Your Next Customer Is an AI Agent: 9 Ways to Make Your Store Agent-Ready
Sep 13, 2026
Something quietly strange is happening to online stores. The visitor landing on your product page is increasingly not a person browsing on a lunch break. It is a piece of software, sent by a person, with instructions to compare six options, check the return policy, confirm the item is actually in stock, and report back.
For a while this looked like a novelty. It is not one anymore. Adobe Analytics measured a 393% year-over-year jump in AI-driven traffic to US retail sites in Q1 2026, and during the 2025 holiday season that figure hit 693%. The more interesting part is what those visitors do once they arrive.
That last visibility figure is the one worth sitting with. A third of what you carefully wrote, photographed, and priced simply does not register with the software now doing the shortlisting.
On 29 September 2025, OpenAI launched Instant Checkout in ChatGPT, starting with Etsy sellers and extending to over a million Shopify merchants. Alongside it, OpenAI and Stripe published the Agentic Commerce Protocol, an open standard that lets an AI agent complete a purchase while the merchant keeps the customer relationship, the fulfilment, and the support.
Google followed with the Agent Payments Protocol. Visa and Mastercard both shipped their own frameworks for letting a verified agent transact on a cardholder's behalf. Four of the largest players in payments effectively agreed on the same thing at the same time: agents are going to buy, and there needs to be a safe way for them to do it.
For store owners, this is not a distant platform shift. It is a checklist. Here is what that checklist looks like.
The most common reason a store is invisible to AI shopping is the least glamorous one. The crawler was blocked, usually by accident.
Check your robots.txt for OAI-SearchBot and GPTBot, then check again at the CDN and firewall layer, because Cloudflare and similar services now block AI crawlers by default in some configurations. A rule you never wrote can be quietly removing you from every AI shortlist in your category.
While you are there, make sure product pages are publicly reachable. No login walls, no region gate before the content loads, no chain of redirects between a link and the item it points to.
Agents do not squint at a photo and infer the size. They read fields. Missing or inconsistent fields are the single largest cause of a product being skipped.
At minimum, every item needs a stable unique ID at the variant level rather than only the parent product, plus brand, GTIN, and MPN. Then the specifics: materials, dimensions, weight, capacity, compatibility, and the attributes that separate one variant from another. Stock count should be tracked per variant, not per product.
This work is unglamorous and it is also the highest-return thing on this list. A catalogue with clean, complete, machine-readable attributes will outperform a prettier store with vague ones.
Schema.org Product markup with a complete Offer block is how you hand a machine the facts without asking it to guess. Price, currency, availability, condition, shipping details, return window, and aggregate rating all belong in the markup, and all of them need to match what a human sees on the page.
Mismatches between your markup and your rendered page are worse than missing markup. They teach the model that your data cannot be trusted.
There are three routes into AI shopping surfaces: publicly crawlable product pages, a structured feed submitted through the Agentic Commerce Protocol, and automatic inclusion via the Shopify catalogue for merchants on that platform.
Crawlable pages are the baseline. A dedicated feed is the upgrade, because it lets you control exactly what the agent sees rather than hoping a crawler interprets your layout correctly. If you are on a custom or headless build, this is a development task, and it is one worth scheduling this quarter rather than next year.
Marketing copy written to create a feeling tends to say very little that an agent can use. "Effortlessly elegant" is not a specification. "Full-grain leather, 34cm wide, fits a 16-inch laptop, 1.1kg" is.
The fix is not to strip out the personality. Keep the opening line that makes a human want the product, then follow it with a factual block covering materials, measurements, use cases, care, and compatibility. Humans skim it. Agents depend on it.
An agent completing a purchase is not entering a card number into your form. It is presenting a delegated payment credential on the buyer's behalf.
Merchants on Stripe can enable agentic payments with relatively little work. Merchants on other processors can participate through Stripe's Shared Payment Token API or the protocol's Delegated Payments specification without replacing their existing stack. Either way, someone has to make the decision and do the integration, and the stores that do it first will be the ones agents can complete a purchase with while competitors return an error.
