Google is testing AI checkout in India, a Flipkart Buy button inside Gemini and AI Mode, TechCrunch reported on 26 September. The checkout is Flipkart's, so Flipkart takes the order. The sale goes to the seller of whichever product the answer names, and a brand the answer leaves out never reaches the checkout.
When the AI answer is also the checkout, the platform picks which marketplace takes the order, and the pages the engine reads shape who gets named. On Gemini, most of those pages are brand websites. In the August 2026 Cited Index, brand websites were 72% of everything Gemini cited and Amazon.in was 0%, while Amazon.in was 9.9% of Google AI Mode's citations.
What is Google testing with Flipkart?
The test is narrow. Per TechCrunch, some users in India see a Buy button on select Flipkart listings in Gemini and in AI Mode in Google Search, for smartphones, electronics and mobile accessories. Tapping it opens a Flipkart-branded checkout without leaving the AI interface. Other users see the same Flipkart listings with no Buy option. Amazon listings appeared alongside Flipkart's in the test, with no way to buy them from inside the answer.
Google gave almost nothing away. A spokesperson told TechCrunch the company is "always testing new features" and had no further details to share. Flipkart did not immediately respond to TechCrunch. TechCrunch's single source said a wider rollout was planned for later in October, ahead of the festive season. As of mid-November we have not seen Google confirm it.
The plumbing behind it is also unclear. TechCrunch notes that it is not clear what technology powers the test. Google's Universal Commerce Protocol, its open standard for AI shopping, describes "agentic actions on AI Mode in Google Search and Gemini, starting with direct buying", and its developer page names no countries. The Flipkart test hands the shopper to a Flipkart-branded checkout, which TechCrunch says looks different from the Google-hosted checkout Google first described for that protocol.
Who gets the sale when AI checkout sits inside the answer?
Three parties touch an order placed this way.
- Flipkart gets the order. The checkout is Flipkart-branded, so the order is placed with the marketplace, not with Google and not with the brand's own store.
- The seller on the listing gets the sale. If your product is the one the answer recommends and the listing is on Flipkart, you sell. If a rival's product is the one shown, they do.
- Everyone the answer skips gets nothing. There is no results page to scroll. A shopper who taps Buy inside a recommendation has already stopped comparing.
That last point is the one I would plan around. In classic search, a brand lower down the results page could still win a click. In an answer with a Buy button, the shortlist is the store. Being named is the first step of the sale, before price, reviews or delivery speed get a vote.
Which AI engines cite marketplaces at all?
Want to know how your brand scores on these same metrics?
We'll run 20 prompts across 3 AI platforms and send your AI Visibility Report within 24 hours.
Here the August data gets useful, because the five engines read marketplaces very differently. The table shows each engine's share of citations that went to Amazon.in next to the share that went to brands' own websites.
| AI engine | Amazon.in share of its citations | Brand websites share of its citations |
|---|---|---|
| Google AI Mode | 9.9% | 54% |
| Google AI Overviews | 4.8% | 54% |
| Perplexity | 1.2% | 50% |
| ChatGPT | 0% | 42% |
| Gemini | 0% | 72% |
Shares are of each engine's own citations in the August 2026 Cited Index (8,493 citations across 12 categories). Amazon.in was 3.1% of all citations across the five engines.
93% of Amazon.in's citations came from Google's two search products, AI Overviews and AI Mode. ChatGPT and Gemini showed no Amazon.in citations at all in the August data. The same marketplace carries real weight in one part of the AI market and none in another. That is why "be on Amazon" or "be on Flipkart" is not, by itself, an AI shopping plan.
Why is Gemini an interesting place for a marketplace button?
Gemini is the engine that leans hardest on brand websites. 72% of its citations in the August edition went to brands' own sites, the highest of any engine.
Put that next to the Flipkart test and the two jobs separate cleanly. On Gemini, the pages that shape the answer are mostly brand websites. The page that takes the money, in this test, is a Flipkart listing. A brand can win the first and lose the second, or the reverse:
- If your own site is strong and your product is not on Flipkart, Gemini can name you and the shopper still has nowhere to tap Buy.
- If your Flipkart listing is strong and your own site is thin, the button is ready, but on the engine that cites brand sites most, the answer may never name you.
On AI Mode the picture tilts the other way. It cited Amazon.in more than any other engine did, at 9.9%. There, the marketplace page can be both what the engine reads and where the order goes.
What should brands selling through marketplaces do now?
GEO (generative engine optimization) and AEO (answer engine optimization) are two names for the same job: getting your brand into the answers AI engines give. A Buy button raises the stakes of that job, because the answer now ends in a purchase. Four moves, in order:
- Keep your own site as the source of truth. In Indian consumer categories, brand websites were 42% of all AI citations in the August edition, the largest single type of source. Your site is the page type the engines cite most in these categories, so the product names, specs and prices on it should be complete and current.
- Make the marketplace listing match it exactly. If the answer names your product from your site and the Buy button opens a Flipkart listing, the two should describe the same product in the same words. A mismatched model name or spec can lose a shopper who is one tap from paying.
- Check which marketplace each engine links to for your prompts. Run the prompts your buyers type in each engine and note which retailer pages show up beside your product. On AI Mode it may be Amazon.in. On Gemini, in this test, the Buy button sits on Flipkart. That tells you where a missing or weak listing costs you most.
- Do not rebuild channel strategy on a test. Google has confirmed nothing beyond testing. Watch whether the button widens to more categories and more retailers before you move budget between marketplaces.
In the 3-Layer AI Visibility Stack, your own site carries the first two layers, Discoverability and Citability: engines can reach it and lift from it. The marketplace listing belongs to the third, Authority, the off-site pages that describe your product, and in an answer with a Buy button it is also where the order lands. For the full per-engine source mix, see how the five engines weight each type of source.
What we do not know yet
Four open questions, any of which could change this read:
- Whether the button reaches every user in India, or stays a test.
- Whether other retailers, including Amazon, get the same button. The test as reported covers Flipkart only.
- Whether Google takes a fee on these orders. TechCrunch's report does not say.
- Whether in-answer checkout changes which sources the engines cite. Our August data predates the test, so it cannot show that.
Two honest caveats
First, the August 2026 Cited Index was measured before we changed how we collect AI answers in late October, so read these shares as a snapshot of that month, not a trend. The 0% for Amazon.in on ChatGPT and Gemini may partly reflect how answers were collected in August. Second, domain-level counts come from each category's top 50 cited domains, so a thin tail of marketplace citations below that line is not counted. What the data does establish is direction: in August, the Google search products cited Amazon.in, and Gemini, the engine now testing a Flipkart Buy button, mostly cited brands' own websites. If the button widens beyond a test, a brand Gemini leaves out of the answer will not get to compete on price at all.
Methodology. 8,493 structured citations in the August 2026 Cited Index, across 339 Indian brands in 12 categories (seven consumer, three B2B software, two education) and five engines: ChatGPT (1,337 citations), Perplexity (1,465), Gemini (1,614), Google AI Overviews (3,159) and Google AI Mode (918). Each engine's share is that engine's citations to a domain or source type divided by all of its citations. Consumer-category figures pool the seven consumer categories. Domain-level counts come from each category's top 50 cited domains. Aggregate and market-wide: no tracked brand is named, only public marketplaces. The Flipkart test is as reported by TechCrunch on 26 September 2026. Google has not published details.
Want to see which sources the AI engines cite for your category, and whether your site or your marketplace listings are among them? Get a free AI Visibility Report, or start a free trial of Cited, the AI search optimization platform.