Executive Summary
Retail is no longer competing for clicks — it is competing for selection. Visibility is shifting from search pages to AI-driven decision layers, where products are not just discovered, but chosen. To remain discoverable, retailers must move beyond optimizing for human browsing and begin structuring their systems for machine interpretation — through high-quality product data, reliable commerce signals, and interoperable infrastructure. The companies that treat structured product data as a strategic asset rather than a hygiene task will preserve customer access when agents mediate the shopping journey.
Key takeaways: discovery will increasingly be mediated by AI systems; product data will become a competitive advantage; emerging protocols and identity layers will shape how systems interact; and Generative Engine Optimization (GEO) will expand the scope of traditional SEO into AI-driven environments.
Conversational AI is rapidly emerging as a new interface layer in retail — shifting interactions from browsing to intent-driven requests. Instead of navigating websites, customers increasingly express needs in natural language (e.g. “find a waterproof jacket under €200”), and AI systems interpret, compare, and recommend products across retailers. In this model, the interface is no longer a webpage — it is a decision layer that sits between the customer and every retailer, determining not just what is seen, but what is selected.
This evolution is already taking shape in platforms such as ChatGPT’s expanding app ecosystem, where AI acts as a gateway to services, tools, and transactions — reinforcing its role as a new interface layer between users and businesses.
Around 40–55% of consumers in key sectors now use AI-assisted search during product discovery, even if final transactions still occur elsewhere.
AI shopping agents are redefining product discovery in retail, shifting the shopping experience from traditional site search and browsing to AI-mediated commerce journeys. In this new model, agentic AI systems interpret intent, evaluate structured product data, and surface curated options based on relevance, trust signals, and real-time availability.
Retailers have long optimized for a predictable trade-off: invest in brand, content, and UX to attract and convert human traffic. That model is no longer sufficient on its own. AI agents introduce a new starting point — one where discovery happens before a user ever reaches your site. Customers no longer browse; they brief. And the agent executes.
Instead of visiting multiple websites, an AI agent compares options across retailers, filters by airline size constraints, weight, durability, and price, and presents 2–3 validated choices, without the user ever browsing category pages.
A recent shift makes this tangible. Retailers such as Walmart, Shopify, and Amazon are actively exploring how AI agents like Sparky in ChatGPT, can reshape product discovery and shopping journeys within conversational interfaces.
While these experiments highlight the potential to collapse discovery and interaction into a single interface, they also expose limitations — particularly around replicating core e-commerce logic such as cart building, comparison, and fulfillment flows.
As a result, many retailers are balancing AI-driven discovery with maintaining control over their own commerce environments. The implication is clear: retailers may no longer control the front door, but they still need to control the store behind it.
For CX leaders, this changes the game. Visibility is no longer just a marketing outcome; it is a product and integration capability. And legacy silos — brand, merchandising, digital — quickly become a liability when they are not connected to real-time, machine-readable commerce signals.
To be strategic, separate where intelligence lands. AI will reshape three fundamental layers - discovery, decision and transaction - and each layer requires different organizational responses.
Here, AI augments traditional keyword search with intent understanding. Agents interpret natural language, infer constraints, and map user intent to product candidates across retailers. The optimization problem is no longer “which page ranks on this query” but “which product record is the best match to a semantically rich intent vector”- in other words, which product best matches what the customer is actually looking for. This means it’s no longer just about having good landing pages. What really matters is having complete and well-structured product data: detailed product features, clear categories, and consistent information.
Instead of showing pages optimized for “running shoes,” AI agents surface products that match cushioning level, support type, price range, and user reviews related to joint impact.
For CX teams, this means catalog completeness, attribute depth, and consistent taxonomies matter more than landing page optimization alone.
This becomes the new merchandising logic. Agents evaluate products using structured signals — compatibility data, verified reviews, delivery reliability, warranty and return policies, price dynamics and contextual user preferences (e.g., sustainability priorities, size constraints). Products are not ranked based on visibility or promotion, but on how well they satisfy intent with reliable, interpretable data. In this context, merchandising increasingly resembles product data engineering: how brand value, trust, and differentiation are encoded into structured signals. AI agents are not influenced by promotional or persuasive language (e.g. ‘only 1 day left’ or other urgency-driven messaging). This highlights an important shift: from emotional persuasion to rational evaluation based on structured product data and user intent.
Behind the scenes, an AI agent does not “see” promotions; it evaluates a structured feature space. Each product is represented as a vector of attributes: battery life (hours), delivery latency (days), return rate (%), review sentiment score, warranty duration, and price.
The user query is also encoded into an intent vector (e.g., prioritizing portability, long battery life, fast delivery).
The agent then computes a relevance score, for example, using similarity metrics or ranking models that compare the user intent vector to each product vector.
Even if one laptop is heavily promoted, the system will rank higher the product whose attributes better align with the user’s intent, such as longer battery life, faster delivery, and stronger reliability signals.
