Agentic Commerce: Will AI Agents Recommend Your Brand?

Agentic commerce is online shopping where an AI agent does the buying for a person: it takes a goal, finds the options, compares them, and completes the purchase. For brands, that rewrites the oldest question in ecommerce. For twenty years the job was to get a human to your page and persuade them. Now the visitor sizing up your product, and increasingly paying for it, is software that does not care about your hero image. The question is no longer whether people can find you, but whether an agent will find, trust, and recommend you, and check out when it does.

What Is Agentic Commerce?

Agentic commerce is a model of online shopping where an AI agent acts as a delegated buyer. The shopper sets the goal and the guardrails ("find a waterproof size-10 hiking boot under $150, delivered by Friday"), and the agent does the searching, comparing, and, when allowed, the buying. You will also see it called a-commerce, and the agents doing the work called AI shopping assistants or shopping agents. Put simply: the agent does not help with the purchase, it is the buyer.

That is the line from regular ecommerce. A traditional shopper opens ten tabs, reads reviews, and fills in a checkout form. An agent collapses that into one instruction and returns a short list or a finished order. The person may never see your product page or your cart.

For a brand, the shift is uncomfortable in a specific way. The thing sizing up your product is a model, not a person, and it weighs your listing on machine-readable signals, not design or copy. The job now is making sure that when an agent shops your category, it can find you, read you correctly, decide you are the answer, and pay you.

How Big Is Agentic Commerce, Really?

Real, but earlier and messier than the headlines suggest. "Agentic commerce" is not defined the same way twice, so the numbers measure different things and land orders of magnitude apart. Read them with the scope attached.

The demand signals are more concrete than the sizing. IBM found 45% of consumers already use AI in the buying journey, Adobe clocked roughly 805% year-over-year growth in AI-referred retail traffic around Black Friday 2025, and Morgan Stanley put Americans who made an AI-assisted purchase last month near 23%.

Now the hedge most explainers skip. Interest cooled from a spring-2026 peak, plenty of "agentic" pilots never complete a purchase, and McKinsey notes adoption tracks "comfort, norms, and credibility" more than raw capability. Treat the trillion-dollar forecasts as a direction, not a budget line, and find out where your brand stands today.

Can an AI Agent Find Your Brand?

Before any of the protocol talk matters, an agent has to answer four questions about you: can it find you, can it understand you, will it trust and recommend you, and can it buy from you. Most coverage jumps straight to checkout. The earlier gates are where brands actually lose.

Start with discovery, because nobody has solved it. The payment protocols define how a transaction completes, not how an agent finds your product in the first place. With no agreed discovery standard yet, visibility comes from the unglamorous basics done well.

Be Reachable

The answer-fetching bots behind ChatGPT and Perplexity, and the agents acting on your pages, have to get in and read you. If your prices, specs, or whole product pages only render after JavaScript runs, many of them see an empty page. Our guide to AI crawlers covers which bots matter and how to control them.

Be in the Feed

Agents lean on structured product data and real-time availability, so a clean, complete product feed (Google Merchant Center, valid GTINs, full attributes) is now table stakes. Often the issue is not a brand problem at all but a catalog-data one: when the same item arrives from several feeds with different part numbers and no shared GTIN, the model sees several uncertain items and cites none of them. Whether a thin file like llms.txt helps is a smaller question; the reachable page and the accurate feed are what actually matter.

Can an Agent Understand Your Products?

Being reachable gets you in the room. Being understood gets you considered. Agents do not read your page the way a person does; they extract structured data and act on it, so the gap between what your page shows a person and what it states in machine-readable form is where you lose.

The consensus across every serious player is the same: move from human-readable to machine-readable. That means schema markup in JSON-LD, GS1 standards, and full attributes (price, brand, variants, sizing, availability, sustainability) in a form an agent can parse without guessing. Different vendors call it different things; it is the same instruction.

Two practical points matter more than the acronyms. First, freshness: agents drop a recommendation when price or stock turns out wrong, so batch-hourly feeds are a liability. Second, the hard one: agents reward accuracy and consistency, not visual flair. If your edge is craftsmanship or sustainability rather than price, encode it into structured signals the model can weigh, or it flattens you to a spec sheet and picks someone cheaper.

