The pitch for agentic commerce is that a shopper tells an AI assistant what they want, and the assistant handles the rest: searching, comparing, checking stock and price, and either handing back a shortlist or completing the purchase outright. No product page visit, no reviews read, no carefully designed homepage seen.
Some of that is now real. A meaningful part of it was tried in 2025 and quietly walked back. Sorting the two is the whole job if you sell products online, so here is where things genuinely stand as of August 2026.
What Changed in 2026
The infrastructure moved fast. Google's chief executive Sundar Pichai unveiled the Universal Commerce Protocol at NRF 2026, co-developed with Shopify, Etsy, Wayfair and Target, with more than 20 payment and retail partners. UCP covers the whole commerce journey from discovery through to post-purchase, and it was deliberately designed to be compatible with the other emerging standards including AP2, A2A and the Model Context Protocol.
The mechanism is straightforward. Rather than an agent having to learn how each individual retailer's checkout works, it queries a manifest at a known endpoint on your domain, which declares which capabilities you support, which payment handlers are active, and where the relevant endpoints are.
Payments caught up in parallel. Mastercard completed its first live agentic transactions between February and March 2026 in Singapore, South Korea and other Asia-Pacific markets. Visa took its Trusted Agent Protocol commercial after sandbox piloting with more than 100 partners. American Express launched its Agentic Commerce Experiences developer kit in April 2026, alongside purchase protection for registered AI agent purchases. All three major card networks now support agent-initiated payments at some scale.
The Correction That Matters Most
Here is the detail that most coverage published before mid-2026 gets wrong, and it should shape your investment decision.
OpenAI deprecated ChatGPT Instant Checkout in March 2026. It had launched with considerable fanfare, powered by Stripe, aiming to let hundreds of millions of weekly users buy without leaving the conversation. The problem was conversion: purchases completed inside ChatGPT converted at roughly a third of the rate of those redirected to the merchant's own site.
The current model under the Agentic Commerce Protocol is therefore product discovery plus merchant redirect. Agents recommend, and shoppers complete the purchase on your site.
🛒 Why this is good news: the expensive version of preparing for agentic commerce, rebuilding your checkout so an agent can transact inside someone else's interface, is not currently the thing to do. The valuable work is making your products discoverable and machine-readable, which is largely product data hygiene you should be doing anyway.
Hold the Forecasts Loosely
The numbers being quoted are large. Morgan Stanley predicts nearly half of online shoppers will use AI shopping agents by 2030, accounting for roughly 25% of their spending. McKinsey estimates agentic AI will influence somewhere between three and five trillion dollars in global retail commerce by the same year.
Those are projections five years out for a category that has already reversed direction once. They are useful for establishing that this is not a passing curiosity, and close to useless for deciding what to spend this quarter. Most agentic capability today still clusters at the top of the funnel: browsing assistance, discovery, product matching and personalised suggestion lists.
What to Actually Do Now
The practical work is unglamorous, cheap relative to a replatform, and pays off in conventional search as well.
- Fix your product data first. Complete and accurate titles, structured attributes, real-world descriptions, current pricing and genuine stock levels. An agent making a confidence-based recommendation will skip a product it cannot fully parse. Incomplete attributes are the most common reason for being passed over.
- Expose a structured feed with product schema. This is the same groundwork that improves conventional visibility, which is why it is worth doing regardless of how agentic commerce develops. Our guide to schema markup for AI visibility covers the implementation.
- Make your policies machine-readable. Returns, shipping and warranty terms buried in prose on a policy page cannot be evaluated by an agent comparing options. Structure them.
- Sync price and inventory frequently enough to be honest. An agent that recommends an out-of-stock item, or one priced differently at purchase time, produces a failed transaction and a customer who blames you.
- Test your actual visibility. Ask ChatGPT and Google's AI Mode to find products in your category and see whether you appear at all. This takes ten minutes and is more informative than most reports you could commission.
- Sort out attribution. Decide now how you will identify sales that originated with an AI agent, because you cannot make a budget case for any of this if the channel is invisible in your reporting.
- Give it an owner. Feed freshness decays. Someone needs to be responsible for it.
How This Connects to What You Already Do
Almost none of the above is unique to agentic commerce, which is the reassuring part. Clean structured product data, accurate feeds and machine-readable policies are the foundations of e-commerce SEO too. If you have been putting off that work, agentic commerce is simply another reason to stop putting it off.
The genuine shift is conceptual. The goal stops being purely to win a click and starts also being to be the option an agent selects, which rewards clarity, completeness and certainty over persuasion. Our guide to generative engine optimisation covers that shift more broadly, and if you are already exploring paid placement in AI surfaces, ChatGPT shopping ads for Australian e-commerce covers the advertising side.
🚀 DigiWolf approach: we treat agent-readiness as a data quality project, not a platform migration. Audit the catalogue, structure what is unstructured, test what the assistants actually return for your category, then decide what if anything needs building. Book a free session if you want that audit run on your store.
The Bottom Line
The protocols have converged faster than most people expected, the card networks are live, and supporting UCP is no longer only viable for the largest retailers. But the checkout-inside-the-chatbot model has already been tried and pulled back, so the sensible position for an Australian retailer in August 2026 is to get the data right and stay close to the standards rather than rebuilding your storefront for a workflow that is still settling.
This is a fast-moving area and the details will keep shifting. The product data work will still be worth having whichever way it goes.
Nauman has spent more than 15 years in digital marketing across Australia, the UAE and Pakistan, working in-house and agency-side before founding DigiWolf. He holds an MBA in Finance & Marketing from the Institute of Business Administration, and works directly with every DigiWolf client on AI search, SEO and paid media strategy.