Your best customer in 2027 might not see your homepage at all. Not the hero video, not the carousel you spent three weeks arguing over. They won’t have shopped, exactly. Their agent will have.
Agentic commerce has stopped being a keynote slide. There’s a working checkout flow behind it now, and it’s scaling faster than anything you’re currently buying media on. Most Australian DTC brands are still arguing about attribution windows while the front door of ecommerce moves into a chat window.
The definition worth keeping is this one: agentic commerce is shopping where an AI agent does the searching and comparing, and increasingly the buying. The human writes the brief. The agent finishes the job.
Which is awkward, because every stage of your funnel was designed for a human who now has no particular reason to turn up.
Two problems, one lazy panic
Founders keep mashing two separate things together and calling the result “AI search.”
The first is that Google is keeping your clicks. Pew Research Center tracked 68,879 real Google searches in March 2025 and found that when an AI summary appeared, people clicked through to a traditional result on 8% of visits, against 15% when there was no summary. Clicks on links inside the summary itself ran at 1%. That’s a channel eating its own referral traffic.
The second is the one nobody’s looking at, and it isn’t really a problem at all: a new channel is being born at the same time, and it converts better than the one everyone’s mourning.
Adobe Analytics has been tracking traffic arriving at US retail sites from generative AI tools since October 2024. The trend line reads like a typo. Up 1,324% since then, still growing 138% year on year as of May 2026, and those visitors convert 31% better than other sources. Over the 2025 holiday period, AI-referred traffic was up 693% year on year with revenue per visit up 254%.
So the doom story is about half right. The old tap is tightening. But the shopper who arrives from ChatGPT or Perplexity has already done the research, in a conversation you weren’t part of, and lands on your product page more or less sold. Fewer visitors, better ones.
Which changes the question. It isn’t how you get your traffic back. It’s how you become the answer.
What happened while you were watching your ROAS
In September 2025, OpenAI and Stripe launched Instant Checkout in ChatGPT, starting with US Etsy sellers and extending toward more than a million Shopify merchants. Glossier, Vuori and SKIMS were among the early names. A buy button, inside the chat, no website visit required.
Alongside it they published the Agentic Commerce Protocol, an open standard for how AI agents talk to merchant checkouts. In plain terms: they built the plumbing so any AI assistant can complete a purchase at any merchant who switches it on.
And the money is following. WPP Media’s mid-2026 This Year Next Year forecast calls generative search the fastest-scaling ad channel ever recorded, growing from US$5.1 billion in 2026 to over US$100 billion by 2030. Fastest ever recorded. Faster than search was, faster than social was. I’d hold the 2030 number loosely, since every 2030 forecast is really a sales document, but the shape of it is hard to argue with.
The last two times a channel scaled like this, the brands that moved early bought attention at prices that never came back, and the ones who waited paid the late tax. You were probably around for at least one of those.
Why most brands will sleepwalk through it
Mainly because it doesn’t show up cleanly in a dashboard, and marketers are trained not to trust anything they can’t put a budget behind. There’s no self-serve ads manager for “be the recommendation,” so there’s nothing to optimise and no line item to defend at the quarterly. Meanwhile the traffic decline arrives slowly enough that it never triggers a board conversation, right up until it’s the only board conversation.
Here’s the uncomfortable part. The agent has no opinion about your brand fonts. It reads your product data, your reviews, what publishers and Reddit threads and forum posts say about you, and synthesises a recommendation out of all of it. Your website becomes a warehouse and the shelf moves into the conversation. If you’ve just signed off a site redesign, that’s a genuinely irritating thing to read.
The Australian timing gap is the actual opportunity
One more thing before the to-do list, because this matters more if you’re reading from Melbourne or Sydney than from New York.
Instant Checkout launched US-only. Etsy first, US Shopify merchants next, wider regions in “future updates.” The recommendation layer, though, doesn’t respect borders. Australians are already asking ChatGPT and Perplexity what to buy, and the engines are already forming views about your category and your competitors. The buying rail gets here later. The reputation is being written now.
That sequencing is a gift, and I don’t think many people here have clocked it. The work that decides who gets recommended (clean product data, pages worth quoting, third-party proof) can all be done before the checkout rail switches on and the category leaders lock in. Nobody sends a memo when the shelf fills up.
The brands I’ve watched treat a platform shift as a quarter of unglamorous preparation, rather than a fortnight of panic when the headline lands, tend to be the same brands whose CAC everyone else envies eighteen months later. This one’s no different. It’s just earlier than usual.
What I’d actually do this quarter
For a $50k–$500k/month Australian brand, none of this needs a new tool subscription.
Start with your product data, and treat it as something a machine is reading, because one is. Clean titles. Descriptions in plain English that answer the questions people actually ask: who it’s for, what it does, what it costs, whether you ship to Perth. Specs in structured form, product and FAQ schema intact. Agents quote pages that answer directly, so put the answer high up, in sentences a fourteen-year-old could repeat back.
Then work on getting cited elsewhere. These engines recommend brands that other sources discuss. Reviews on platforms the models crawl, press and publisher mentions, a real presence in the communities where your category gets argued about, comparison content on your own site that names competitors honestly. If the only place your product is described is your own website, a system built on cross-referencing has nothing to cross-reference.
Make sure you can be bought. If you’re on Shopify, watch the ACP rollout and switch it on when it reaches AU merchants. The margin conversation about agent fees is a real one, but it’s a later one. Being absent from the shelf costs more than the commission does.
And instrument it before you judge it. Segment AI referrals (ChatGPT, Perplexity, Copilot, Gemini) in GA4 now, so that in six months you’re reading a trend rather than guessing. With the AU food and consumer brands I work with, the volume today is tiny, low single digits of sessions. That’s precisely why now is when you set it up. Learning is cheap while the numbers are small.
Keep running your ads through all of it. Agentic commerce doesn’t replace demand creation. Someone still has to make a human want the thing before an agent goes and fetches it. The brands that come out of this well will be doing both jobs at once: building memory with people, staying legible to machines.
If you want a second pair of eyes on how findable your brand is to AI engines, and whether your product pages would survive an agent reading them, book a growth call with Thrive. Twenty minutes, no deck.
FAQ
What is agentic commerce? Agentic commerce is online shopping where an AI agent searches, compares and completes purchases on a person’s behalf. The shopper states what they want; the agent recommends and, through systems like OpenAI and Stripe’s Agentic Commerce Protocol, can finish the checkout inside the chat interface.
Is AI search really sending ecommerce brands traffic? Yes, and it’s growing fast. Adobe Analytics reports AI-referred traffic to US retail sites grew 1,324% between October 2024 and mid-2026, and those visitors convert 31% more than other traffic sources. Volumes are still small relative to Google, but the growth rate and conversion quality are unmatched.
Do AI Overviews reduce clicks to websites? Yes. Pew Research Center found users clicked a traditional search result on only 8% of visits when Google showed an AI summary, versus 15% without one, and clicked links inside the summary on just 1% of visits.
How does an ecommerce brand get recommended by ChatGPT? Make product pages that answer buyer questions directly in plain language, keep structured product data and schema clean, and build third-party evidence: reviews, publisher mentions, community discussion. AI engines preferentially cite sources that are clear, current and corroborated elsewhere.
