A customer never visited your store today. No one scrolled your homepage, read your About page, or compared your shipping policy against three other tabs. Instead, an AI agent did all of that in about four seconds, cross-referenced your return policy against a review site, and either added your product to a cart or moved on to a competitor who was easier to parse. You didn't lose the sale to a better product. You lost it because your site was unreadable to the thing doing the shopping.
That's the uncomfortable reality of agentic commerce in 2026. Over a third of people in OECD countries used generative AI tools last year, and during last Valentine's Day, 42% of shoppers used AI to research gifts before buying. Google, OpenAI, and Shopify have all shipped agentic commerce infrastructure this year, and the protocols behind it — OpenAI and Stripe's Agentic Commerce Protocol (ACP) and Google and Shopify's Universal Commerce Protocol (UCP) — are no longer pilot projects. They're live, and analysts expect them to matter a lot: McKinsey projects agentic AI will influence $3 to $5 trillion in global retail commerce by 2030, and Morgan Stanley expects nearly half of online shoppers to be using AI shopping agents by then, representing roughly a quarter of their spending.
Small retailers don't need to chase every headline about this shift, but they do need to understand the handful of things that determine whether an agent recommends them at all — and the new bookkeeping wrinkle that comes with letting a machine complete the transaction.
Why Your Store Might Be Invisible to AI Agents
Here's the part that catches most small business owners off guard: AI shopping agents don't read your site the way a human does. They don't appreciate your hero image, skim your brand story, or forgive a beautifully designed page that hides the actual specs in a carousel. An agent evaluates data feeds and executes against APIs. If your price, stock status, size options, and return policy aren't sitting in a structured, machine-readable format, the agent often can't act on them at all — not "won't recommend you," but literally cannot parse what you're selling.
This is a bigger deal than old-school SEO ever was, because there's no equivalent of ranking #4 on a search results page. An agent either has clean enough data to include you in its shortlist, or it silently drops you and moves to the next retailer whose feed it could actually read. One industry analysis found that sites following agentic SEO best practices saw 28% higher inclusion rates in AI-curated recommendation lists compared to sites that didn't.
The reassuring news: this is one of the more fixable problems in e-commerce. It's a data-hygiene project, not a redesign.
Five Things That Actually Move the Needle
1. Make your catalog machine-readable
This is the foundation, and skipping it makes everything else irrelevant. Implement structured data — JSON-LD using Schema.org's Product vocabulary — for every item in your catalog: price, availability, GTIN/MPN, material, dimensions, shipping details, and your merchant return policy (hasMerchantReturnPolicy is a real field agents check). Pair that with an AI-friendly sitemap and, if your platform supports it, a product catalog API or feed rather than relying on agents to scrape your storefront pages.
If you're on Shopify, note that "Agentic Storefronts" now syndicates your catalog to connected AI platforms — ChatGPT, Microsoft Copilot, Google's AI Mode, and Gemini — automatically for eligible merchants, with no app install and no extra fee. If you're on another platform, you can often sync your feed into Shopify's Catalog to get the same distribution without replatforming.
2. Write like a customer talks, not like a keyword list
Old SEO copy optimized for phrases like "running shoes men's size 10." Shoppers querying an AI agent ask, "Which trainers are best for marathon training if I have flat feet?" Your product pages and FAQ content should answer real, specific questions in plain language, because agents are increasingly synthesizing answers from that conversational content rather than matching keywords.
A practical exercise: pull up your last twenty customer service emails or chat transcripts. The actual questions people asked you — about fit, durability, compatibility, care instructions — are the exact content gaps an agent is trying to fill when it decides whether to recommend you.
3. Build credibility outside your own website
Agents don't just trust what you say about yourself — they cross-reference it. Roughly 86% of AI shopping citations trace back to brand-controlled sources, but "brand-controlled" includes more than your homepage: reviews, business listings, and media mentions all count, and agents weight consistency across all of them. A retailer with accurate, matching information on Google Business Profile, a handful of trade or local press mentions, and a healthy review volume on third-party sites builds the kind of external corroboration an agent looks for before it recommends a purchase.
