Somewhere in your organisation, a product that outperforms on every metric that matters could vanish from the only ‘shelf’ that will matter in five years: agentic commerce.
Generative agents such as ChatGPT and Gemini are setting the future norms of shopping today, with ChatGPT’s 900 million weekly active users already making 50 million shopping-intent prompts every dayi. These agents will increasingly sit between the retailer and the customer and when asked by a customer to find the ‘best option’, they only return what they can confidently describe.
The uncomfortable truth most retail leadership teams haven't fully confronted yet, with agentic commerce, is that the winning product isn't the best product. It is the product that is best understood by the LLM.
You didn't lose the shelf. You lost the description.
For decades, retail competed for physical and digital space. And while space was finite on the shelf, products could still be found. Likewise, products could still be found on retailers’ catalogues on their websites or app, plus, Google has a page 2, page 3, and so on. A mediocre listing could still get browsed and still convert on price or packaging alone.
However, agents don't browse or have multiple pages. They reason and resolve. A shopper asks a question once and the agent returns somewhere between five and eight answers. There is no "let me just have a look". The competitive unit has shrunk from the shelf and catalogue to the sentence: the handful of machine-readable facts an LLM can retrieve, trust and stitch into a recommendation in the half-second it has to answer.
Most retailers and brands are still resourced to win shelves. Almost none are resourced to win sentences.
The real threat isn't Amazon - your own data debt is
Many Boards may be concerned about the wrong competitor. The biggest risk to win with generative agents is the quality, coverage and timeliness of dataii and not whether others will out-compete you.
Every retailer already has an "AI shadow" - the composite picture an LLM has assembled of your brand and products from everything it's read, retrieved and inferred about you. Nobody designed or approved it. Indeed, most executives have never even seen it. Yet it's already shaping which of your products get recommended and which remain invisible.
Try it now. Ask a leading agent which supermarket has the best own-label ready meals for a midweek family dinner, or which grocer to trust for a first-time low-sugar shop. You will get an answer, confident and specific, built entirely from public reviews, forum threads, press coverage and metadata assembled over years. Your merchandising team did not write it nor did your brand team approve it. And if the products you would want an agent to recommend are missing from that answer today, they are quietly being written out of the category for tomorrow.
Here's the part that catches most leadership teams off guard: that shadow isn't built from keywords. It's built from reasoning. An LLM does not index your product data, it interprets it. It infers the occasion, the trade-off, the substitute, the "what this is actually for" in the same way a good shop assistant would, by reading context and drawing conclusions. That means the SEO instinct to stuff titles and descriptions with search terms will not work for agentic commerce. A model isn't matching strings; it is asking whether your brand’s or product's story resolves the customer's problem with enough confidence to stake its own recommendation on it.
Most product data was written to be scanned, not reasoned about. Most product data states what a product is, rarely why someone would choose it, what it competes against or what it is bought alongside. LLMs need narrative and context as they draw inferences about a customer’s needs and mindset (think: intent) on the basis of their prompt. Agents are risk-averse by design. Faced with a thin, ambiguous story, or no story at all, they don't guess for the customer, they default to recommending the brand or product with the fullest narrative to reason with.
This is part of the reason that company Boards now need to get comfortable with building and sharing behavioural attributes derived from their data, to help agents understand their products and the customers that buy them, so those very retailers and brands become the de facto recommendation by agents.
This is a data issue in a marketing costume. And it is fixable in quarters, not years, which is exactly why it deserves Board attention now, before a competitor fixes it first.
You now have two customers to win
Here's what most loyalty strategies haven't caught up to: the agent and the human it is shopping for are separate audiences with separate criteria and you have to earn the loyalty of bothiii. Your customer still wants what they've always wanted: trust, value, and relevance. The agent only requires unambiguous, structured, verifiable facts and descriptions it can act on.
You can have a customer who loves your brand and still lose the sale because the agent making the recommendation on behalf of the customer is not swayed by affinity, only by evidence and reasoning from the information available to it.
Conversely, you can win the agent's recommendation and still lose the customer if what arrives does not match what was promised, because the agent staked its own credibility on you and a broken promise damages both relationships at once. Loyalty has split into two separate contracts, running in parallel and most organisations are only resourced to manage one of them.
Agents never forget
Agents are now using memory which means every recommendation leads to the formation of machine habitiv. Once an agent has recommended, selected and had a product validated by a customer, that preference gets reinforced into future decisions. The agent starts building its own loyalty - a second, separate loyalty, running alongside the relationship between the customer and the brand or retailer. Win the first few recommendations in a category and you compound. Lose them and you're not just invisible today; you are structurally locked out of tomorrow, fighting to unseat a default recommendation to a competitor.
That is a moat no Board has modelled yet. It will be far more decisive and far cheaper to build than another loyalty app or another loyalty percentage point. To succeed, brands and retailers need to come to terms with the fact that they need to use their data openly to influence agents to win the recommendation and earn the loyalty of the LLM. Until this point, brands and retailers were wary of sharing their data – this has to change.
The question for your next leadership meeting
Boards have moved beyond asking "what is our agentic commerce strategy". That is too broad and the market has moved on. The sharper question is: if an agent were asked right now to find the best version of our top ten products, would it find us, would it get us right and where would we rank?
A CEO may not know who is responsible for answering the question and owning the outcomes. Solving the ownership gap, not the technology or data, is the first step to creating a competitive edge.
The retailers and brands that treat discoverability in agentic commerce as a structured, funded, board-level programme will spend the next decade being the default answer. Everyone else will spend it trying to buy their way back into a conversation their competitors already won.
Where to start now
Pick your top ten SKUs. Prompt ChatGPT, Gemini and Perplexity the way a customer would and record the following
1) whether you appear and where you rank;
2) how you are described and;
3) whether the agent gets you right.
The gap between what you sell and what the agent sees is your AI shadow. Closing this gap is the first move worth making, before the machine habits harden and the default answer becomes someone else’ brand or product.