Plenty of modern storefronts assemble their product content in the browser after the page loads. A shopper never notices. A crawler that does not execute your JavaScript sees an empty shell where your price and stock status should be.
Server-side rendering or static generation for product content is no longer a preference. If the price, availability, and specifications are not in the initial HTML response, assume a meaningful share of AI systems will never see them.
Agents cross-check. When an agent finds a price in your feed that does not match your page, or recommends an item that turns out to be out of stock, the correction is not just an embarrassment for that one sale. It is a signal about your reliability that affects future recommendations.
Feed refresh intervals measured in hours were acceptable when the audience was human. They are not acceptable when the audience compares your stated stock against your live page in the same second.
This is the tension nobody enjoys. Visa reported a 25% spike in malicious bot-initiated transactions, with a 40% surge in the US over six months. Accenture found 78% of financial institutions expect AI-linked fraud to rise. The instinct to tighten everything is understandable.
The answer is not a blanket block. It is verification. The new agent protocols exist precisely so that a legitimate agent can identify itself and prove it is acting for a real cardholder. Configure your protection to distinguish verified agents from unverified traffic, rather than treating every non-human request as an attack.
For a while this looked like a novelty. It is not one anymore. Adobe Analytics measured a 393% year-over-year jump in AI-driven traffic to US retail sites in Q1 2026, and during the 2025 holiday season that figure hit 693%. The more interesting part is what those visitors do once they arrive.
The numbers that should get your attention
AI-referred traffic now converts 42% better
In March 2026, visitors arriving from AI tools converted 42% better than standard traffic. One year earlier, the same traffic converted 38% worse. That is an eighty-point swing in twelve months.They spend more
Adobe recorded 37% higher revenue per visit from AI referrals in March 2026, reversing a gap that human traffic used to dominate.They read more carefully
AI-referred visitors spend 48% more time on page and browse 13% more pages per visit.Roughly a third of your product page may be invisible
Adobe's Content Visibility Checker found average language-model visibility of just 66% on retail product pages.The forecasts are not small
Bain projects a $300 to $500 billion US agentic commerce market by 2030. McKinsey puts orchestrated US retail revenue at $1 trillion by the same year.
That last visibility figure is the one worth sitting with. A third of what you carefully wrote, photographed, and priced simply does not register with the software now doing the shortlisting.
What actually changed
On 29 September 2025, OpenAI launched Instant Checkout in ChatGPT, starting with Etsy sellers and extending to over a million Shopify merchants. Alongside it, OpenAI and Stripe published the Agentic Commerce Protocol, an open standard that lets an AI agent complete a purchase while the merchant keeps the customer relationship, the fulfilment, and the support.
Google followed with the Agent Payments Protocol. Visa and Mastercard both shipped their own frameworks for letting a verified agent transact on a cardholder's behalf. Four of the largest players in payments effectively agreed on the same thing at the same time: agents are going to buy, and there needs to be a safe way for them to do it.
For store owners, this is not a distant platform shift. It is a checklist. Here is what that checklist looks like.
1. Confirm the agents can actually reach you
The most common reason a store is invisible to AI shopping is the least glamorous one. The crawler was blocked, usually by accident.
Check your robots.txt for OAI-SearchBot and GPTBot, then check again at the CDN and firewall layer, because Cloudflare and similar services now block AI crawlers by default in some configurations. A rule you never wrote can be quietly removing you from every AI shortlist in your category.
While you are there, make sure product pages are publicly reachable. No login walls, no region gate before the content loads, no chain of redirects between a link and the item it points to.
2. Fix your product data before you touch anything else
Agents do not squint at a photo and infer the size. They read fields. Missing or inconsistent fields are the single largest cause of a product being skipped.
At minimum, every item needs a stable unique ID at the variant level rather than only the parent product, plus brand, GTIN, and MPN. Then the specifics: materials, dimensions, weight, capacity, compatibility, and the attributes that separate one variant from another. Stock count should be tracked per variant, not per product.
