AI Is Upending Marketing on Two Fronts

We’re in the middle of two concurrent revolutions that will reshape how companies compete for customers. One is about how consumers search for information. The other, just getting started, is about who makes purchasing decisions. Marketing management has always been about moving people through the funnel, from awareness to consideration to purchase. That journey depends on how consumers learn about the marketplace, and that learning process is exactly where these revolutions are hitting hardest. For decades, marketers have built awareness through mass media and then captured demand through search engines, but conversational AI—systems such as ChatGPT and Gemini that interact with users using natural language—is now disrupting both stages at once. What follows draws on the most recent academic research on this topic. The specific details will change, of course, but the direction of change is clear. The First Revolution: Learning Without Websites AI chatbots are dealing retailers a double blow. Think about the old model: Someone searches on Google, gets a list of links, clicks through to various websites, compares options, and eventually buys. But consumers are now turning to chatbots for information. When someone asks ChatGPT or Claude for product recommendations, they get a complete answer without ever seeing a brand’s carefully crafted web presence, and the chatbot mentions far fewer options than a traditional search results page would show. Google’s response to ChatGPT made things worse for many retailers. Those AI overviews at the top of search results? Many users read them and never scroll down to the links below. Brands that spent years building their search presence are watching that traffic erode. People are finding what they need through chatbots and never landing on a webpage. The aggregate data, as discussed last year in The Economist, backs this up: Web traffic patterns are shifting fast. AI is killing the web. When consumers today ask a chatbot for recommendations, they get a curated answer shaped by training data and algorithms. Brands that show up in these responses get disproportionate visibility, and everyone else might as well not exist. Researchers have started quantifying these shifts. One Boston University study, for example, looked at Stack Overflow, a platform popular with software developers. After ChatGPT launched, traffic on the platform plunged. Why? Because developers had figured out that they could get answers from ChatGPT and didn’t need to ask other people. But here’s where things get interesting: The researchers also looked at Reddit, another social platform popular with developers, and there they found no comparable effect in the subreddits (or communities) popular with developers. The difference comes down to purpose: People go to Stack Overflow for specific information, something that chatbots can also give them, whereas they go to Reddit for community, discussion, and the experience of interacting with other people, which chatbots can’t give them. The strategic implication of this emerging division is clear: Brands built primarily on delivering information are exposed, whereas those built on community, emotional connection, or experience have more protection. A study by researchers at London Business School and UCLA took this line of research further. Using ComScore’s Web Behavior Panel, a dataset with over a million users and URL-level browsing data, they tracked individual behavior and found, after controlling for confounds, that online searches drop roughly 20% after people adopt ChatGPT. Smaller websites suffer the most, because they lack brand recognition that prompts direct navigation, and the effect is strongest among high-retail-activity users. The upshot is that the customers worth the most are shifting their behavior the fastest. From SEO to GEO These shifts are going to demand a rethinking of digital strategy. The techniques that worked in the search engine era won’t necessarily translate to conversational AI. Keyword optimization, link building, metadata refinement—none of that maps directly onto how chatbots work. Instead, what’s happening is that we’re moving from an era of search engine optimization, or SEO, to one of generative engine optimization, or GEO. The problem is that nobody really understands GEO yet. The rules are still being written, and companies that figure out effective GEO strategies early will have a real edge before best practices spread. The key will be to understand something important: that optimizing for conversational AI isn’t a technical SEO exercise. It requires grasping how these systems process and synthesize information, which sources they favor, how they construct responses. It means rethinking content strategy entirely—not just what you say but how you structure information so that AI systems can parse it, contextualize it, and recommend it. Structured data, clear categorization, and comprehensive content that directly answers common queries probably matter more than the link-based authority signals that have dominated SEO. (In 2024, in an article for HBR.org, we took a first crack at explaining why chatbots choose to mention the brands they do.) Where Conversational AI Sits in the Funnel To anticipate where this is heading, you need to understand how conversational AI will make money. Consider how the