What Happens to Advertising When AI Starts Choosing What We Buy?
For most of advertising history, the basic assumption has been that the person receiving the advertising is also the person making the decision. A brand creates an advertisement, a human sees it, something about the message influences that person's perception, and eventually the person may decide to buy. Every major development in me…
That does not mean millions of people are about to hand their credit cards to autonomous robots and stop thinking about what they buy. Agentic commerce remains early, particularly when it comes to allowing AI to complete transactions without human approval. But the earlier parts of the buying process are already changing quickly because people are using ChatGPT, Gemini, Perplexity, retailer assistants, and other AI systems to research products, compare choices, narrow shortlists, understand specifications, plan purchases, and decide which brands deserve consideration. McKinsey reported in 2026 that 68 percent of surveyed U.S. consumers had used at least one AI tool during the previous three months, while 19 percent of AI users said they had used it specifically to discover or decide between brands, products, or services. That is already enough to matter because marketing does not begin at checkout.
The shift becomes much more interesting when AI moves from adviser to agent. An assistant that tells you which laptop is best for your needs is influential, but an agent that can compare current prices, check stock across merchants, apply your stated preferences, put the best option into a basket, and eventually complete the transaction is doing something much closer to shopping on your behalf. McKinsey described 2026 as the year AI agents stopped being merely experimental and became part of everyday shopping behavior, particularly around assembling options, resolving trade-offs, and moving consumers closer to action. Deloitte now defines agentic commerce as an environment in which AI agents can search, decide, and transact across channels for the consumer. The customer is still the customer, but another decision-maker is beginning to sit between the customer and the brand.
For marketers, that creates a question the industry has never really had to answer at scale. What happens to advertising when the audience evaluating the advertisement is partly a machine? Human beings respond to humor, aspiration, fear, identity, celebrity, aesthetics, status, nostalgia, social proof, storytelling, and hundreds of other psychological cues that advertising has spent generations learning to use. An AI agent may care much more about whether the product is available, compatible, competitively priced, well reviewed, deliverable on time, easy to return, appropriate for the user's stated constraints, and supported by evidence it can verify. Advertising may therefore need to work simultaneously on two very different audiences: the human who wants to feel something and the machine that wants to understand something.
We Are Moving From Search to Delegation
Search engines changed shopping because they gave consumers enormous control over product discovery. Instead of relying solely on whatever brands retailers stocked or whatever advertising happened to reach them, people could actively search for almost anything and compare multiple sellers within minutes. Ecommerce expanded that choice further, while marketplaces created enormous catalogs that allowed customers to move between thousands of options without leaving the same website. The problem is that choice eventually became work. More products, more reviews, more comparison pages, more sponsored results, and more tabs created an environment in which consumers gained access to almost everything while increasingly needing help deciding between all of it.
AI shopping assistants are attractive because they promise to absorb some of that cognitive workload. Instead of opening fifteen browser tabs comparing noise-canceling headphones, someone can describe how much they want to spend, where they plan to use them, what phone they own, whether microphone quality matters, and which features they do not care about. The assistant can reduce hundreds of products to a small shortlist and explain the trade-offs. The user can then refine the request conversationally rather than restarting the research process every time a new preference appears. That is not simply another search interface because the AI is increasingly participating in the reasoning itself.
ChatGPT's shopping experience already works in this direction. When someone expresses shopping intent, ChatGPT can surface product options with images, prices, descriptions, reviews, merchant choices, and in some cases Instant Checkout so the customer can complete a purchase without leaving ChatGPT. Product selection is separate from advertising, according to OpenAI, and considers factors such as the user's request, conversational context, availability, price, reviews, and other relevant product information. Merchants can also provide product feeds so their inventory, pricing, availability, and seller information are represented more accurately. The shopping assistant is therefore beginning to look less like a search box and more like an intermediary capable of understanding the shopper and the catalog simultaneously.
Google is moving in the same direction from a very different starting point. In May 2026, Google introduced Universal Cart as part of a broader agentic commerce strategy built around its Shopping Graph, which contains tens of billions of product listings. The vision is that Google can help people discover products, track possibilities, assemble choices across merchants, and increasingly manage more of the shopping process from one intelligent environment. Google has also been developing the Universal Commerce Protocol as a common language that lets agents and businesses exchange the structured information needed to conduct commerce. The traditional search engine is therefore evolving toward a system capable of helping the user act on search rather than merely producing results.
ChatGPT Is Already Becoming a Shopping Layer
OpenAI's strategy makes the direction particularly visible because ChatGPT combines conversation, product discovery, merchant data, checkout, and advertising inside the same environment while keeping those systems formally separate. In March 2026, OpenAI expanded the Agentic Commerce Protocol to support richer product discovery, allowing merchants to provide more accurate catalog information and enabling ChatGPT to compare products visually and conversationally. Major retailers including Target, Sephora, Nordstrom, Lowe's, Best Buy, The Home Depot, and Wayfair were already integrated into the discovery system, while Shopify's catalog made millions of Shopify merchants available through the shopping layer. The significance is not simply that ChatGPT can display products. It is that product discovery is becoming native to the conversation rather than requiring the user to leave and restart the journey somewhere else.