This is the same discipline as generative engine optimization, pointed at the purchase instead of the citation.

Will an Agent Trust and Recommend You? (And How to Check)

Discovery and structure get you eligible. Being recommended is the prize, and it is the part the enterprise playbooks skip.

The question brands actually ask, in their own words: why does the agent recommend an inferior competitor while my better, cheaper, in-stock product gets skipped? Sometimes it is the structured-data gap from the last section. Often it is trust signals the model can verify: reviews, ratings, return and warranty terms, consistent specs, even structured sentiment that lets it warn a shopper a shirt "runs small." Get those right and the agent keeps recommending you; get them wrong and it quietly moves on.

The deeper problem is that the engines do not show you this. No native dashboard inside ChatGPT or Gemini tells you how often agents recommend you versus a competitor, or why you lost the comparison, so that signal has to come from outside the engines. That blind spot is why agentic visibility feels like guesswork. You cannot fix a recommendation you cannot see.

Be careful what you assume carries over. Your backlink profile and domain authority, the moat you spent years on, count for far less here. Models lean on third-party best-of lists, reviews, and awards, and a given answer names only a few brands, closer to winner-take-all than the ranked page of links you know.

In our experience auditing brands for AI visibility, most have never run the check, and the first scan is usually a surprise: a competitor named in the exact spot they assumed they owned. The fix is to treat it like rank tracking for a new surface. Measure your AI visibility across engines, watch your AI share of voice on the prompts that matter, and find the places where AI cites a competitor instead of you, so you know where the recommendation is leaking.

Can an Agent Actually Buy From You?

This is where the hype and the reality split, and it pays to be precise because most write-ups are months out of date.

What's Actually Live

The headline version, "you can buy things inside ChatGPT," is now only half the story (the picture below is mid-2026). OpenAI launched Instant Checkout in ChatGPT in September 2025 on the Agentic Commerce Protocol, starting with Etsy and Shopify merchants, then pulled the native in-chat checkout back around March 2026 after only a tiny fraction of merchants went live: shoppers researched in ChatGPT but finished the purchase on the store. ChatGPT now leans discovery-first, comparing products and handing the shopper to the retailer's own checkout. Google went the other way: its Universal Commerce Protocol, launched in January 2026 with Shopify, Target, Walmart, Best Buy and others, runs end-to-end buying inside Gemini and Google's AI Mode. So the real picture is two competing rails, not one, and merchants are not listed automatically on either. They apply.

The Protocols Doing the Plumbing

Underneath, a stack of protocols does the plumbing, from OpenAI and Stripe's ACP to Google's Agent Payments Protocol.

Of these, Coinbase's x402 is drawing the most developer attention right now as a stablecoin settlement rail; we break down what x402 is and how it works separately.

A second route skips the protocols entirely: agentic browsers like Comet and ChatGPT Atlas drive your existing site the way a person would, which makes "does my checkout work for a machine" its own question, covered in is your site agent-ready. And because Gemini's and Perplexity's checkout flows are still moving fast, treat any single vendor's claim as a snapshot. The smart play is to be ready on the open rails through your existing platform and keep earning citations in ChatGPT, since discovery now does most of the work.

Your Agentic Commerce Readiness Checklist

You do not need to rebuild your stack. Clear the four gates in order, and most of the work overlaps with the GEO and technical SEO you should already be doing:

1. Reachable. AI crawlers can fetch your pages, and nothing critical hides behind JavaScript.
2. Understandable. Product schema in JSON-LD, GS1 standards, valid GTINs, full attributes.
3. Accurate. Price and stock are correct at query time, not on an hourly batch.
4. Trustworthy. Reviews, ratings, and return and warranty terms exposed and consistent.
5. Payable. Your platform supports agent-initiated checkout on the open rails like ACP and AP2.
6. Measured. You can see whether agents recommend you, and where they pick a competitor.

If you do one thing this week, do the first and the last: confirm you are reachable, then check where you stand.

