OpenAI Abandoned Checkout. Google and Perplexity Went All-In on Agents.
OpenAI dropped checkout. Google expanded booking. Perplexity built agents. Each path changes what brand visibility means on that platform.
OpenAI abandoned checkout. Google and Perplexity went all-in on agents.
In March, OpenAI quietly shelved ChatGPT’s Instant Checkout feature. Users were researching products in massive numbers, but almost nobody completed a purchase inside the app. Sales tax compliance wasn’t built. The experiment died.
That same month, Perplexity’s revenue jumped 50% after pivoting from search to AI agents that book, shop, and send emails on behalf of users. Google expanded AI Mode restaurant booking to eight new countries and started testing hotel and flight bookings with Booking.com, Expedia, and Marriott.
Three platforms. Three opposite bets on what AI should do when someone asks about a product or service.
This matters more than any individual feature launch because it determines what “brand visibility” even means on each platform. Getting cited in a ChatGPT answer is a different game than getting booked through Google AI Mode, which is a different game than getting recommended by a Perplexity agent completing a task. And most brands are running one playbook across all three.
Where each platform stands right now
The split happened fast. Six months ago, all three platforms were converging on the same model: answer questions, cite sources, maybe show some products. Now they’re heading in three different directions.
| Platform | Commerce approach | What happened | What “visibility” means |
|---|---|---|---|
| ChatGPT | Discovery only | Killed checkout in March. Focusing on research and recommendations | Getting cited when users ask purchase-related questions |
| Google AI Mode | Agentic booking + ads | Restaurant booking in 9 countries. Hotel/flight booking coming. Direct Offers ads live | Being in the merchant network agents can access, plus citation in AI answers |
| Perplexity | Full agent mode | Computer does tasks: shopping, email, social media. Revenue up 50% | Being in the product data and merchant APIs that agents query |
OpenAI’s retreat is the most telling move. According to Skift, ChatGPT users asked plenty of product questions but almost never bought. The company shifted its strategy to discovery and research, leaving transactions to third-party developers.
Google went the opposite direction. AI Mode now handles restaurant reservations across nine countries through OpenTable, Resy, and Tock. Users describe what they want in natural language — “Italian restaurant near downtown, party of four, Saturday at 7pm” — and AI Mode searches multiple platforms to surface bookable options. Google is working with Booking.com, Expedia, Marriott, and others to bring hotel and flight booking into AI Mode next.
Perplexity’s bet is the most aggressive. Perplexity Computer, launched in late February, acts on your behalf — shopping, comparing products, even sending emails. It’s not answering questions about products. It’s buying them. The pivot pushed Perplexity past $500 million in annualized revenue with more than 100 million monthly active users.
Why this split changes your visibility strategy
When ChatGPT answers “what’s the best CRM for a 50-person team,” it retrieves content, synthesizes an answer, and cites sources. The brand that wins has the best content, the strongest third-party mentions, and the most citation velocity. That’s search mode. The optimization levers are content quality, freshness, and authority.
When Google AI Mode books a restaurant, it doesn’t write an essay about dining options. It queries reservation APIs, checks availability, and presents bookable results. The restaurant that wins is the one connected to OpenTable or Resy with accurate availability data. Content quality is irrelevant if you’re not in the merchant network.
When Perplexity Computer shops for running shoes, it compares product data across merchant platforms, checks pricing and reviews, and can complete a purchase. The brand that wins has clean product feeds, accurate pricing, and structured data that agents can parse.
Three visibility problems that barely overlap:
| Mode | What determines visibility | Optimization levers |
|---|---|---|
| Search/citation (ChatGPT) | Content quality, authority, freshness | GEO, earned media, content structure, chunk-level architecture |
| Agentic booking (Google AI Mode) | Merchant network access, data accuracy, availability | Platform partnerships, reservation system integration, structured data |
| Agentic task completion (Perplexity) | Product feed quality, API accessibility, pricing accuracy | Schema markup, product feeds, merchant APIs |
The invisible shelf problem
Google Cloud coined a phrase for this: the invisible shelf. When an AI agent shops on behalf of a user, there’s no physical shelf, no search results page, no category browse. The agent queries product data and decides. If your data isn’t structured, complete, and accessible to that agent, your product doesn’t exist.