That changes where brands compete for visibility. In a conventional ecommerce environment, marketers compete for search rankings, paid placements, category pages, retail-media inventory, reviews, and attention inside product listings. In an AI-mediated shopping environment, the first competition may be whether the agent considers the product worthy of mentioning at all. A brand can have an excellent ecommerce site, expensive search campaigns, beautiful packaging, millions of followers, and enormous awareness, yet still lose visibility if the AI system cannot confidently determine that the product meets the user's requirements. Discovery therefore starts to depend on a new layer of machine interpretation before the traditional landing page is ever reached.
OpenAI's current shopping documentation offers a glimpse into the criteria that can matter. ChatGPT may consider structured product metadata, prices, availability, product descriptions, reviews, merchant quality, whether a seller is the manufacturer or primary seller, and the specific constraints the user has expressed. The user can also refine the conversation by saying that price matters less than durability, or that a particular style, feature, or brand attribute should be excluded. Those instructions can change which products appear. The agent is effectively creating a personalized shelf every time the customer asks.
This is why the future ecommerce battle may not be about earning the top position on one universal digital shelf. There may be no universal shelf anymore. Two people asking for the same product category may receive completely different recommendations because their budgets, preferences, previous conversations, timing, location, delivery requirements, and stated constraints differ. The brand's challenge becomes getting onto millions of dynamically generated shelves created for individuals rather than winning one static ranking. Personalization moves from rearranging products on a webpage to constructing the consideration set itself.
Welcome to the Algorithmic Shelf
Deloitte uses the phrase algorithmic shelf to describe this emerging environment, and it captures the change remarkably well. Traditional retail shelves are shaped by assortment decisions, physical placement, trade negotiations, shopper behavior, and brand power. Ecommerce shelves added search algorithms, sponsored listings, reviews, product-page quality, and recommendation systems. The algorithmic shelf goes one step further because an AI agent may not show the consumer a shelf at all. It may inspect many options privately, eliminate most of them, and present only the few products it believes fit the customer's instructions.
That turns product consideration into an increasingly invisible competition. A person may never consciously reject your brand because your product may never appear in the shortlist they receive. The AI agent may determine that another product offers better compatibility, more complete information, stronger reviews, a clearer warranty, faster delivery, or a better match against the customer's stated preferences. From the consumer's perspective, the missing brand simply never entered the conversation. From the marketer's perspective, this is more concerning than losing a click because the company may not even know that an evaluation occurred.
The concept has an obvious parallel with SEO. For years, companies optimized websites so search engines could crawl, interpret, rank, and present their information. The rise of answer engines and generative AI extended that work into AEO because brands increasingly wanted to become sources that AI systems could cite and recommend. Agentic commerce takes the same logic much closer to the transaction. Product information must become understandable not simply so a search engine can rank a page but so an automated buyer can evaluate whether the actual product belongs in a customer's basket.
This is where marketers need to become much more interested in structured product data than many traditionally have been. Product identifiers, dimensions, compatibility, materials, pricing, stock status, delivery options, return rules, reviews, certifications, sustainability claims, images, technical specifications, and other attributes can become machine-readable evidence supporting a recommendation. OpenAI's product-feed documentation explicitly says that feed quality improves discovery relevance and recommends rich information including reviews, media, performance signals, pricing, inventory, and fulfillment options. The dullest-looking spreadsheet in the ecommerce department may eventually influence brand visibility as much as the glossy campaign produced by the advertising agency.
Advertising to Machines Is Not Really About Making Ads for Robots
The phrase "advertising to AI agents" makes it sound as though marketers will start designing banners with slogans intended for machines. That is probably the wrong mental model. An AI agent does not need to be emotionally persuaded by a television commercial or notice a beautifully designed billboard. What matters is the information environment the agent uses when deciding what deserves consideration. Advertising to the machine therefore becomes partly about ensuring that the brand's claims are visible, verifiable, current, structured, and supported by enough external evidence that the agent considers them credible.
Deloitte argues that brands competing on the algorithmic shelf need to translate brand equity into machine-readable proof points. That phrase deserves attention because traditional branding frequently operates through associations that are difficult to quantify. A company can spend decades building a reputation for durability, quality, innovation, or customer service. An AI agent may still ask what evidence supports those associations when comparing products against a user's stated criteria. The marketing claim "built to last" becomes stronger when the system can also see warranty duration, independent reviews, material specifications, repairability information, return rates, customer satisfaction, and long-term performance evidence.