What Agentic Commerce Means for Your SEO and GEO Work

The anxieties underneath this are real. Brands worry about disintermediation, becoming a faceless supplier while the AI keeps the customer and the data. They worry about pay-to-play, the "discovery premium" McKinsey hints at, where being the agent's default becomes a paid placement the way ad slots did. And they worry the whole thing is hype.

On disintermediation, the defensible move is to be both the recommended product and the merchant of record. You keep fulfillment, returns, the post-purchase relationship, and first-party data when the sale runs through your own agent-ready storefront, not only inside a platform's catalog. And not every channel is opening up: some marketplaces, including eBay and Amazon, have moved to restrict or block unauthorized third-party shopping agents, guarding their own customer relationship. Your site and the open rails are where you have control; marketplaces are a moving, sometimes hostile, target.

The grounded read on the rest is calmer. Agentic commerce is not a new channel that replaces your SEO; it is the same core job (be found, be understood, be trusted) extended to the purchase. The structured data, clean feed, reviews, and citations you build to win AI search are most of what gets you recommended too. The discipline carries over even where the old ranking signals do not, and for the first time you can measure it. If you have not mapped how the pieces fit, GEO vs AEO vs SEO is the place to start.

You Don't Have to Guess

The brands that handle agentic commerce well are not the ones with the biggest budgets. They are the ones who looked. Most never check whether agents can reach their pages, read their products, or recommend them over a competitor, so they find out the slow way, when the traffic moves and the cause is invisible in their analytics.

You can close that gap in an afternoon. geotoolbox is built for exactly this: a free AI-readiness score tells you whether agents can reach and read your site, and the visibility tools show where AI recommends you, and where it does not.

Frequently Asked Questions

What is the difference between agentic commerce and agentic AI? Agentic AI is the general capability: software that can plan and take actions on its own. Agentic commerce is that capability applied to shopping, payments, and merchants, where an agent finds products, compares them, and completes a purchase for a person.

How big is the agentic commerce market? Estimates vary widely because the term is defined inconsistently. Grand View Research put the market at $5.71 billion in 2025, growing to $65.47 billion by 2033, while McKinsey projects agents could orchestrate up to $1 trillion in US B2C retail and $3 to $5 trillion in global goods spend by 2030. Treat the long-range figures as direction, not precision.

Is ChatGPT Instant Checkout still live? Not in its original form. OpenAI launched native in-chat Instant Checkout in September 2025 and pulled it back around March 2026 after it struggled to scale. ChatGPT now shows and compares products and sends shoppers to the retailer's own checkout via the Agentic Commerce Protocol (Etsy and Shopify merchants), while Google's rival Universal Commerce Protocol runs similar buying inside Gemini.

Will I have to pay to be recommended by AI agents? For now, ChatGPT shows product results as organic and unsponsored, and merchants pay a small fee only on completed purchases, not for placement. Whether a paid "discovery premium" emerges later is an open question, and a reason to build organic agent visibility now, while it is still earned rather than bought.

How do I check whether AI agents recommend my brand? Run the prompts a customer would and see who gets named, then track it across engines over time. A tool that monitors your AI share of voice and flags where AI cites a competitor instead of you turns a one-off spot check into something you can manage.

What is an example of agentic commerce? You tell an assistant "reorder my coffee and find a grinder under $80 with good reviews." It checks your history, compares grinders across retailers, picks one, and either buys it or hands you a one-tap checkout. You may never open a product page.

Sources

• Buy it in ChatGPT: Instant Checkout and the Agentic Commerce Protocol - OpenAI
• OpenAI's ChatGPT Instant Checkout plan just changed - Search Engine Land
• Google launches new protocol for agent-driven purchases (AP2) - TechCrunch
• Agentic Commerce Market Size & Share Report - Grand View Research
• The agentic commerce opportunity - McKinsey
• What Is Agentic Commerce? - IBM
• Under the Hood: Universal Commerce Protocol (UCP) - Google Developers
• 43% of Retailers Are Piloting AI Shopping Agents - PYMNTS
• Agentic Commerce Market Impact Outlook - Morgan Stanley