A field guide to agentic commerce, the new content supply chain, and why structured meaning beats prettier pages.
Short answer
When AI becomes the customer, the first round of product discovery moves from the human shopper to a machine. The agent reads the request, checks price, stock, delivery, reviews and product content, then builds the shortlist. The human still owns the need. The agent increasingly owns what reaches them.
AI systems become the first filter in product discovery.
Definition: agentic commerce
Agentic commerce is commerce where AI agents research, compare, shortlist or buy products on behalf of customers. It changes the role of ecommerce content because the first reader may be a machine interpreting the product before a human ever sees it.
In simple terms: your product page used to explain the product to the shopper. In agentic commerce, your content may need to explain the product to an AI agent first.
What it means
For two decades, the brand site sat at the centre of the commerce model. Brands controlled the page, the layout, the product story and the call to action. That model is now bending.
Search, social, email and paid all funnelled people back to a surface the brand owned. The product now shows up in marketplaces, retail media networks, social feeds, comparison engines, search snippets and answer engines. The customer may never see the carefully designed product page. More importantly, the customer may not be the one doing the first round of evaluation.
Discovery did not just hop to a new channel. The decision-maker changed. A model now sits between the need and the brand.
What it does not mean
Humans do not leave commerce. They still bring the budget, the taste, the deadline and the half-remembered brand name.
Decisions are not fully automated. The agent narrows. The human chooses.
SEO is not finished. SEO alone is no longer enough when the reader is a model.
Why this matters now
AI search, agent-led shopping and answer engines all consume content the same way. They read it. They extract from it. They form a view. Then they hand a summary back to the customer.
The product is still in the result. The brand version of the product is not. Most teams react as if it is a formatting issue. More copy. More variants. More feeds. Useful. It also misses the point.
AI does not display your content. It forms an opinion about it.
The issue is no longer where the content appears. It is who, or what, interprets it before the customer ever sees it.
What changed in the old model
The old test was simple. Did the page convert.
The new test starts a step earlier:
Can the product be found by an agent.
Can it be understood without the page around it.
Can it be compared without losing the difference.
Can it be trusted by a system that does not care about brand guidelines.
Can it be selected for the right reason, not the loudest one.
If the answers are weak, the prettiest PDP in the category is sitting in the wrong part of the funnel.
Structured data is necessary. It is not enough.
Machine-readable data gets you considered. Meaning-rich content helps you get chosen.
Agents need facts they can verify. Price, availability, size, delivery, returns, reviews, attributes. They also need context they can interpret. Who the product is for. Which use case it fits. Which trade-offs matter. Why one option suits a specific need.
Operational example: a sofa for a small flat with two retrievers
The product data might say:
220cm wide, 95cm deep
pet-friendly boucle in oat
pocket-sprung seats
FSC hardwood frame
removable, washable covers
£1,299, 4-week lead time
Useful. Now ask the agent for a sofa for a north-facing London flat shared with two muddy golden retrievers, up two flights of narrow stairs. Attributes alone fall short. The agent needs to reason about hair release on light fabric, fade in low daylight, seat firmness ten years in, the diagonal of the doorway on the landing, working-from-home posture versus Friday-night sprawl, and the trade-off between deep lounging and an upright sit.
Some of that lives in data. Most of it lives in content. Buying guides. Fit notes. Comparison copy. Reviews. Care guidance. Use-case content. Alt text. Video. FAQs. Support content.
Definition: the semantic layer in commerce
The semantic layer in commerce is the meaning-rich content and metadata that helps AI systems understand product fit, use cases, trade-offs and context. It sits alongside structured product data, not instead of it.
In agent-led discovery, the semantic layer stops being garnish. It becomes commercial infrastructure.
The next content problem is not volume. It is interpretability.
Most brands still treat AI as a production tool. Generate more copy. Generate more variants. Generate more posts. The team can now produce more content than anyone has time to review. What looked like a production problem quickly becomes a governance problem.
The harder question is whether a system acting on behalf of a customer can read the content correctly:
Can the agent tell who the product is for.
Can it explain the use case.
Can it compare two products without flattening the difference.
Can it tell the trade-off between technically suitable and commercially right.
Can it separate “pet-friendly“ from “survives one labrador“ from “still presentable after three winters with two retrievers“.
A lot of ecommerce content fails this test. It was written for a human standing on a page surrounded by images, filters, reviews and merchandising. Pulled apart by a model, the meaning leaks out.