This does not mean brand storytelling becomes irrelevant. The human consumer still cares about meaning, design, identity, aesthetics, aspiration, reputation, and trust. But storytelling may increasingly need a parallel layer of proof underneath it. The human sees the story while the machine evaluates whether the factual structure surrounding that story supports the claim. The strongest brands may be those where the emotional promise and the observable evidence reinforce one another.
That could ultimately make branding more accountable rather than less important. If a company has spent years claiming superior customer service while customers consistently complain about support, an agent summarizing reviews may expose the gap much faster than a traditional advertisement can conceal it. If a hotel claims to be family friendly but reviews repeatedly mention that children have little to do, AI can synthesize those patterns. If a product claims premium quality but competitors provide longer warranties and better reliability, an agent may recognize the discrepancy. Agentic shopping gives marketers another reason to align the brand promise with the actual customer experience.
The Agent May Become the Most Important Influencer
Influencer marketing became enormous because people often trust recommendations from other people more than they trust corporate advertising. AI assistants are now creating a new kind of influence because the user may see the recommendation as personalized, analytical, and less commercially motivated than a traditional advertisement. BCG reported in July 2026 that among consumers who use AI tools during product research, 85 percent ranked them among their five most influential touchpoints in the purchase decision. That puts generative AI alongside social media as a major source of influence and well ahead of many traditional media channels for those users.
That does not mean consumers blindly trust AI. The distinction between recommendation and transaction remains enormous. BCG's October 2026 research found that more than 60 percent of surveyed consumers trusted information from AI shopping agents, but only 4 percent were ready to let an agent complete a purchase entirely on its own. The battleground, at least for now, is therefore much more about recommendation than autonomous checkout. AI may choose the shortlist while the human still chooses the winner.
This distinction is strategically important for advertisers. Marketers do not need to wait until autonomous purchasing becomes normal before adapting because the agent can already influence which brands receive consideration. A recommendation does not need permission to complete a transaction to alter market share. If millions of consumers ask AI systems which products deserve attention, brands that perform well in those recommendations gain an advantage long before machine-controlled checkout becomes mainstream.
The dynamic may resemble the rise of search marketing in an unexpected way. Brands did not need Google to make purchases for consumers before ranking on Google became commercially essential. Google simply had to become influential enough in discovery that visibility there affected what consumers considered. AI agents may follow the same pattern. Their power could arrive first through recommendation and only later through transaction.
In Qatar, the Gap Between Assistance and Autonomy Is Already Visible
The difference between using AI and trusting AI to spend money is particularly visible in regional research. Visa reported in June 2026 that 90 percent of surveyed consumers in Qatar had used AI to assist with shopping, while 95 percent said AI tools made online shopping faster and easier. Yet only 28 percent said they currently trusted AI agents to complete checkout. The market is therefore not waiting for AI-assisted commerce to arrive because consumers are already using it heavily, but full delegation remains a much larger psychological step.
This pattern makes sense because recommendation errors and purchasing errors carry different consequences. If an AI assistant recommends the wrong pair of shoes, the consumer can ignore the suggestion. If an autonomous agent purchases the wrong pair of shoes with the customer's money, the problem involves authorization, refunds, fraud protection, customer service, and liability. Trust therefore has to increase dramatically as the agent moves closer to financial action.
Marketers should not mistake that hesitation for evidence that agentic commerce will fail. Consumers routinely adopt technology in stages. People once distrusted entering credit-card numbers online, then became comfortable storing cards in browsers, then moved toward digital wallets, contactless payments, recurring subscriptions, and one-click checkout. Autonomous purchasing may develop through similarly gradual delegation, beginning with low-risk repetitive purchases before expanding into more consequential decisions.
The likely transition is not from complete human control directly to complete autonomy. A shopper may first ask the agent for recommendations, then allow it to build the basket, then allow it to purchase only after receiving approval, then pre-authorize purchases within a budget or from certain merchants, and eventually allow particular recurring purchases to happen automatically. Agentic commerce therefore exists on a spectrum. Marketing strategy should prepare for the entire spectrum rather than imagining one dramatic moment when humans stop shopping.
Payments Are Being Rebuilt for Machines
For agents to become genuine buyers, payment infrastructure needs to distinguish legitimate automated purchases from fraud. Payment networks therefore have to solve questions that conventional ecommerce did not need to ask in the same way. Did the consumer authorize this agent? What is the agent allowed to buy? Is there a spending limit? Can it purchase only from specified merchants? What happens if the product is wrong? Who handles the dispute?