This work is unglamorous and it is also the highest-return thing on this list. A catalogue with clean, complete, machine-readable attributes will outperform a prettier store with vague ones.
3. Mark up every product page with structured data
Schema.org Product markup with a complete Offer block is how you hand a machine the facts without asking it to guess. Price, currency, availability, condition, shipping details, return window, and aggregate rating all belong in the markup, and all of them need to match what a human sees on the page.
Mismatches between your markup and your rendered page are worse than missing markup. They teach the model that your data cannot be trusted.
4. Publish a proper product feed
There are three routes into AI shopping surfaces: publicly crawlable product pages, a structured feed submitted through the Agentic Commerce Protocol, and automatic inclusion via the Shopify catalogue for merchants on that platform.
Crawlable pages are the baseline. A dedicated feed is the upgrade, because it lets you control exactly what the agent sees rather than hoping a crawler interprets your layout correctly. If you are on a custom or headless build, this is a development task, and it is one worth scheduling this quarter rather than next year.
5. Write descriptions that survive being read by a machine
Marketing copy written to create a feeling tends to say very little that an agent can use. "Effortlessly elegant" is not a specification. "Full-grain leather, 34cm wide, fits a 16-inch laptop, 1.1kg" is.
The fix is not to strip out the personality. Keep the opening line that makes a human want the product, then follow it with a factual block covering materials, measurements, use cases, care, and compatibility. Humans skim it. Agents depend on it.
6. Get your checkout ready to accept a delegated payment
An agent completing a purchase is not entering a card number into your form. It is presenting a delegated payment credential on the buyer's behalf.
Merchants on Stripe can enable agentic payments with relatively little work. Merchants on other processors can participate through Stripe's Shared Payment Token API or the protocol's Delegated Payments specification without replacing their existing stack. Either way, someone has to make the decision and do the integration, and the stores that do it first will be the ones agents can complete a purchase with while competitors return an error.
7. Treat rendering as a visibility problem, not a performance one
Plenty of modern storefronts assemble their product content in the browser after the page loads. A shopper never notices. A crawler that does not execute your JavaScript sees an empty shell where your price and stock status should be.
Server-side rendering or static generation for product content is no longer a preference. If the price, availability, and specifications are not in the initial HTML response, assume a meaningful share of AI systems will never see them.
8. Keep pricing and inventory genuinely real-time
Agents cross-check. When an agent finds a price in your feed that does not match your page, or recommends an item that turns out to be out of stock, the correction is not just an embarrassment for that one sale. It is a signal about your reliability that affects future recommendations.
Feed refresh intervals measured in hours were acceptable when the audience was human. They are not acceptable when the audience compares your stated stock against your live page in the same second.
9. Block the bad bots without blocking the buyers
This is the tension nobody enjoys. Visa reported a 25% spike in malicious bot-initiated transactions, with a 40% surge in the US over six months. Accenture found 78% of financial institutions expect AI-linked fraud to rise. The instinct to tighten everything is understandable.
The answer is not a blanket block. It is verification. The new agent protocols exist precisely so that a legitimate agent can identify itself and prove it is acting for a real cardholder. Configure your protection to distinguish verified agents from unverified traffic, rather than treating every non-human request as an attack.
What this means for your roadmap
There is a version of this shift that is genuinely uncomfortable, and it is worth naming. When an agent does the comparing, your homepage hero, your brand video, and your carefully art-directed landing page matter less to the buying decision than your GTINs do. Presentation stops being the differentiator. Data quality becomes it.
The stores that win the next few years will be the ones whose product data is clean enough to be trusted, whose pages are readable without a browser, and whose checkout can accept a payment from software. None of that is exotic engineering. All of it is work that has to be scheduled, owned, and finished.
Most catalogues we look at fail at step two. The data has been accumulating in a CMS for five years, entered by four different people, with size in the title for some products and in a custom field for others. That is fixable. It is just not fixable in an afternoon.
