed incorrectly, which made it look like Spark had been asked for the cheapest and had gone to the dearest. 

The agent was right, and we were wrong, and I only caught it because I opened the sheet Spark had exported instead of trusting the write-up of it.

2) Finding invoices buried in Gmail

Three years of hosting invoices were found and pulled into a Sheet with a link back to every Gmail thread in 6 minutes and 40 seconds.

The first sheet link threw a permissions error. Spark noticed on its own, diagnosed the Drive sharing setting, and fixed it before I’d finished reading the error. That’s the most impressive thing it did all week, and it’s also why the task took nearly seven minutes instead of four.

3) Researching upcoming phone launches

This is the one that should make you cautious. Spark built a clean sheet of twelve phones launching in India between August and September 2026, with specs, expected prices, and source links. It looked great. Two of the twelve rows were wrong.

It listed the Huawei Nova 16 SE launching on 5 August, and Huawei doesn’t launch phones in India. It listed the OnePlus Nord 5 as a September 2026 launch. The Nord 5 came out in 2025.

The cause is sitting right there in the sheet’s own source column. Spark scraped two “upcoming mobiles” aggregator pages, a price-list site, and a blog post from a Polish mobile operator. Not one manufacturer announcement.

That matters more than the error count. Spark didn’t reason badly; it sourced badly. Point it at a topic where the top results are content farms, and you get the content farms’ mistakes back in a tidy spreadsheet, which is worse than getting them messily. The formatting makes them look checked.

Limitations of Gemini Spark

1) It could not unsubscribe from anything

Spark read my inbox and sorted the promotional senders properly into financial offers, e-commerce, entertainment, and subscriptions. Genuinely good categorisation. Then it unsubscribed from none of them.

This is the failure that bothers me most, because the hard part worked. Working out which of a few hundred senders are junk is the bit I can’t do quickly. Clicking unsubscribe is the bit I can. Spark did the thinking and stopped at the doing, which is the wrong half to get right when the pitch is that it acts for you.

So I changed the instruction and asked it to move promotional mail to the trash instead. That worked, with a catch. Every message threw a ‘Confirm Trash’ dialog showing the sender and subject, and I had to click Allow on each one.

When I asked it to allow everything at once, it paused partway through.

2) There is no local Chrome option in India

Google shipped the ability to drive your own desktop Chrome, using your logins and saved passwords, on 30 July 2026. It is US-only. Google’s auto browse requirements state it plainly: you must “be 18 or over and in the US”.

There was no option to use my local Chrome. Everything ran through Google’s remote browser, which is why the cart task had to type a PIN code in rather than use an address already sitting in my account.

Worth knowing for when it does reach India: auto browse is capped at 20 requests a day on AI Pro and 200 on Ultra, and it doesn’t work in Incognito, iPhone, & iPad. If you’ve read a US review praising Spark for working inside your real browser session, you don’t get that version here.

Is Gemini Spark’s Scheduling Reliable?

In my testing, yes. A Spark task set to run at 7 PM daily produced a document every evening from 3 August through 10 August 2026. Eight consecutive days, no re-prompting, no re-authorising, nothing to click.

The job was to fetch the day’s tech news and draft a Hinglish video script in Google Docs. It set the whole thing up from one prompt plus a single refinement. Eight days is a week, not a quarter, and I’ll say plainly that I don’t know what happens on day sixty.

But every piece I’ve read about Spark lists scheduling as a feature and then goes quiet about whether it holds. Mine held for eight days while I wasn’t looking, and that’s the thing I’d actually pay ₹1,950 for.

Should You Pay for Gemini Spark in India?

Pay for it if you have recurring work. A weekly report pulled from a sheet, a daily digest, a monthly invoice chase. Jobs where the output lands in a document you read later, and where being wrong twice in twelve costs you a correction rather than a mistake.

Don’t pay for the browsing demos. They work. They also need you watching, which makes them a party trick rather than automation. I wouldn’t hand Spark anything that spends money or sends mail on my behalf without reading every line first, and I don’t expect that to change while it’s still asking me to approve each email individually. If ₹1,950 a month is more than you want to commit, the cheaper Indian plan to compare it against is ChatGPT Go at ₹399 a month as of 11 August 2026, and if you would rather spend nothing at all, Google Omni Flash is free to use.

The same logic applies to automating emails with Make.com and to saving reusable prompts as Skills in Chrome. Repeatable workflows are module 4 of AI For Working Professionals, though the setup above is complete on its own. You do not need the course to run any of it.

I will re-test this when Gemini Spark leaves Beta, and again if Chrome auto-browse reaches India, because that changes what the cart and booking tasks can do. If your scheduled task has run longer than eight days, or if it quietly stopped, tell me how long it lasted. That is the number I most want and the one I cannot get in a single week.

Conclusion

In this article, we saw what Gemini Spark is, how it works, how to use it, and some of its major use cases. But before you jump in, there are a few things worth keeping in mind. Read everything it researches: in my testing, I found two wrong rows out of twelve. So, treat its output as a first draft with citations attached, not a final answer you can blindly trust. 

Also, don’t worry too much about the 15-task limit that Google documents. I peaked at just three concurrent tasks, so you’re unlikely to run into that ceiling during normal use. Spark also asks for permission before anything destructive, which is good for safety but means “automate my inbox” isn’t completely hands-off. 

And remember, it’s still in Beta and has already changed twice in three months, so some of this may evolve quickly. Finally, if you’re using free Gemini, a work or school account, or you’re in the UK or EEA, Spark isn’t available to you right now. There are no workarounds for these restrictions.

FAQs