Visa has built Visa Intelligent Commerce around these problems, including tokenization, agent recognition, authorization controls, fraud protections, and systems for allowing consumers to define what agents are permitted to do. In July 2026, Visa announced that AI agents were already carrying out live transactions at participating European merchant sites rather than operating only in controlled demonstrations. Visa's broader architecture describes agent-native checkout as a transaction in which the agent acts according to consented instructions with a fully auditable authorization process. The point is not simply enabling machines to spend money but allowing them to prove that they are spending it legitimately.
Mastercard is building similar infrastructure. Its Agent Connect system allows participating AI agents to discover merchant catalogs, confirm prices and fulfillment, recommend items, and complete transactions using secure payment credentials while merchants maintain control over pricing, customer relationships, fulfillment, and business rules. Mastercard also introduced Agent Pay and mechanisms designed to verify that an AI-initiated purchase reflects the consumer's actual intent. In September 2026, the company described agentic commerce as a shift that will eventually affect everything from discovery to disputes.
PayPal has also entered the space with agentic commerce services that let merchants connect products to AI-powered shopping environments, manage carts, and complete purchases across stores and platforms. These developments matter for marketers because payments infrastructure usually follows consumer behavior rather than advertising hype. Visa, Mastercard, PayPal, Google, OpenAI, retailers, ecommerce platforms, and AI companies are all investing in ways to let agents participate safely in commerce. That does not guarantee rapid adoption, but it makes the idea much harder to dismiss as a temporary AI experiment.
Retailers Are Dividing Into Open and Closed Camps
Retailers face a difficult strategic choice because third-party shopping agents are simultaneously a source of demand and a threat to customer ownership. If a consumer asks ChatGPT to find the best product, the retailer may gain a sale it would otherwise have missed. But if the entire discovery process happens inside ChatGPT, the retailer loses some control over the relationship, merchandising, advertising environment, loyalty program, recommendations, customer data, and opportunity to encourage additional purchases. The agent can deliver customers while weakening the retailer's role as destination.
BCG's October 2026 retailer research found that 80 percent of surveyed retailers had already taken a position on external shopping agents. Forty-two percent were making their data discoverable to any agent, 32 percent were allowing access only to selected agents, and 6 percent were blocking agents altogether. The remaining retailers had not yet established a clear position. That diversity demonstrates that agentic commerce is not simply a technological implementation decision. It is a strategic decision about distribution and customer control.
Walmart has taken a particularly open approach. On October 21, 2026, it announced new infrastructure that allows third-party agents to shop more effectively on behalf of customers, including agent profiles for tracking authorization and spending limits and gateways through which approved agents can retrieve structured merchant information rather than scraping a website intended for human browsing. Walmart's approach contrasts with Amazon, which has been more restrictive toward external shopping agents while simultaneously building its own AI capabilities within the Amazon ecosystem. The two companies are effectively making different bets on whether the winning retailer of the agentic era should welcome external agents or keep more of the shopping intelligence inside its own environment.
For brands, this fragmentation means there may not be one agentic commerce strategy. A company may need to optimize for ChatGPT, Google, marketplace agents, retailer-owned assistants, social commerce, and conventional ecommerce simultaneously. Each environment may expose different product information, use different ranking signals, provide different advertising opportunities, and preserve different amounts of customer data. The supposedly frictionless AI shopping future could therefore become extraordinarily complicated behind the scenes.
Advertising May Split Into Human Persuasion and Machine Evidence
The traditional creative brief assumes a human audience. Who are we trying to reach? What do they believe now? What do we want them to believe after seeing the advertisement? Which emotional or rational message is most likely to move them? Those questions remain important, but marketers may increasingly need a parallel brief for the AI systems mediating discovery.
That second brief asks different questions. Can the agent clearly determine what the product is? Can it identify who the product is suitable for? Are the product's differentiators expressed as facts rather than vague marketing adjectives? Is current pricing accessible? Is stock information reliable? Are delivery promises accurate? Are reviews and ratings sufficiently strong? Are warranty, returns, compatibility, dimensions, ingredients, materials, and certifications structured clearly enough for machines to use?
This may create two layers of brand communication operating simultaneously. The human layer persuades through narrative, imagery, identity, emotion, design, entertainment, and cultural relevance. The machine layer persuades through accessibility, consistency, evidence, structured information, availability, performance, reputation signals, and relevance to the user's constraints. The best marketing will not choose between these layers. It will make them reinforce one another.
A premium luggage brand, for example, can still advertise travel, aspiration, craftsmanship, and identity to people. But when an AI agent compares suitcases for a customer who travels weekly, it may also evaluate weight, dimensions, warranty, wheel durability, repair options, customer reviews, airline compliance, delivery time, and price. The emotional brand creates preference while the machine-readable evidence creates eligibility. Lose either side and the competitor may win for different reasons.
Product Feeds May Become the New Media Asset
Marketing departments traditionally treat product feeds as technical ecommerce infrastructure rather than glamorous marketing assets. That is likely to change. A product feed contains exactly the type of structured information AI agents need to compare options at scale. If the feed is incomplete, inaccurate, stale, or poorly structured, the brand may become harder for AI systems to recommend even when the underlying product is excellent.