A Fortune article from March reported that some brands already attribute 10% of their revenue to agentic channels. McKinsey projects agentic commerce could drive $900 billion to $1 trillion in US B2C retail by 2030.
But the same query — “best running shoes for flat feet” — triggers search mode on ChatGPT and could trigger agent mode on Perplexity. The user’s intent is identical. The platform’s behavior is completely different. One writes a recommendation. The other starts shopping.
Your brand needs to be visible in both modes. And the tactics for each barely overlap.
What to do about the search mode side
If you’ve been following AI visibility trends, you know most of this playbook already. For platforms operating in search/citation mode (primarily ChatGPT, and the citation layer of Google AI Overviews):
- Publish content that directly answers the questions your audience asks AI platforms
- Build third-party authority through earned media, reviews, and community mentions
- Keep content fresh — 50% of AI citations come from content less than 13 weeks old
- Structure pages so each section works as a standalone chunk that AI retrieval can pull independently
- Monitor citation rates across models, because GPT-5.3 and GPT-5.4 cite completely different sources
The ChatGPT checkout failure actually reinforces the value of this approach. OpenAI learned that users treat ChatGPT as a research tool, not a store. The share of travelers using ChatGPT for trip planning rose 124% year-over-year (from 13% to 30%), even as checkout went nowhere. People trust ChatGPT to help them decide. They don’t trust it to handle their credit card.
That means citations and recommendations in ChatGPT carry real influence on purchases that happen elsewhere. Discovery visibility on ChatGPT might be worth more now than it was when everyone assumed checkout was coming.
What to do about the agent mode side
This is where most brands have a blind spot. Agent mode visibility requires a different set of investments.
Get into the merchant networks agents can access. Google AI Mode books through OpenTable, Resy, Tock, Booksy, and others. If you’re a restaurant not on any of those platforms, you’re invisible to AI Mode’s 75 million daily users, regardless of how good your website is. The same logic applies to hotels (Booking.com and Expedia integration coming), event venues (Ticketmaster, StubHub), and e-commerce (Shopify’s UCP integration).
Treat product data as your most important marketing asset. When an agent compares your product to a competitor’s, it reads structured data — prices, specs, availability, ratings, return policies. Incomplete or inaccurate data means you lose the comparison before a human ever sees the results. Google Cloud’s invisible shelf framework puts it bluntly: your data is your packaging now.
Expose clean APIs and product feeds. Perplexity Computer and similar agents need machine-readable access to your inventory. If your product information lives only in JavaScript-rendered pages that require human interaction to access, agents can’t read it. Schema markup, XML sitemaps, and standardized product feeds (Google Merchant Center, Shopify APIs) are the minimum.
Monitor pricing accuracy obsessively. AI agents cross-reference prices across sources. If your site says $99 but a third-party listing says $89, the agent will surface the discrepancy — or recommend the cheaper option. Price consistency across channels matters more when the buyer is a machine that can check every source simultaneously.
The monitoring gap
Most AI visibility tools (ours included) are built for search mode. They track citations, mentions, and share of voice in AI-generated text responses. That’s one half of the picture now.
The other half — whether your brand appears when an AI agent books a restaurant, shops for a product, or plans a trip — is largely unmonitored. There’s no standard metric for “agentic visibility.” No dashboard tells you that Google AI Mode recommended three competitors for a reservation query in your city and never surfaced your restaurant.
This gap will close. But right now, the brands gaining an edge are the ones manually testing: asking Google AI Mode to book in their category, running purchase queries through Perplexity Computer, checking whether their structured data actually shows up when agents go shopping. Tedious work. Also the only way to know.
What comes next
The platform divergence isn’t temporary. It reflects a deeper bet each company is making about what users want from AI.
OpenAI thinks users want a thinking partner — a research assistant that helps them make better decisions. Google thinks users want a doing partner — an agent that handles logistics and transactions. Perplexity thinks users want both, bundled into a single tool that can research, decide, and act.
All three could be right for different segments of the market. That’s the problem for brands. There’s no single AI visibility strategy anymore. There’s a citation strategy for search mode, a data strategy for agent mode, and a monitoring approach for each.
The brands that figure out both playbooks first will have an advantage that compounds over the next two years. Everyone else will be optimizing for a version of AI platforms that’s already outdated.
Want to know what AI platforms say about your brand? Try RivalHound free and find out.