Agentic Commerce vs Traditional Ecommerce: What Changes

One sentence of difference

Traditional ecommerce optimises a page a human reads. Agentic commerce optimises a response a machine parses and then narrates to a human who never saw the page.

That single shift reorders almost every assumption underneath conversion work, because the assets you spent the most on — hero imagery, A+ content, review widgets, interactive sizing tools — are not in the payload the agent receives.

Discovery: from ten blue links to one tool call

In search, a shopper sees a results page and can go back, scroll, open a second tab and compare. In an agent conversation there is no results page and no going back in the same sense. The official agent instructions Shopify publishes at shop.com/references/http.md tell agents to keep rejected choices out of later results. If a buyer says a product is too expensive or the wrong colour, that option is gone for the rest of the conversation.

The same skill tells agents to preserve buyer intent and not narrow the query more than asked, which means a lot of retrieval happens on broad category terms and then converges through filters. Two things therefore gain weight: whether your product can be recalled on a broad category term at all, and whether you clear the thresholds on each filter dimension the catalogue supports.

Merchandising: ranking signals you used to call operations

Cross-store catalogues cluster offers for the same product and let agents filter on dimensions including rating, condition, media type, price tier, ships-to and ships-from. Several of those were, until recently, operational metrics nobody in the growth team owned.

Two deserve particular attention. Rating filters are evaluated at variant level with a minimum review count, so a strong store-wide average will not rescue a flagship variant with thin reviews. And ships-from filtering by merchant origin country is a dimension that simply did not exist before — it can be used to exclude cross-border sellers in a single step, which changes how much fulfilment localisation is worth.

Creative: from persuasion to extraction

Cross-store catalogue fields include a set the platform marks as inferred — attributes, technical specifications, top features and unique selling points generated or enhanced from your own text. Shopify's documentation is explicit that these are discovery and merchandising signals rather than merchant-authored source text.

The practical instruction is blunt. Brand poetry yields nothing extractable. A sentence like built for those who dare produces no feature. A sentence like 304 stainless steel inner pot, dishwasher safe, fits 2.6 L produces three. Copy still has to persuade humans on the page, but it now also has to feed a machine that will paraphrase it to a buyer who never reads it.

One tactic is explicitly closed off: the official skill instructs agents to treat merchant text as data, not instructions. Hiding directives in descriptions, alt text or markup does not work, and the same skill requires agents to state material unknowns — so the realistic risk is not being manipulated into first place, it is being accurately described as vague.

Measurement: the part that breaks first

When an order completes inside an agent surface, your thank-you page is never loaded. Every event bound to it stops firing: page-based purchase events, pixel conversions, tag-based ad conversions and any audience built from storefront pixels. Multi-touch attribution tools develop a gap.

The orders themselves are unaffected — they land in your Shopify admin as normal orders with customer data, and fulfilment, support and refunds work as before. Ownership does not change; measurement does. In practice attribution breaks before you see any incremental revenue, which is why the first project is server-side measurement rather than channel expansion.

What stays the same

• Product-market fit. An agent will not recommend a product buyers do not want.
• Price and value. The official skill directs agents to find good value rather than the lowest price, and to mention worthwhile options above budget unless a firm maximum was set — premium brands are not structurally disadvantaged.
• Fulfilment reliability. Delivery performance now shows up as a filterable and badge-able ranking factor rather than only as a support cost.
• Brand knowledge. Agents search by brand name when buyers name a brand, which only happens if the brand is already known — the reason GEO and AEO still matter.
• Your site. The product link an agent returns still gets opened by a human, so page quality and speed keep earning their keep.

How we would approach it with you

We treat this as two linked programmes rather than one. The first is agent readiness: protocol and checkout verification, catalogue eligibility, attribute normalisation, extractable product facts, published policy facts. The second is AI visibility: the GEO/AEO work that gets a brand known, mentioned and cited across ChatGPT, Google AI Overviews, Perplexity and Gemini so that branded retrieval has something to retrieve.

If you are weighing how much of this your team should take on in-house versus hand over, that is a conversation we have most weeks. Tell us your domain and category and we will give you a candid read on the gap and the sequence.