OpenAI's Agentic Commerce documentation makes this relationship explicit. Merchants can provide regularly refreshed product data containing identifiers, descriptions, pricing, inventory, media, fulfillment information, reviews, and additional product attributes. OpenAI says richer and more current feeds improve relevance and user trust. In other words, the quality of a company's product data directly influences the quality of its representation inside an AI shopping environment.
Google's Shopping Graph and agentic commerce infrastructure are moving in the same direction, while Walmart's agent gateways similarly give automated systems structured access rather than expecting them to interpret consumer-facing webpages. Mastercard's Agent Connect is designed partly around helping merchants make catalog information available across external AI ecosystems. The direction is consistent across companies: the internet was originally designed for humans clicking through websites, while agentic commerce increasingly requires businesses to expose structured information that machines can retrieve directly.
This does not make the website irrelevant. Websites still provide brand experiences, information, customer service, transactions, content, and authority. But companies may increasingly need to think of their product information as infrastructure existing independently from the presentation layer of the website. The human visits the storefront. The agent accesses the inventory system behind it.
Brands May Have to Optimize for Recommendation, Not Traffic
Digital marketing has traditionally treated website traffic as an important intermediate objective. SEO tries to earn visits, paid search buys clicks, social advertising drives landing-page traffic, and content marketing attempts to move audiences toward owned properties. AI-mediated journeys can break that sequence because the customer may receive enough information to make a decision without visiting the company's website at all. A product can be discovered, compared, evaluated, and perhaps eventually purchased inside an AI environment.
That does not make the marketing influence less valuable. It makes traffic a less complete proxy for influence. A brand could appear frequently in AI recommendations, gain consideration, and produce sales through marketplace or agentic checkout flows while receiving fewer conventional sessions. Conversely, a company might maintain strong website traffic while gradually disappearing from AI-mediated shortlists. Traditional analytics could therefore make the first company look weaker and the second stronger even as competitive power is moving in the opposite direction.
This is similar to the measurement problem created by AI search, but commerce makes the stakes higher. A citation inside an AI answer matters because it can shape knowledge. A recommendation inside a shopping agent can directly shape revenue. Marketers will need new measures for recommendation visibility, share of AI-generated consideration sets, agent referral traffic, product-feed performance, merchant selection, AI-assisted conversions, and eventually agent-initiated transactions.
The industry will undoubtedly invent a large and confusing collection of acronyms for this. The principle is simpler: marketers need to understand whether AI systems consider their brands relevant when customers describe the problems those brands solve. Ranking number one for a keyword will not matter much if the agent handling the customer's purchase never considers you.
Retail Media Will Have to Follow the Shopper Into AI
Retail media became one of advertising's fastest-growing categories because retailers possess valuable purchase data and can place advertisements close to the transaction. Sponsored products, marketplace placements, display advertising, in-store media, and off-site targeting allow retailers to monetize their customer relationships while brands pay for visibility at high-intent moments. Agentic commerce creates a problem because the shopper may no longer browse the retailer's sponsored listings in the traditional way.
If an AI agent requests product information through an API, evaluates the options, and returns three recommendations to the user, where does the sponsored product go? Does the retailer tell the agent which items are sponsored? Can brands bid to become part of the shortlist? Must paid recommendations remain visibly separated from organic recommendations? Could an AI agent reject a sponsored product because it is objectively less suitable for the user?
These questions are already moving from theory into experimentation. BCG reports that Dutch ecommerce company bol integrated retail media into its shopping agent from launch in September 2026 specifically because it wanted to learn how advertising works in a conversational environment. The industry therefore faces the same problem ChatGPT Ads already raises at a broader level: how do you monetize commercial intent without compromising the user's trust in the recommendation?
The answer will probably involve clearer separation between paid placement and organic recommendation, but the commercial pressure will be enormous. Retailers make significant revenue from brands paying for visibility, and those revenues do not disappear merely because the interface changes. Advertising will follow shopping into AI because advertising has followed shopping into every previous environment. The difficult part will be making sure the agent does not become a recommendation engine whose advice quietly belongs to the highest bidder.
The Battle Over Neutrality Will Become Huge
Consumers may trust AI recommendations partly because they believe the system is evaluating options according to their needs rather than according to which company spent the most money. That expectation creates a tension with advertising. Search engines solved the same issue imperfectly by labeling sponsored results, while social platforms label paid partnerships and sponsored posts. AI systems will need equally understandable boundaries between recommendation and promotion.