How this compares with adjacent terms
SEO and AEO. SEO optimises web pages so a human types a query, scans a list of blue links and clicks through. AEO, answer engine optimisation, optimises content so an AI system such as ChatGPT, Perplexity, Google AI Overviews or Gemini can extract a defensible answer, cite the brand inside the response, and shape what reaches the customer when no link is ever clicked.
Traditional CMS, headless CMS and agentic CMS. A traditional CMS welds content to a page template. A headless CMS separates content from the front end and serves it as structured data through an API. An agentic CMS goes further. It models content as semantic objects, exposes them through AI-ready APIs and orchestrates assembly in real time so any channel, including an agent, gets the right meaning at the right moment.
Headless commerce and agent-ready commerce. Headless commerce decouples the storefront from the commerce engine so brands can deliver to any channel. Agent-ready commerce adds two things on top. Structured semantic content the agent can interpret. Live data the agent can act on.
Personalisation and agent commerce. Personalisation adapts a page for a known segment using rules or models. Agent commerce composes a recommendation for a one-off prompt the brand has never seen before, often with no prior signal at all.
DAM, PIM and the semantic layer. A DAM stores assets and asset metadata. A PIM stores product attributes and inventory data. The semantic layer is the meaning-rich content that connects them so an agent can reason about fit, use case and trade-offs, not just list specs.
Why this matters in ecommerce content operations
If AI is the customer, the job of content shifts. It no longer just sells on the page. The brand is now feeding evidence to a system that summarises, compares and recommends for humans.
That changes the operating model:
Content has to carry meaning outside the page.
It has to survive being summarised.
It has to be specific enough to be interpreted, not generic enough to be ignored.
It has to connect to live data without losing context.
It has to explain what the product is and why it fits a particular need.
Live data carries the facts that change. Content carries the meaning that does not. The agent needs both.
The practical work, in three moves
Find the questions customers are asking before they reach the brand.
Check whether AI systems are answering those questions, with which sources, which competitors and which claims.
Close the gaps with comparison, use-case and structured product content connected to live price, stock, delivery and policy data.
Where Amplience fits
Amplience is built for agent-led commerce.
Amplience brings together three connected product areas: CMS, DAM and Workforce.
Amplience CMS models content as structured, AI-ready semantic objects. Amplience DAM stores and serves assets with the metadata machines need to interpret them. Amplience Workforce is the orchestration and operations layer that runs the content supply chain across production, governance and delivery.
Live commerce data sits outside those product areas. It comes from the brand’s systems of record, including the commerce engine, PIM, ERP and inventory systems. That is where live price, stock, promotion, delivery and policy data already sit.
Meaning comes from Amplience. Live truth comes from the brand’s systems of record. Agents need both, and the join is what lets them find, understand, compare and recommend products correctly.
From idea to everywhere. Including the answer the agent gives the customer.
Related reading and viewing:
Original LinkedIn article by John Williams, co-CEO, Amplience: What happens when AI becomes the customer.
Companion video on the content explosion that follows, when content stops being scarce: What happens when AI becomes the customer (video).
FAQ
What does “AI becomes the customer“ mean?
It means an AI system, not the human shopper, performs the first round of product discovery, comparison and shortlisting. The human still chooses. The agent decides what reaches them.
Does AI replace the human customer?
No. The human still has the need, the budget and the final say. The agent narrows the options before the human sees them.
What is the semantic layer in commerce?
The semantic layer in commerce is the meaning-rich content and metadata that helps AI systems understand product fit, use cases, trade-offs and context. It sits alongside structured product data, not instead of it. Buying guides, fit notes, comparison copy, reviews, FAQs and use-case content all sit here.
How is AEO different from SEO?
SEO optimises for people finding pages through search results. AEO optimises for AI systems summarising and recommending on behalf of people.
What should brands do first?
Three steps. Find the questions customers ask before they reach the brand. Check whether AI systems are answering them, and with which sources. Close the gaps with comparison, use-case and structured product content connected to live data.
Why does generic brand language fail this test?
A line such as “the sofa that fits any home“ gives an agent nothing specific to interpret. Specificity wins. Trade-offs win. Vague brand language loses.
What is agentic commerce?
Commerce in which AI agents act on behalf of customers to research, compare and shortlist products, sometimes completing the transaction. Brands need machine-readable data and meaning-rich content to be selected.
Where does Amplience fit?
Amplience provides three connected products for agent-led commerce. Amplience CMS for structured, AI-ready content. Amplience DAM for assets and the metadata machines need to interpret them. Amplience Workforce as the agentic orchestration and operations layer that runs the content supply chain end to end. Amplience integrates with the brand’s own commerce engine, PIM, ERP and inventory systems for live price, stock and policy. Full detail in the section above.