Agentic Commerce, Explained Without the Hype.

Agentic commerce describes any shopping transaction where software acting on a person's behalf performs the discovery, comparison, or checkout work that a human used to do manually. The agent is usually a large language model wrapped in tools: a retrieval layer that pulls product data, a ranking layer that weighs options against the shopper's stated goals, and a payment layer that can place the order when the shopper approves. ChatGPT with Shopping, Perplexity Shopping, Amazon Rufus, and Google Gemini's shopping flows are the current consumer-facing examples.

The consumer layer is dominated by OpenAI, Anthropic, Google DeepMind, Perplexity, and Amazon. The payment and checkout layer is being built by Stripe, Shopify, Visa, and Mastercard, each publishing their own agent-payment primitives through 2025 and 2026. The retrieval layer runs on Schema.org data, Google Merchant Center, and marketplace APIs. The protocol layer is still consolidating: Anthropic's MCP has the widest adoption for tool connections, while OpenAI's ACP and Google's Agent Payments Protocol (AP2) compete in the commerce-specific slice.

Pew Research and Adobe both published studies in 2025 showing double-digit percentages of shoppers already starting product research inside an AI assistant. Gartner forecasts that by 2028 a meaningful share of online retail will move through agent-mediated flows. The brands that show up in those flows will be the ones whose data is clean, structured, and verifiable today. Waiting until the behavior is mainstream means waiting until the rankings have already settled.

There is overlap but it is not the same. SEO optimizes for a ranked list of blue links that a human scans. Generative Engine Optimization, or GEO, optimizes for the moment an LLM summarizes, cites, or ranks your product inside an answer. SEO cares about keywords and backlinks. GEO cares about entity clarity, factual density, and how cleanly a model can extract a single quotable sentence from your page. A brand that invests only in SEO will lose ground inside AI answers; a brand that invests only in GEO will miss organic search. Both matter.

Start with Schema.org markup validated by Google's Rich Results Test. Add Product, Offer, Brand, AggregateRating, and GTIN properties to every listing. Push a clean feed to Google Merchant Center and keep it healthy. Make sure inventory and price on your site match the feed. Offer plain-text specifications in tables rather than images of specs. Expose a read API for inventory and price if your platform supports it. These are unglamorous data-plumbing tasks, and they are the work that actually moves retrievability.

For most small and mid-size brands, no. The protocols are still stabilizing and most commerce happens through the retrieval and feed layer, not through direct protocol connections. What you need today is a clean feed, valid schema, and a consistent brand entity. Once your platform (Shopify, BigCommerce, Amazon) natively supports a given agent protocol, enabling it is a configuration step, not a rebuild. Spend your 2026 budget on the data foundation, not on premature protocol engineering.

Ask the assistants directly: run queries in ChatGPT, Perplexity, Gemini, and Rufus for your category and see whether your brand surfaces. Track HTTP referrer strings from those domains in your server logs. Monitor Merchant Center diagnostics for feed errors. Watch branded search volume, which often rises when AI assistants start name-checking you. Add a one-question survey at checkout asking how the customer heard about you. None of these are perfect, but together they give you a defensible readiness signal.

Myth one: agents will replace search entirely. Reality: they are adding a layer on top of search, not removing it. Myth two: you need to hire an AI engineer to be ready. Reality: most of the work is structured data, feed quality, and content positioning, the same disciplines that serious e-commerce teams have always done, executed with higher rigor. Myth three: it is too early to invest. Reality: the cost of cleaning a catalog grows with its size, so brands that defer will pay more later. Myth four: agents only care about price. Reality: reviews, brand entity, and availability weigh at least as heavily in current rerankers.

For a mid-size catalog (500 to 5,000 SKUs) a focused team can reach a respectable baseline in 60 to 90 days. Schema coverage and feed hygiene are usually the first 30 days. Content rewriting and entity work are the next 30. Measurement instrumentation and iteration run in parallel. Larger catalogs take longer, mostly because of the content rewrite load, and benefit from templated approaches and careful use of LLM-assisted copy generation with human review.

A small brand with a clean Shopify catalog and decent existing content can often get to a defensible baseline with five-figure investment spread over a quarter, most of it in audit and rewrite labor. A larger multi-channel brand with legacy data and inconsistent attributes commonly spends in the low six figures in the first year to rebuild the feed infrastructure and content layer. The right benchmark is not a fixed dollar number, it is the cost of being invisible in a channel that will carry meaningful share of purchases.