AI systems are beginning to support transaction flows — such as assembling carts, monitoring prices, or triggering reorders — but full automation remains limited.
Recent industry developments show that completing purchases directly within AI interfaces is still challenging, with many experiences redirecting users back to retailer-owned environments to maintain control over checkout, fulfillment, and customer experience.
As a result, retailers must prepare for more agent-driven interactions by exposing reliable, secure interfaces (e.g., APIs and integrations) and designing flexible, permissioned transaction flows.
Seen together, these layers highlight why a single “AI” initiative is insufficient. Optimizing for one layer while neglecting the others creates an imbalance: a retailer may be discoverable but not trusted, or trusted but not easily transactable.
When machines choose, what counts as “visibility” changes. For humans, visibility is a mix of brand, UX, and persuasion. For agents, visibility is fidelity of representation, how clearly and reliably a product can be understood and evaluated. Product intelligence becomes the primary ranking signal.
Agents need rich, unambiguous metadata — dimensions, materials, compatibility matrices, usage constraints. Sparse or inconsistent attributes are not just inconvenient; they reduce the likelihood that a product will be considered at all.
Trust signals influence agent recommendations. Verified reviews, return policies, shipping reliability, and historical fulfillment performance all factor into an agent’s risk assessment. This shift is already reflected in consumer behavior: studies suggest that around 40–55% of consumers in key sectors now use AI-assisted search during product discovery, even if final transactions still occur elsewhere.
When multiple products satisfy the same intent, the outcome is no longer driven by visibility alone but by which option can be selected with the most confidence.
Contextual relevance increasingly outweighs brand awareness in agent-driven environments. A well-known brand can lose to a better-fitting SKU if the latter more precisely matches user intent and operational constraints. This shifts decades of marketing logic: brand equity still matters, but increasingly upstream — encoded within product data, policies, and fulfillment capabilities — not only downstream as page design or messaging.
This is not purely a marketing challenge, but a cross-functional one. It exposes long-standing silos: product teams managing catalogs, operations managing inventory and fulfillment, and marketing shaping messaging. If these functions do not align around a shared product-data foundation, organizations incur a “late information” tax, in which incomplete, inconsistent, or delayed signals reduce their likelihood of being selected by AI systems.
In practice, this means demand is increasingly routed toward retailers that provide clearer, more reliable, and more actionable machine-readable signals.
The companies that will lead are those that treat product data as a strategic asset — a moat you build over years, not a checklist to be outsourced.
Structured product knowledge enables more reliable interpretation by AI systems. A well-designed product graph that expresses compatibility, substitutions, variants, and use cases enables agents to match products to intent more accurately. This is the difference between a product being selected for “jogging shoe for marathon training” versus being surfaced simply because it contains the keyword “shoe.”
Reliable commerce signals increasingly outweigh surface-level messaging. AI systems rely on price accuracy, inventory availability, detailed specifications, and verified customer experience data to evaluate options. Creative presentation still matters — but it must be supported by consistent, machine-readable facts.
Proprietary product intelligence becomes a durable advantage. The richer your internal product model — the more provenance and real-world performance you can attach to SKU records — the more easily agents can select your catalog confidently.
This is not a fast follower problem: constructing accurate, interoperable product graphs and the processes that keep them fresh is hard and requires sustained, organization-specific effort. In practice, this becomes a data moat.
The companies that will lead are those that treat product data as a strategic asset — a moat you build over years, not a checklist to be outsourced.
Structured product knowledge enables more reliable interpretation by AI systems. A well-designed product graph that expresses compatibility, substitutions, variants, and use cases enables agents to match products to intent more accurately. This is the difference between a product being selected for “jogging shoe for marathon training” versus being surfaced simply because it contains the keyword “shoe.”
Reliable commerce signals increasingly outweigh surface-level messaging. AI systems rely on price accuracy, inventory availability, detailed specifications, and verified customer experience data to evaluate options. Creative presentation still matters — but it must be supported by consistent, machine-readable facts.
Proprietary product intelligence becomes a durable advantage. The richer your internal product model — the more provenance and real-world performance you can attach to SKU records — the more easily agents can select your catalog confidently.
This is not a fast follower problem: constructing accurate, interoperable product graphs and the processes that keep them fresh is hard and requires sustained, organization-specific effort. In practice, this becomes a data moat.
Agentic AI changes the participant model: retail interactions are no longer simply Human → Retailer.
Increasingly they are Human → Agent ↔ Retailer, with AI systems acting as intermediaries that influence discovery, evaluation, and, in some cases, transaction flows. That shift alters incentives.
Agents can reduce friction throughout the journey by monitoring price changes, recommending reorders, and assisting with purchase decisions within user-defined constraints. For retailers, this creates new revenue and data opportunities — but also new requirements. AI systems depend on reliable access to product, pricing, and policy data, as well as clear signals around eligibility, fulfillment, and loyalty.