4. Keep brand facts relentlessly consistent
Treat your pricing, stock levels, shipping windows, and policies the way you'd treat inventory: something that needs constant reconciliation, not a one-time setup. If your website says a product ships in 2 days but your Google listing says 5, or your stated return window doesn't match what a customer actually experiences, that inconsistency doesn't just annoy a human — it trains the AI system to treat your data as unreliable and quietly deprioritizes you next time. Nearly half of retail AI citations now originate from first-party websites specifically, which means your own site's accuracy carries more weight than ever.
5. Get ready for agents to actually complete the purchase, not just recommend it
Discovery is only half of agentic commerce. The newer layer is checkout: ACP handles the transaction handshake (structuring a cart, generating a payment token, completing checkout) and has been adopted by 25+ platforms including Salesforce, Squarespace, and Adobe Commerce. UCP goes further, covering the full journey from discovery through post-purchase support, and its most recent update added multi-item carts, live catalog queries, and loyalty program integration. If your storefront can't support one of these protocols yet, an agent may still recommend you in conversation but bounce the shopper to a competitor for the actual purchase because your checkout can't complete the transaction machine-to-machine.
The Bookkeeping Question Nobody's Asking Yet
Here's the part that gets skipped in most "how to win at agentic commerce" advice: when a machine completes the purchase instead of a person clicking through your checkout, your books need to keep up.
You're still the merchant of record. That doesn't change just because an AI agent initiated the transaction on a customer's behalf — you remain liable for fraud, chargebacks, and disputes exactly as you would for a purchase made by a human. And the fraud picture is genuinely different here: friendly fraud already drives roughly 75% of chargebacks and an estimated $132 billion in annual merchant losses, and security researchers warn that autonomous refund and return flows create new attack surfaces — bot-driven "return fraud" at a volume no human customer service team could generate. If your accounting can't tell you at a glance which sales came through an agent channel, you'll have a much harder time spotting an anomalous spike before it drains your cash reserves.
The fix isn't complicated, but it does require intention: tag agent-mediated sales distinctly in your books from the start, keep the transaction metadata (which protocol, which platform, timestamp, intent token if your processor provides one) alongside the sale rather than letting it disappear into a generic "online sales" bucket, and reconcile these channels at least as often as you reconcile your marketplace or POS sales. A comprehensive audit trail is exactly what makes chargeback disputes cheaper and faster to resolve — and it's a lot easier to build that trail from day one than to reconstruct it after a dispute lands.
This is where plain-text, version-controlled bookkeeping earns its keep. When every transaction is a line in a file you control — not buried in a black-box dashboard — adding a new tag for "agent-originated" sales, or writing a query that isolates them for a monthly review, is a small change rather than a system migration.
Getting Started Without Overhauling Your Whole Stack
You don't need to solve all five priorities in one sprint. A reasonable order for a small retailer:
- Audit your product data first. Pick your ten best-selling products and check whether their price, stock, and return policy are actually available as structured data anywhere an agent could read them. If not, that's your first fix.
- Rewrite your FAQ page around real customer questions, not keywords — this is copywriting work you can do this week without touching your tech stack.
- Check your listings for consistency across your own site, Google Business Profile, and any marketplace you sell on.
- Ask your platform (Shopify, BigCommerce, WooCommerce, etc.) what agentic commerce support already exists — many platforms are shipping this as an automatic feature, not something you have to build.
- Add a simple tag or category to distinguish agent-originated sales in whatever system you use to track revenue, even before volume from that channel is significant.
Keep Your Books Ready for Whoever — or Whatever — Is Buying
Agentic commerce doesn't change what good bookkeeping looks like; it just adds a new channel that needs the same discipline you'd apply to any other sales source. Beancount.io's plain-text accounting makes it straightforward to tag, track, and reconcile new transaction sources as they emerge, with a full audit trail you control instead of a locked dashboard. Get started for free and keep your books as adaptable as your storefront needs to be.