two big digital marketing channels differ. Meta and other social media companies monetize attention—advertising in feeds, targeted by demographics and behavior. That’s top-of-funnel marketing, building awareness among people who aren’t actively looking for anything in particular. Google, on the other hand, monetizes intent. Type a query and you’ve signaled active interest, which puts you much further down the purchase journey. That intent makes search ads valuable for driving conversions. In a new paper with Erik Hermann and David Schweidel, we argue that conversational AI sits in between those two extremes. Users express interest through prompts, which is more targeted than passively scrolling social feeds, but often that happens earlier in the decision process than, say, a Google search for information. Someone prompting a chatbot about a topic relevant to a product category may be interested in it but is also probably not ready to buy. Recent data seems to confirm this assumption. Researchers from Germany used first-party data from over a thousand websites and looked at the value of referrals, including referrals from ChatGPT, which constituted a small percentage of all referrals. They found that ChatGPT referrals generate revenue per session that falls in the middle between social media referrals (top of funnel, low intent) and search (high intent). That middle position creates interesting dynamics. For brands, it means a new touchpoint where consumers are engaged and curious but haven’t yet developed specific purchase intent. That’s valuable real estate in the customer journey. It’s also why the advertising battle for conversational AI is going to be fierce. OpenAI recently announced that advertising is coming to ChatGPT, a move the company has to make if it hopes to continue offering a free plan to hundreds of millions of users. The company also published their advertising principles, which it said are designed to prioritize consumer trust and avoid the damage that social media advertising has created. The principles are admirable, but that middle-funnel position creates strong incentives to push farther down toward purchase intent, which, in turn, is likely to lead to more aggressive tactics. It’s going to be hard to resist the temptation to capture more eyeballs. The Second Revolution: Bots as Customers The search revolution is disruptive enough, but a second transformation is coming that could matter even more. To see it, you need to start making a distinction between two terms that marketers often use interchangeably: customer and consumer. A customer is somebody who makes a purchase decision, whereas a consumer is somebody who uses a product. Sometimes they’re the same person, but sometimes they’re not: Parents buy for children, doctors prescribe for patients, procurement directors buy supplies for other departments. This distinction matters now because AI agents are starting to make purchasing decisions. The question is no longer just “Who is your customer?” It’s also “What is your customer?”—because your customer might be an algorithm. In the age of AI agents, the decision-maker and the end user may no longer be the same entity. Marketing will have to attend to both. Imagine a busy professional who tells their AI assistant, “I need a new laptop for work. My budget is two thousand dollars. I care about battery life and display quality. Order one for me.” The AI agent researches options, weighs tradeoffs, and makes a purchase. No person ever visited a product page or read a review. The entire customer journey happened inside an algorithm. When an AI agent can research options, evaluate alternatives, negotiate prices, and complete purchases for a person, marketing as we know it will need to be rebuilt. We’re in the early stages of this transition. It starts with asking for recommendations but will evolve toward transactions. AI agents already have most of the technical capabilities to do such autonomous work. What’s missing is infrastructure, trust frameworks, and adoption. But the trajectory is obvious: Just as smartphones forced a shift from desktop-first to mobile-first thinking, AI agents will force many marketers to make a shift to machine-customer-first strategy. Companies that start preparing now will be ready when the market matures. Designing for Machine Customers Our web infrastructure was built for people. We navigate through visual interfaces, clicking tabs, scrolling menus, selecting options. This works because we’re visual creatures with fine motor skills. But AI agents are different. For a bot to use a typical website, it needs sophisticated computer vision and interaction simulation. Alternatively, the bot could much more efficiently read structured data formats that it can parse directly. The future web will therefore rely heavily on dense files and structured data that AI agents can navigate without pretending to be human browsers. In the meantime, brands face a practical problem: maintaining web presences that work for both people and AI agents. Architectural changes will be required to ensure that product information, pricing, and availability are exposed in formats both audiences can access. Most companies aren’t ready for this. Their websites assume a person is browsing. Their checkout processes assume a person is clicking. Their customer-service systems assume a person is on the other