Browsing is not dead, for now. I don't think it is going to be the default behaviour that defines how purchase decisions get made in the future, though. For decades, the act of shopping meant scanning shelves (online or in person), opening tabs or walking to new stores, comparing specs, reading reviews or asking in-store assistants and gradually narrowing options. What will likely change is who or what performs those tasks for us.
Something that has changed quite radically in the last 24 months, when you need an answer to a question today, for many people, our first instinct is of course not to find an encyclopaedia or even type a single keyword into a search box. If you are at work drafting an email, helping your child with GCSE maths, or debating something in a pub, the fastest path to resolution is to ask an AI system directly. The interface is conversational, so we therefore ask fully formed questions, full of spelling mistakes. And our expectation in return is immediate synthesis or a complete answer, not blue links that require further exploration.
Going back thirty years, problem solving meant books, magazines, or finding subject experts. Fifteen years ago, it meant search engines, blue links and reading content online. Now, it increasingly means delegation to AI. The surfaces we use to answer questions tend to be the same surfaces we use to evaluate and select the products we buy. That evolution is from physical shops/mail order magazines to websites and now AI platforms. The shift matters because it changes the mechanics of discovery quite a lot. The browsing requirement becomes more optional and we increasingly are able to delegate part of the workflow. Gartner has explicitly forecast a decline in search as users shift to AI chatbots and virtual agents.
Will Browsing Become a Fully Delegated Task?
Browsing, both in physical shopping and on the web, has always been about refinement of choices to enrich the selection process. It is the process of exploring a wide field of options, applying constraints, clicking browse nodes or ranking price low to high, and arriving at a decision that feels justified for the purchase to take place. Arguably it is cognitively expensive, in relative terms. Over the past few decades, consumer markets have expanded significantly. The ease of launching a brand has simplified and the number of choices in a lot of retail spaces has somewhat exploded. That means what used to be a small set of clearly differentiated options, which was easy to browse, has now become an overwhelming matrix of price tiers, features, claims and bundles.
My favourite comparison is ketchup in a modern supermarket. In a Walmart store there are up to 30 different options available. You are not just choosing ketchup. You are choosing brand, ingredient profile, sugar content, packaging format, sustainability claims and price per kilogram. Whereas in a discount store they normally offer one. I think this is why discount stores continue to grow, despite the price match offers from major grocery chains, people like to keep the cognitive load in check.
On the topic of low cognitive load, humans often, or nearly always, gravitate toward the path of least resistance. When choice overload reaches a certain threshold, tools that reduce decision friction inevitably get traction. This is consistent with classic research on choice overload showing that increasing choice can reduce purchasing and satisfaction in certain contexts. I think that AI agents represent the next logical step in this evolution, as they can actively perform the narrowing process on your behalf. And this is not a new concept, is it? Wealthy consumers have used personal shoppers or assistants to filter options and execute purchases since consumerism existed. The difference now, I think, is the scale and access being provided. Agentic shopping, the process of delegating some part of the shopping process to AI, makes it accessible to everyone who has internet access. This optionality is embedded directly into the interfaces people already use to ask questions and manage tasks in day-to-day life.
What an Agent Does Differently from a Chatbot
The distinction between chatbots and agents is operational and about taking action. A chatbot is designed primarily for interaction. It answers questions, clarifies information and simulates a human conversation. Its output is mostly just text and its only job is to provide an explanation or answer to a specific question.
An agent extends beyond explanation into actual execution. It is designed to take goals (buy something), interpret constraints (at a certain price, delivery timing) and complete multi-step actions (add to basket, add discount code, check out with payment). This "chat to act" framing aligns with how McKinsey has described agentic AI in the retail space, namely systems designed to execute tasks, not just generate responses.
For an even simpler differentiation, think of a chatbot as a knowledgeable adviser. Think of an agent as your adviser who also has the authority to act. The difference is subtle but there is a more significant consequence. I think that once systems can reliably move from recommendation to specific actions, browsing (as a process for selection, not enjoyment) becomes something that happens inside the agent's reasoning process, behind the scenes, rather than on a consumer's screen. We are already seeing mainstream payment players position for this eventuality, with Bloomberg reporting on Mastercard's initiative to let AI agents shop and pay on consumers' behalf.
So what does this all really mean? For brands, this means visibility is no longer just about being seen by a human scanning results across a retail platform or in Google Shopping results. It is about being legible to an AI system, or set of systems, that is making structured decisions under constraints that have been delegated to it by humans.