OpenAI currently states that product results in ChatGPT's shopping experience are selected independently and are not advertisements, while advertisements appear separately and are clearly labeled. That distinction is important because it allows the customer to understand when a product was selected because the system considered it relevant and when a brand paid for exposure. Whether this separation remains commercially sustainable as agentic shopping becomes larger will be worth watching closely.
The temptation to blur those lines will be strong because being the first product an AI assistant recommends could become extraordinarily valuable. Imagine the commercial value of being the default hotel, insurance provider, laptop, skincare product, restaurant, or airline recommendation across millions of high-intent conversations. The advertising industry will certainly attempt to buy access to those moments. Regulators and consumers will equally want to know when commercial relationships influence what the machine says.
Trust may therefore become the scarce currency of agentic commerce. If users believe the assistant consistently recommends inferior options because advertisers paid for them, they will stop delegating decisions to it. The agent has to preserve enough credibility that the recommendation remains useful. Paradoxically, this may force advertising to become more transparent than many existing digital formats rather than less.
Will Brand Loyalty Survive the Perfect Comparison Machine?
Agentic shopping also challenges one of marketing's oldest objectives: building enough preference that customers stop reconsidering alternatives every time they buy. A loyal customer does not perform a complete market analysis before ordering another bottle of the shampoo they like or booking the airline they usually use. Familiarity reduces cognitive effort. Brands benefit because switching requires a reason.
AI agents can remove much of that switching effort. A customer might instruct an agent to reorder a familiar product unless another option is at least 20 percent cheaper and receives better reviews. The agent can continuously check the market without the customer personally investing any effort. Loyalty therefore becomes conditional in ways that are difficult for traditional brands to control.
This could be particularly disruptive in functional categories where products are easily comparable. Batteries, household supplies, office equipment, commodity electronics, basic financial products, travel components, and many B2B supplies may become increasingly vulnerable to machine comparison. If the customer cares primarily about specifications, price, delivery, and reliability, an agent can optimize those criteria ruthlessly. Deloitte specifically notes that agentic commerce may favor standardized products where options can be compared objectively more than highly emotional categories where human preference plays a larger role.
The defense is not simply spending more on awareness. Brands need to create reasons the customer explicitly wants them, whether through superior product performance, distinctive design, emotional attachment, membership benefits, loyalty rewards, ecosystem compatibility, customer service, status, trust, or experiences competitors cannot easily replicate. The future of brand loyalty may involve convincing the customer to tell the agent, "I prefer this brand unless there is a compelling reason not to switch." That is a remarkably literal expression of brand preference.
The Customer Database Becomes Even More Valuable
Agentic commerce also strengthens the argument for owning customer relationships. A retailer that knows a customer's purchase history, sizes, preferred brands, allergies, loyalty status, delivery address, return behavior, and budget can provide a much more useful agent than an external system encountering the customer for the first time. The richer the first-party relationship, the more intelligent the shopping assistant can become. CRM therefore stops being merely a communications database and becomes potential memory for automated commerce.
BCG argues that retailers now need to play two games simultaneously. The destination game involves remaining valuable enough that shoppers continue visiting the retailer directly, while the evaluation game involves ensuring that external agents still consider the retailer's products and services competitive when shopping on the customer's behalf. Loyalty programs, marketplaces, personalization, customer data, and retail media all operate differently depending on which game the retailer is playing.
Brands face a similar choice. If every transaction is intermediated by an external agent, the business risks becoming a commodity supplier while the AI platform owns the customer relationship. If the brand has a strong direct relationship, it can use its own customer knowledge to create better experiences, offers, service, replenishment, and personalized recommendations. The return of CRM therefore becomes even more strategically significant in an agentic world.
This may eventually create competition between agents themselves. A customer could use ChatGPT for broad discovery but prefer a retailer's own agent for repeat purchases because that retailer knows their preferences better. A hotel chain's agent might understand loyalty status, previous rooms, preferred destinations, and upgrade history better than a general assistant. Companies therefore need to ask not only how they appear to external agents but whether they should build useful agents of their own.
Customer Experience Becomes Machine-Visible
One of the most consequential changes is that bad customer experiences become harder to hide behind good advertising. AI systems can synthesize reviews, customer discussions, return policies, delivery performance, public complaints, product comparisons, and other evidence much faster than individual consumers traditionally could. BCG argues that brand stewardship in the agentic era will increasingly depend on observable performance rather than solely on marketing claims. The promise and the experience therefore need to align much more closely.
Consider how many brands currently depend on customers not investigating too deeply. The advertisement promises exceptional service, while thousands of reviews complain that customer support never responds. The website claims easy returns, but online forums describe a difficult process. A hotel presents luxurious photographs, while recent guests repeatedly report noise and maintenance problems. A telecom provider advertises seamless connectivity, while public complaints tell a different story.
