Decolar closes its first sale with agentic AI, and SEO is worthless in this new game

Decolar told Neofeed (a Brazilian business news outlet) about what it calls its first agentic sale in Brazil: less than 12 hours after turning on the feature, a customer asked SOFIA (the company's generative AI agent) for lodging suggestions, and the tool searched, recommended, walked through the booking, generated a QR code, and received payment via Pix, all within the conversation. No traditional checkout, no switching screens.

It's a nice demo. But the fact that matters for people running growth and martech isn't the transaction, it's the line CEO Rafaela Rezende dropped midway through the interview: "being relevant to an agent is very different from being relevant to traditional Google SEO". That's the real bet, and it reorganizes how you think about channel, acquisition, and media.

The number that backs the story (and what it still doesn't back)

Before buying into the narrative: SOFIA already runs at scale. According to Neofeed, there are around 2 million conversations per month, and the agent takes part in journeys tied to 6.7% of Decolar's B2C revenue. That figure is what lends credibility to the story: it's not a lab experiment, it's volume in operation.

What the numbers don't say: 6.7% is "journey participation," not an end-to-end executed sale. The full agentic sale (the one that closes on its own, with payment) has just been born, a single transaction. The gap between "the agent took part in the journey at some point" and "the agent sold without a human" is exactly the hardest part, the one about payment and who's accountable for it. It's worth holding back the enthusiasm here.

The business backdrop: Prosus paid US$1.7 billion for the company in 2025 and wants to take it to US$6 billion by the end of the decade, tripling the group. Brazil already accounts for nearly 40% of Despegar's business. Agentic AI is one of the levers behind the goal of growing 40% a year in the country. In other words: there's valuation pressure pushing the agenda, which tends to inflate demos at launch.

Two channels are born here, and the second one is what matters

Neofeed is clear in separating two fronts that Decolar has opened, and they require quite different marketing skill sets:

| Front | What it is | Skill set it requires | |---|---|---| | SOFIA as salesperson | Decolar's own agent guides and closes the purchase within Decolar's channels | Conversational UX, chat conversion optimization, payment integration | | Inventory exposed to external agents | Decolar products findable and sellable via ChatGPT and other agents | Product data structuring, "agent optimization," presence beyond Decolar's own front end |

The first is an evolution of conversational e-commerce, more comfortable. The second is the rupture. Decolar is preparing for the traveler to not even open the app: they ask ChatGPT, and the agent pulls a flight, hotel, or package from Decolar's inventory. The consumer stops being "traffic" and becomes a query that happens in an environment the brand doesn't control.

For anyone who lives off channels, this breaks the mental model of the funnel. There's no more landing page to optimize, no more single place where you drop a pixel and measure. The question shifts from "how do I get the user to reach me and convert" to "how do I make my product the answer the agent picks".

GEO, the discipline that replaces SEO in this scenario

What Rezende calls "being relevant to an agent" has an emerging name in the market: GEO (Generative Engine Optimization). The logic differs from SEO on points that directly affect the work of people doing media and content:

• There's no SERP with ten links. The agent picks one option (or a few) and executes. There's no "second page" to fight for, no CTR to optimize. Either you're the choice, or you don't exist in that transaction.
• What matters is structured data, not persuasive copy. Price, availability, cancellation policy, and attributes need to be clean and accessible via API/feed for the agent to compare. A catchy headline doesn't convince a machine.
• Intermediation becomes the battlefield. Whoever controls the agent (OpenAI, Google, Amazon) controls the selection criteria. The brand loses the direct relationship with the customer and becomes dependent on how the agent ranks and presents things.

In practice, this means a budget reversal: part of what today goes to traditional search performance tends to migrate toward securing structured presence in agentic environments, and toward negotiating the rules of that presence. The warning case has already shown up: Neofeed notes that Amazon's Buy for Me, launched in April 2025, was accused by small businesses (speaking to Business Insider) of including them in the service without consent. Being "found by an agent" may not be a brand's choice, it's a power-negotiation position.

Why Pix changes the Brazilian game

The most Brazilian technical detail in this story is the payment rail. Abroad, agentic commerce is being built on cards: Visa Intelligent Commerce (April 2025) and Mastercard Agent Pay connect verified agents to card networks, with authentication and spending controls. In Brazil, Banco do Brasil and Visa carried out the first agentic transaction in March, still in a controlled environment.

Edson Santos, from Colink Business Consulting, points out to Neofeed the part that matters: if the regulator opens an agentic payment rail on top of Pix, and not just on cards, payment could go out via Pix or TED (a Brazilian wire transfer), which reduces concentration among a few players. Decolar closing the sale via Pix isn't a UX choice, it's a bet on which infrastructure will dominate Brazilian agentic commerce. For sellers, that changes transaction cost and changes who holds the purchase data.

The size of the market, and the counterpoint

The projection numbers backing the race: Morgan Stanley estimates that agent-guided purchases will move between US$190 billion and US$385 billion in US online retail by 2030 (10% to 20% of US e-commerce). McKinsey puts global agent-intermediated consumption at US$3 to 5 trillion by the end of the decade.

The honest counterpoint: these are long-term projections about a behavior that doesn't yet exist at scale. Travel is an emotional, comparative, high-ticket purchase category, exactly the kind of decision where consumers resist delegating. Booking a vacation hotel on "autopilot" is different from reordering detergent. Santos himself admits that "we're still at the beginning." Decolar might be right about the direction and too early on timing, and arriving early in an expensive channel has a cost.

What this leaves in the growth reader's hands

Three moves this case makes concrete:

1. Treat your product feed as a channel asset. If the agent sells from structured data, dirty data means invisibility. This applies to any e-commerce, not just travel.
2. Start measuring presence in agentic environments. There's no consolidated GEO metric yet, but you can test it today: how does your product show up when someone asks ChatGPT about your category? Who gets chosen, and why?
3. Reassess your dependence on owned traffic. The Decolar-iFood partnership (iFood, Brazil's dominant food delivery platform), which already accounts for 25% of new-customer base growth, with cross cashback, shows that ecosystem and integration gain weight when the direct front loses control. Ecosystem-driven acquisition lowers CAC in a world where the agent steals the relationship.

Decolar's agentic sale is a demo. The bet behind it, preparing inventory to be sold by agents the brand doesn't control, is what growth professionals need to start budgeting for now.