The Trust Problem
Delegating purchasing decisions, like any form of delegation, raises the question: can the system be trusted? Large language models are excellent at generating fluent answers that parrot human language but occasionally contain inaccuracies, which have been dubbed hallucinations. Humans also do the same, we call it making things up or "BS", but we are accustomed to verifying human claims through things like reputation, context and the requirement of accountability. In the case of retail, you typically have the option to return an item to a seller if they have misrepresented the product.
I believe this places pressure on the quality of the underlying data. Many brands historically treated product data as an operational necessity rather than a strategic asset. As such, product data accuracy across the web is riddled with mistakes and omissions, including the obvious of not always having a unique product identifier like a GTIN. In an AI-driven environment, identifiers like GTINs matter because they help systems recognise and distinguish products more reliably.
Agents optimise for reliable signals, kind of like humans, focusing on a path of least resistance. Why trawl multiple websites to understand the product "truth" if there is one verifiable source that provides it from another similar product? Put simply, if a catalogue is messy, incomplete, or ambiguous, the agent will either downgrade confidence or prefer alternatives with cleaner data. Data architecture becomes the basis for highly trusted signals and therefore product recommendations.
The Product Truth Layer: Complete and Accurate Attributes for the Win
At the core of agentic commerce is product attribute truth. It is quite amusing that LLMs are very good with unstructured data but clearly structured data makes life simpler and gives a better output. "Rubbish in, rubbish out" still holds true.
To expand on that, while modern AI systems can easily extract meaning from unstructured content, they still prefer the path of least resistance. A well-defined attribute table is far easier to reason over than scattered prose across multiple web pages. When an agent needs to answer a question like "will this charger work with my device?" or "which variant is caffeine free?", structured data provides immediate clarity.
For brands and retailers, I believe investing in this layer is one of the highest-leverage moves available today. It reduces ambiguity, improves consistency across channels and gives agentic systems a stable foundation for recommendation, as well as helping humans who also want those questions answered simply.
In practical, straight-talking terms, that means things like disciplined taxonomy management, rigorous compatibility mapping and clear variant logic. This is not glamorous work, which is why it gets ignored by most agencies and anyone looking to get promoted inside an organisation. Which is also why it presents an opportunity for the brands that do invest in it.
Content That Converts Agents, Not Just Humans
It is definitely tempting to assume that content optimised for agents must look radically different from content written for people, maybe like the 1s and 0s in green and black from The Matrix. In practice, the overlap is pretty substantial from what I have seen. Both humans and agents benefit from clear language and well-structured information that is easy to read and easy for an agent to "parse", meaning turn into structured information it can reliably interpret and use.
In this new world, brand narrative still matters a lot. The brand voice and storytelling will influence human perception and long-term loyalty, as it always has done. In the version of agent-mediated decision making, the system is going to prioritise verifiable facts over feelings. However, content that blends both the persuasive human framing with specific, structured detail is more likely to survive the filtering process for both paths to purchase.
How Brands Get Filtered Out
You might think that the most common failure mode, of not getting recommended by AI, is the accidental (or purposeful) inclusion of malicious misinformation. It is not that however, it is the basic inconsistency in brand data. Duplicate SKUs with conflicting attributes, outdated specifications, unclear variant naming, or missing compatibility data all introduce a layer of friction for the agent. As does a website which isn't easily readable by AI. It's not hard to understand why an agent faced with uncertainty will often default to safer alternatives, just like a human would too.
Preparing for a World of Delegated Commerce
Agentic shopping will not replace human browsing overnight, because humans won't change their habits that quickly. Hybrid behaviours will most likely propagate in the near term. People will still explore, compare and impulse buy themselves like they always did. The invention of the internet did not completely replace physical stores. The rise of delegated commerce will not completely erase human browsing. But I do believe that delegation will steadily absorb the more routine shopping decisions, especially where it is easy for the human to outline the constraints and the time-saving benefit is more than moderate.
What Can You Do Now to Prepare for the Near Future?
First, tighten the product data layer. Ensure attributes are complete and consistent across channels. Doing this will certainly not impact your existing business negatively and may in fact improve performance through traditional channels as well, but it comes down to time, resource and opportunity cost.
Second, I think all brands should think about aligning content with real user questions. That looks like writing for humans, but structuring it well for machines.
Third, experiment. AI is changing the world in many ways already and shopping behaviour will evolve too. We can't be 100% certain what works today will work tomorrow, so like those email titles and image tests, keep A/B testing everything going forward.
In Summary
Browsing is not ending. Humans will still do it, for fun. The more choreful versions will be internalised by AI. The process of comparison and refinement will still happen but increasingly behind the scenes, in systems designed to compress the complexity and time for humans. For brands, I think the implication is quite straightforward. If you want to be chosen, you must be understandable to the entity doing the choosing, whether that is a human or an agent. Creating clarity for both audiences is not just good practice, it is the competitive advantage.