AI reduces the work required to discover those contradictions. A customer can ask the agent specifically whether people complain about returns, battery life, sizing, customer service, hidden fees, delivery delays, or product reliability. The assistant can summarize patterns that once required reading dozens of reviews manually. Reputation therefore becomes increasingly queryable.
This means marketers cannot treat post-purchase experience as somebody else's department. Customer service, logistics, product quality, returns, warranties, retail execution, and operations increasingly feed the information environment from which AI recommendations are built. Marketing's influence expands because the brand is being evaluated through evidence produced across the entire organization. The advertising department can no longer comfortably separate message from reality.
The Cheapest Product Will Not Automatically Win
There is a common assumption that AI shopping agents will destroy branding by reducing every purchase to price comparison. That is possible in some commodity categories, but it misunderstands how people actually make decisions. Consumers rarely optimize for one variable. They balance price against quality, convenience, trust, aesthetics, delivery, compatibility, service, familiarity, ethics, warranties, personal preferences, and countless other considerations.
An AI agent can incorporate those complexities rather than eliminate them. If a user says they prefer environmentally responsible products, the agent may prioritize sustainability evidence. If someone values fast delivery more than the lowest price, the ranking changes. If a traveler prefers a particular airline because they have status and accumulated miles, the agent can account for that. If someone refuses to buy from certain companies, those companies can disappear from the shortlist entirely.
This may actually make differentiated positioning more valuable. Brands need to understand which attributes customers genuinely value enough to include in their decision criteria. A product that costs more can still win if the higher price is supported by characteristics the customer cares about. The machine does not necessarily commoditize the category; it makes unsupported differentiation more difficult.
The real danger belongs to brands whose differentiation exists only in marketing language. If two products are functionally almost identical and one costs significantly more without stronger evidence, AI comparison may expose that quickly. Brands that genuinely deliver something distinctive have less to fear. Brands that primarily advertise distinctiveness may have considerably more.
How Marketers Should Prepare for Agentic Commerce
The first step is surprisingly unglamorous: clean up your product information. Make sure product names, descriptions, identifiers, categories, dimensions, technical specifications, compatibility, availability, pricing, delivery, returns, warranties, images, reviews, certifications, and other important attributes are accurate and consistent across your own site, marketplaces, merchant feeds, retailer systems, and AI integrations. An agent cannot reliably recommend information it cannot understand. Product-data management is becoming part of marketing whether creative teams find it exciting or not.
The second step is to audit what AI systems currently say about your products and category. Ask ChatGPT, Gemini, Perplexity, and other relevant assistants the questions customers are likely to ask. Do not search only your brand name because customers often begin with a problem rather than a company. Ask for comparisons, alternatives, best options for particular use cases, common complaints, reasons not to buy, and recommendations under different constraints.
The third step is to identify why your brand wins when a rational comparison occurs. If the answer is simply awareness, you may have a problem. Strong machine-readable differentiators include objectively better warranties, exclusive features, superior compatibility, verified performance, stronger customer ratings, local availability, faster delivery, certifications, better after-sales service, loyalty benefits, proven durability, transparent pricing, or something else the agent can recognize. Emotional distinctiveness remains valuable, but rational proof gives AI systems more reasons to keep you in the consideration set.
The fourth step is to build stronger first-party relationships. Encourage customers to create accounts, join loyalty programs, register products, communicate preferences, subscribe where useful, and interact directly with the brand in ways that create value for them rather than merely harvesting data. The richer that relationship becomes, the more personalized your own future AI services can become. Customer data increasingly becomes an asset for both marketing and agentic experience design.
The fifth step is to prepare technically for agent access. Investigate whether your commerce platform already supports emerging protocols and product-feed standards used by OpenAI, Google, Shopify, payment networks, and other ecosystems. Decide which agents you want to allow, what information they should access, how often feeds need updating, and how transactions will be authenticated. This work will increasingly involve marketing, ecommerce, IT, security, payments, legal, and operations rather than belonging to one department.
The sixth step is to think about advertising beyond the human click. Your future media plan may need to include conventional creative advertising, search visibility, AI answer visibility, retail media, product-feed quality, agent integrations, conversational advertising, and sponsored opportunities inside AI commerce environments. Those activities will sit across different teams if organizations maintain today's structures. The companies that adapt fastest may be the ones that recognize they are all parts of the same demand system.
Marketers May Need to Win the Human Before the Agent Arrives
There is another defensive strategy that sounds almost old-fashioned: build such strong brand preference that the customer gives the agent instructions favoring you. An AI shopping assistant may compare dozens of coffee machines, but if the user explicitly says, "I usually buy De'Longhi," the brand enters the evaluation with an enormous advantage. The same principle applies to Nike, Apple, Toyota, Emirates, LEGO, Sephora, or any brand strong enough to become part of the customer's criteria rather than merely one option among many.