end. Retrofitting all of this for machine customers will take significant investment and a mindset shift that many organizations haven’t even begun. The Emerging Field of Bot Psychology When people were the only customers, marketers leaned heavily on psychology—to understand the impact of cognitive biases, emotional triggers, decision heuristics, social influence, and design go-to-market strategies and related messaging. If bots become customers, marketers will also need to understand bot psychology. That will be trickier than it sounds. AI systems have consistent behavioral patterns, but those patterns are often surprising, inconsistent across models, and counterintuitive. Research on this is just beginning, but the findings should make every marketer nervous. One study had both people and AI write advertising copy, then had both rate the results. Researchers found what they call AI-AI bias: AI systems rated AI-generated content higher than people-generated content. Think about what that means. In a world where AI agents make purchasing decisions, people-created marketing might lose out not because it’s worse but because of structural biases in how AI evaluates information. We’ve spent decades worrying about human biases in decision-making and building systems to counteract them. Now we need to worry about machine biases too. Researchers from Columbia and Yale, working with an AI startup, built a sandbox to study how agents behave on e-commerce platforms. What they found reveals a complicated landscape. Like people, AI agents have position effects; that is, they favor products in certain display locations. Those preferences vary wildly across models. The researchers tested GPT, Claude, and Gemini and found that all three prefer products in the top row of a display. But within that row? GPT strongly favors the first position, Claude prefers the middle, and Gemini favors the right side. If you’re a retailer trying to optimize page layout, good luck figuring that out. The researchers also found that AI agents penalize “sponsored” tags, reward endorsements, and discount information when they detect commercial influence. In some ways, bots behave more rationally than people, who are famously susceptible to advertising even when they know it’s advertising. AI agents also respond to price and quality ratings, but the weighting varies by model. Presented with the same products and same information, different AI models made substantially different purchase decisions. What Marketing Leaders Should Do Now These changes are bewildering and scary, because they’re happening so fast and because of where they seem to be taking us. So how should marketing leaders respond? Here are steps that they can take to navigate the two revolutions: Start by auditing your exposure to the search revolution. What share of traffic and conversions comes from search engines? How much of your digital strategy depends on people clicking through to your site? High dependence on information-seeking traffic means high near-term risk. Build expertise in generative engine optimization. The field is young and moving fast. Experiment with structuring content for chatbot visibility. Track how your brand appears in AI responses. Develop internal knowledge before your competitors do. Double down on what AI can’t easily replicate. The Stack Overflow versus Reddit finding matters here: Community, emotional connection, and experience are harder to displace than information delivery. Think about how your brand can deepen those dimensions. Prepare for AI customers. Start thinking about how an algorithm would evaluate your products. What structured data would it need? How would your value proposition translate into the factors AI systems weight? Watch the research. Academics and analysts are generating new findings fast. Building the capacity to track and act on these insights will pay off. Rethink content strategy for a dual audience. Your content needs to resonate with human readers while also being structured so AI systems can process and cite it. This doesn’t mean writing for machines instead of people. But it means recognizing that the same content will be consumed through very different pathways. Treat this as a leadership issue, not a marketing-department problem. These transformations touch how companies connect with customers at every level. They require coordination across technology, content, product, and customer experience. Organizations that keep these capabilities siloed will struggle, whereas those that integrate them at the executive level will move faster. . . . Marketing has survived disruptive transitions before: print to broadcast, broadcast to digital, desktop to mobile. Each created winners and losers. This transition goes even deeper. We’re changing not just how marketing reaches human decision-makers but potentially also who those decision-makers are. The early evidence suggests that we have a rough ride ahead, given variations in AI agent behavior across models and the opacity of how chatbots weight information. We’re heading toward a period of uncertainty and experimentation. But uncertainty is where opportunity lives. The companies that develop expertise in conversational AI marketing and that figure out how to optimize for both human and machine customers will reap large rewards.

AI Is Upending Marketing on Two Fronts

Summary.

Partner Center