On 5 October 2026, TikTok added an in-app checkout and an AI Shopping Assistant to its app. The company said these were the first of several agentic commerce features on the way. The assistant answers questions about sizing, shipping and stock, and then lets people buy without leaving the conversation.
If that sounds like a small product update, look at the rest of the year. Google launched the Universal Commerce Protocol at the NRF retail show in January, so AI agents can read merchant catalogues and complete purchases through one open standard. OpenAI and Stripe's Agentic Commerce Protocol has been live in ChatGPT since September 2025. Shopify now lets merchants sell inside ChatGPT, Microsoft Copilot, Google AI Mode and Gemini at the same time.
So the shop window has moved. Some of the time it's your website. More and more, it's the inside of an AI answer.
What agentic commerce actually means
Strip away the jargon and it's simple. A person tells an AI assistant what they want, and the assistant does the legwork. "Find me trail running shoes under 9,000 rupees that can reach Pune by Friday." The agent reads product catalogues, compares specs, checks reviews and delivery promises, and comes back with two or three options. In some setups it can also place the order. The important change is who is doing the reading. For twenty years, your website was written for a person who scrolled, clicked and got distracted. An agent doesn't get distracted. It doesn't care about your hero video. It wants clean, structured, consistent facts, and it moves on quickly if it can't find them.
How big is this, honestly?
The numbers going around are big, and some are a stretch, so here's a fair reading.
Shopify reported that orders coming from AI-powered searches grew 15 times year on year through 2025. Morgan Stanley expects nearly half of online shoppers to use AI shopping agents by 2030, putting about a quarter of their spending through them. Those are real signals.
On the other side, commercetools' autumn 2026 review of the market calls fully agent-run checkout a 2030 bet rather than a 2026 one. Most people still click through and pay on the merchant's own site.
We think both views are right. Discovery has already moved into AI answers. Checkout will follow, more slowly and unevenly. The part you can't afford to ignore today is the first part: being found, understood and recommended. If an agent never puts you on the shortlist, it doesn't matter how good your checkout is.
It isn't only a retail story
Most coverage is about shoes, groceries and gadgets. B2B companies should pay closer attention, because their buyers moved first.
G2 surveyed 1,076 business software buyers in early 2026. Just over half (51%) now start their research in an AI chatbot more often than in Google, up from 29% a year earlier. And 69% said an AI chatbot pointed them to a different vendor than the one they had planned to pick.
Then there's procurement. Commercetools expects AI agents to take over a share of B2B ordering, approvals and even price negotiation. Think of a distributor's buying team asking an agent to source 500 industrial valves with a certain rating, from a supplier who can deliver within three weeks. If your catalogue lives in a PDF and your lead times live in a salesperson's head, you won't be in that answer.
This is why we keep telling manufacturers and IT services firms that "we're B2B, AI shopping doesn't apply to us" is the wrong conclusion. It applies to you first.
What an agent checks before it picks you
Nobody outside the AI companies knows exactly how each agent ranks options. But the protocols and the merchant guidance published so far point to the same handful of things:
- Structured product or service data: names, prices, specs, stock and variants in a format a machine can read, such as schema markup and clean product feeds.
- Delivery and returns facts. One logistics firm put it well this year: in agentic commerce, delivery stops being an after-sales detail and becomes a ranking signal before the sale.
- Third-party proof. Reviews, ratings, comparisons and mentions on sites the AI already trusts.
- Consistency. If your website says one price, your marketplace listing says another and your Google profile says a third, an agent has no reason to trust any of them.
A clear position. If your own pages can't say in one sentence who you're for, the agent will guess, and it may guess wrong.
Look at that list again. Only one item is purely technical. The rest are brand, marketing and content problems that happen to show up in a technical way.
Why this breaks when five vendors run your business online
Here's a situation we see often. A consumer brand runs a festive sale. The performance agency launches ads with the new price. The website team updates the product page. Nobody tells the person who manages the product feed, because that's a different freelancer. For a week, Google and ChatGPT see one price on the site and another in the feed.
A human shopper might shrug and click through. An agent sees conflicting data and moves to a competitor whose numbers agree.
Nobody did anything wrong in their own lane. The failure lives in the gap between lanes. Agentic commerce makes those gaps visible, because machines check everything against everything else, every time.
That's the practical case for having brand, marketing, digital experience and technology designed as one system. Not because it sounds neat on a slide, but because AI agents read your business as one system whether you run it that way or not.
A 90-day plan to get ready
You don't need to rebuild your company for robots. You need to make the facts about it easy to find and impossible to misread. Here's the order we'd work in.
- Weeks 1 and 2: see what the agents see. Ask ChatGPT, Gemini, Perplexity and Google AI Mode the ten questions your buyers actually ask. Note whether you appear, how you're described, and which competitors show up instead. This takes an afternoon and is usually humbling.
- Weeks 3 to 6: fix the data. Add or repair product and organisation schema. Clean your product feed and Google Merchant Center listing. Connect stock and pricing so the site, feed and marketplaces update together. Publish delivery times and return terms in plain text, not only inside images or PDFs.
- Weeks 7 to 10: fill the answer gaps. Write pages that answer comparison questions directly ("X vs Y", "best option for Z", "how much does it cost"). Start a review programme so recent customers leave feedback where AI tools look.
Weeks 11 to 13: test and measure. If you're on Shopify or a platform that supports agent storefronts, switch them on and watch. Set up analytics to separate AI referral traffic from ordinary search. Repeat the week-one questions and compare.
What we wouldn't do yet
A few warnings, because this space is full of people selling certainty.
Don't buy a tool that promises to "rank you number one in ChatGPT." Nobody controls that, and the answers change from one week to the next.
Don't hand an AI agent authority over your pricing or discounts without a person signing off. The payment networks are still building the trust and protection layers. American Express only announced purchase protection for registered agent purchases this year.
And don't tear down a website that converts well today. Most of the work above happens underneath the design: data, structure and consistency.
The bottom line
Agentic commerce is less about AI and more about whether your business tells one consistent story across every place a machine might look. That has always been good practice. It has now become a sales requirement.
If you'd like to see how your business looks to AI systems today, our free AI Growth Assessment scores five dimensions in 5 to 7 minutes. Or book a strategy session and we'll run the week-one test with you.
Sources: PYMNTS on TikTok's agentic commerce launch · nShift on agentic commerce and delivery · commercetools: 7 AI trends shaping agentic commerce · commercetools agentic commerce radar, autumn 2026