Early adoption patterns reflect this shift. While only around 22% of users have completed a purchase directly within AI tools, nearly 50% report making a purchase after using AI for product research — reinforcing that AI is already reshaping discovery, even if transactions are still evolving.
Over time, agent-driven interactions are likely to increase transactional efficiency at the margin: fewer abandoned carts, more “zero-click” reorders, and a growing share of long-tail purchases initiated without a traditional browsing session.
However, this comes with trade-offs. As interface control shifts toward AI systems and platforms, retailers risk losing direct access to the customer relationship and may face margin pressure if access to demand becomes intermediated.
This is not only a product challenge, but it is also a platform strategy decision.
If SEO optimized pages for human attention, GEO optimizes product data for machine selection.
🔎 What is Generative Engine Optimization (GEO)? Expand for definition..GEO refers to the practice of structuring product data, content, and systems so they can be interpreted, trusted, and selected by AI shopping agents, not just discovered by human users. In AI-mediated commerce, visibility is no longer about ranking pages, but about being retrieved and chosen within AI-driven product discovery flows.
From keywords to product intelligence
Structured, machine-readable product data becomes a primary driver of relevance. In AI-driven discovery, selection depends on how well a product’s attributes match user intent, not just how well a page ranks for a query.
From crawling to structured access
AI systems still rely on web content, but increasingly benefit from structured and consistently organized data sources. Retailers already expose product, pricing, and inventory data through APIs and feeds, originally designed for internal systems and partners. The shift is now toward making this data more accessible, reliable, and usable in AI-driven contexts, particularly for real-time decision-making.
From ranking to selection
Visibility is no longer only about appearing in a list of results; it is about being selected by AI systems. Products that are better structured, more trustworthy, and easier to interpret are more likely to be chosen.
| Old (SEO) | → | New (GEO) |
| Keywords | → | Product intelligence |
| Crawling | → | Structured access (APIs) |
| Ranking | → | Selection |
In this context, structured product models become critical. Clear identifiers, consistent attributes, and well-defined relationships between products allow AI systems to interpret offerings with greater confidence and accuracy.
At the same time, traditional web crawling and scraping remain part of the ecosystem, but they are insufficient for reliable commerce. APIs and structured integrations are emerging as a more reliable complement to traditional crawling, particularly for real-time commerce data such as pricing, inventory, and availability. As recent industry developments show, incomplete or outdated product data leads to poor user experiences and failed transactions.
The shift is not simply toward APIs, but toward structured, real-time, and interoperable commerce systems. Emerging standards such as Google’s Universal Commerce Protocol and OpenAI’s Agentic Commerce Protocol signal a move toward more coordinated, system-level integration across discovery, decision, and transaction layers.
For retailers, this means rethinking how product, pricing, and inventory data are exposed, not just for internal systems or partners, but for AI-driven environments that require accuracy, consistency, and trust.
The practical answer is not to abandon human UX — it is to design dual-readable commerce systems.
Build conversational discovery experiences that degrade gracefully: natural-language interfaces for customers and structured backends for agents. Integrate with AI ecosystems and marketplaces while preserving owned channels via tokenized identity and loyalty primitives. Encode brand differentiation into machine signals: loyalty tiers, sustainability credentials, return policies, and service guarantees should all be expressible as attributes in your product graph.
Organisationally, this requires collapsing legacy silos. Product, tech, operations, and marketing must adopt shared KPIs: “agent discoverability”, “agent transaction success rate”, and “agent-attributed lifetime value.” Without governance and a cross-functional operating model, pilots will live in Babel — pilot purgatory — and the company will pay a structural disadvantage.
Designing for AI-mediated commerce requires more than optimizing websites.
It requires aligning product data, systems, and customer experience across the organization.
At ML6, we work with retail and ecommerce companies to design and deploy AI systems that improve how products are discovered, evaluated, and delivered — from structuring product data and enabling more reliable decision-making, to integrating AI into operational workflows across commerce, supply chain, and customer experience.
AI in Retail & Ecommerce at ML6 → Explore our industry expertise
The transition from human-centric discovery to agent-mediated commerce is no longer theoretical. It represents an architectural shift with real, near-term commercial implications. The companies best positioned for this shift will be those that treat product intelligence as strategic, design their systems — including APIs and data layers — to support AI-driven interactions, and reorganize to remove the legacy friction tax of siloed teams.
There is a latency to this change — technology moves in waves — but the competitive window is short. The next five years will determine who can be easily discovered and trusted by machines. Those who wait for “plug-and-play” solutions will find themselves at a structural disadvantage; those who invest in a rigorous, cross-functional foundation today will own access to demand tomorrow.
For leaders in Customer Experience and Digital Transformation, three actions can create immediate momentum:
The future of retail visibility depends less on banners and more on the quality of the signals you expose. Build them now — the future is being trained on what you publish today.
If this is something you’re navigating, feel free to reach out to our team.