This is why agentic commerce does not eliminate upper-funnel marketing. It may make it more important. Advertising, sponsorships, creators, entertainment, physical experiences, packaging, PR, cultural relevance, and distinctive brand assets shape preferences before the customer asks an AI system to help them shop. The agent can optimize the purchase, but it may begin from preferences marketing created years earlier.
Brand building therefore becomes one way of influencing the machine indirectly. You are not persuading the algorithm to love the brand. You are persuading the human to tell the algorithm that they do. That may be the most powerful targeting signal of all.
This also creates a fascinating division of labor between brand and performance marketing. Brand marketing creates the customer's priors, while machine-readable performance information helps the agent validate the choice. The company wins twice when the human prefers the brand and the agent finds enough evidence to agree.
What Happens to the Marketing Funnel?
The traditional funnel imagines awareness leading to consideration, purchase, and loyalty. Digital journeys already made that model messier because customers jump between stages constantly, but agentic commerce could compress several of them together. A consumer can become aware of a product, compare it with alternatives, read summarized reviews, check pricing, and prepare a transaction within one conversation. The distance between discovery and purchase becomes shorter even when the reasoning behind the decision becomes deeper.
McKinsey argues that customer experience in the agentic era will shift from predefined journeys toward dynamic orchestration. Instead of companies designing one ideal sequence of touchpoints, agents may construct journeys in real time based on the customer's goal and context. One person might need extensive research, while another simply wants the cheapest available replacement delivered tomorrow. The path changes according to what the customer delegates.
This makes conventional funnel attribution even harder. A company may influence awareness through social media, become validated by ChatGPT, appear through a retailer's product feed, get compared by Google, and complete payment through an agentic checkout flow. Which channel created the sale? The answer is increasingly that the sale emerged from an ecosystem of influences no single platform can completely claim.
Marketers will need to care less about assigning perfect credit and more about ensuring the brand performs well across the entire decision environment. That means building demand, being discoverable, being understandable, being credible, being available, and removing friction when the decision finally turns into action. Agentic commerce makes the journey less visible but not less strategic.
Advertising Is Becoming Infrastructure
Perhaps the biggest lesson from this three-part series is that advertising itself is becoming increasingly infrastructural. In Google Ads, AI is taking over more of the bidding, targeting, matching, creative variation, and campaign execution that marketers once controlled manually. In ChatGPT Ads, advertising is entering conversations where the platform understands the customer's problem in more detail than a conventional keyword ever could. In agentic commerce, AI may increasingly influence whether the product enters consideration and eventually participate directly in the purchase.
That moves marketing away from thinking exclusively in terms of discrete advertisements. Product information becomes advertising infrastructure. CRM becomes advertising infrastructure. Reviews become advertising infrastructure. Customer experience becomes advertising infrastructure. Inventory accuracy, delivery performance, merchant feeds, APIs, and structured data all influence whether the brand can participate effectively in machine-mediated demand.
This does not make creativity obsolete. In fact, it may make creative work more important because everything machine-readable is easier for competitors to imitate than a genuinely distinctive identity. Products still need stories, brands still need meaning, and humans still need reasons to care. The difference is that beautiful communication increasingly sits on top of an invisible technical foundation that determines whether AI systems can understand and trust what the brand is saying.
Marketing therefore becomes both more technical and more human at exactly the same time. Machines handle increasing portions of execution and evaluation. Humans remain responsible for meaning, positioning, taste, ethics, creativity, emotional connection, and deciding what the business should stand for.
The End of Advertising to People Alone
Advertising is not about to stop targeting human beings. People will continue watching videos, walking past billboards, following creators, searching Google, visiting stores, reading reviews, attending events, opening emails, noticing packaging, and talking to other people about what they buy. But humans may no longer be the only audience that matters between the marketing message and the sale. AI systems are becoming another participant in the decision.
The immediate challenge is recommendation. Can the AI find you, understand you, trust your information, and determine that your product deserves to be considered? The next challenge is transaction. Can the agent verify availability, confirm pricing, understand your rules, complete payment securely, and transact within the customer's instructions? Eventually, the challenge may become retention, with agents continuously deciding whether the customer should keep buying from you or switch to something better.
That future should make marketers uncomfortable in useful ways. It exposes weak differentiation, inaccurate product information, poor customer experiences, artificial loyalty, vague positioning, and brand promises unsupported by evidence. At the same time, it creates extraordinary opportunities for companies whose products genuinely solve problems and whose information makes that value easy for both humans and machines to understand.
The brands most vulnerable to agentic commerce may not be the ones with the smallest media budgets. They may be the ones that depend most heavily on customers not comparing them too carefully. The strongest brands will still create emotional preference, but they will also make that preference rationally defensible when an AI system investigates.
The future of advertising therefore may not be a battle between humans and machines. It may be a two-audience problem.
You still need the human to want you.
You increasingly need the machine to understand why.
