Agentic commerce: how to be discoverable and win the agent’s attention

For decades, the retail mantra was ‘location, location, location’. In the digital age, that shifted to ‘page one of Google’. We are entering a third age: the era of agentic commerce.

Imagine a customer asking an AI assistant – be it Gemini, Copilot, or Claude – to find ‘the perfect 40th birthday gift for a whisky lover who values sustainability’. The agent doesn't return a list of ten blue links for the user to browse. Instead, it synthesises an answer, presenting a small set of highly confident recommendations, and with a single ‘best’ option pre-selected, complete with price, retailer and a ready-to-act purchase path.

In this world, being ‘highly ranked’ isn't enough. You are either the recommendation, or you are invisible. Discoverability – the science of ensuring AI models choose you as the fulfilment partner – is the next great frontier for retail survival and growth. The research supports this; according to eMarketer, between 80% to 95% of products discovered through AI assistants like Gemini or ChatGPT will still be purchased through a retailer’s site directly.

The anatomy of an AI recommendation

To To win the battle for discoverability, it needs to be understood that AI-driven recommendation systems – often powered by Large Language Models (LLMs) are not traditional search engines. They don't just crawl; they synthesise an ‘AI Shadow’ based on the information available to them. While real-world systems are complex and evolving, it is useful to think about AI-driven recommendations as operating across four interacting layers:

1. The foundation

This is the model’s ‘world view’. If your brand is frequently mentioned in premium gift guides, press releases, and high-authority articles, the LLM develops a ‘prior’ bias towards you. However, this layer is fundamentally descriptive, not behavioural. It reflects what is said about products (a narrative), not how they are actually chosen or what they are. 2. The pulse (Retrieval-Augmented Generation - RAG)

At the time the query is made, the model retrieves fresh signals. These could be product feeds, availability, pricing and structured attributes. This layer determines whether a product is eligible to be recommended. Most feeds today describe products in isolation. They lack context on:

• What need-state the product fulfils
• What it is typically bought with
• What it is substituted against

This creates a gap between product data and the context of the decision. 3. The reasoning layer

This is where the model interprets user intent. When a user asks for ‘a gift for a scotch whisky lover who values sustainability’. The model must infer:

• The underlying need-state (gift-giving, milestone, premium but ethical)
• The trade-offs (heritage vs sustainability, price vs uniqueness)
• The choice set (which products are credible substitutes or complements)

Today, this reasoning is largely inferred from patterns in text: how words, products and concepts co-occur in language, rather than from real behavioural data. This is the critical gap. Without grounded signals, models can:

• Over-index well-described brands
• Misinterpret substitution
• Miss context-specific relevance (what people buy for this exact occasion)

4. The incentive − the conversion layer

Companies like Google and Microsoft haven't abandoned their business models. Sponsored listings, affiliate relationships, and preferred merchant programmes still whisper in the AI’s ear, shaping which retailer gets the final "nod".

Winning in agentic commerce requires more than visibility. It requires alignment across all four layers: from pre-brand selection, to real-time eligibility, to behavioural relevance, to commercial viability.

From SEO to GEO

Traditional Search Engine Optimisation (SEO) was about keywords. Generative Engine Optimisation (GEO) is about context and confidence.

From a financial perspective, the stakes are binary. In traditional search, a second place result might still garner a 17[1] percent click-through rate. In agentic commerce, the dynamic becomes increasingly winner-takes-most. If an agent manages the transaction, only a small set of options capture the majority of value. Investing in discoverability today is about lowering your long-term Customer Acquisition Cost (CAC). By becoming the default recommendation for specific categories, you aim to bypass the expensive, repetitive bidding wars of legacy PPC (Pay-Per-Click) and enter the realm of organic, AI-driven loyalty.

It is important to recognise that these systems are probabilistic and platform-controlled. Retailers are not directly optimising the model, but improving their likelihood of being selected within it.