July 2026 | Marketing

Five Things About AI Every Marketer Needs to Know

Marketing has always been about persuading, reassuring, and winning over a buyer. And until very recently, that buyer has always been a person. But increasingly with AI, the buyer who is comparing options and clicking on “buy” is not a person at all, but instead an AI agent acting on a customer’s behalf. That change — and others brought about by AI — is sending shock waves through marketing departments, whose carefully and creatively constructed brand stories are now falling on literally deaf ears.

Consider an ordinary errand. You need a flight to Seoul, within set dates and a budget, and you want the best deal across several airlines’ loyalty programs. Instead of taking the time to book it yourself, you hand the job to an AI agent: software that doesn’t just answer your questions, but acts on your behalf, not only weighing the options but also completing the purchase.

“2026 is proving to be the takeoff year for personal agents,” says Wharton’s Stefano Puntoni, who co-directs the school’s Human-AI Research center. And the fact that an agent cannot be charmed removes marketers’ oldest advantage while raising a question the field has never had to answer: does a brand still count for anything to a buyer that feels nothing, or does the agent simply pick the cheapest option that does the job?

AI is forcing more than this one question on marketers. It is reshaping the function in at least five ways, most not yet part of how marketers think about the technology. Taken together, they read less as a forecast than as a correction: of the hype on one side and the gloom on the other.

Efficiency Is the Easy Story

AI changes the company before it ever reaches out to the customer, and most firms start in the same place: cost. They ask how AI can do today’s work faster or with fewer people, because that return is the easiest to see and the easiest to justify. The bigger prize is harder to picture. As Eric Bradlow, Wharton vice dean of AI and analytics, puts it, “You can only drive cost to zero, but you can drive revenue as high as you want.” The deeper insight is that the two aren’t symmetrical. Savings stop at zero, and the closer you get the less is left to take; revenue has no such limit. Aiming AI at cost means working toward a fixed, shrinking target while the larger prize goes untouched.

That’s not surprising, since the pace of AI adoption itself works against that bigger prize. Timelines are so compressed, Puntoni says, that companies never give themselves a chance to step back and ask what a function could do differently. “You have to do it next week,” he says, “and next week you can only put the plaster on.” The firms that pull ahead will be the ones that pause long enough to rethink the work — what AI might actually be able to do for the firm — not just speed it up.

AI Rewards the People with Taste

AI does not improve everyone’s work the same way. It sharpens judgment more than it supplies it, so what a marketer gets out of the technology depends on what they bring to it. “It can lead to improved decisions,” says Annie Wilson, Wharton faculty director of the Marketing Undergraduate Student Establishment (MUSE), corporate AI advisor, and instructor in the AI in Marketing: Creating Customer Value in an AI-Driven Enterprise program along with Puntoni and Bradlow. “But the important caveat is that the result comes when AI is paired with good judgment or good taste.” The people who already know a good idea from a bad one can tell a strong AI output from a merely plausible one, and can see which uses are worth pursuing and which are dead ends. That filtering instinct, the thing that has always separated strong marketers from the rest, is exactly what AI rewards.

Wilson sorts users into two groups. "There are the Thinks and the Think-nots,” she says. The Thinks use AI to make sharper, better-informed decisions, and to reach them faster. The Think-nots “use it to replace thinking and judgment,” deferring to whatever the model returns. Over time, that produces a split: the strongest work gets stronger, while the rest settles into the competent and forgettable. AI lifts the floor, raising the baseline anyone can reach, but the floor is not where marketers compete. “The floor has definitely already gone up,” Wilson says. “So, the question is how high the ceiling is.” When competence is everywhere, the advantage belongs to whoever can push the ceiling, and that is most likely to be the marketers who already have the taste and discernment to separate what's good from what only sounds good.

The Funnel Shifts at Both Ends

The clearest shifts a marketer will feel are at the two ends of the buyer’s journey: how people find products and how they buy them. Both are moving toward AI.

At the top, discovery is migrating into AI-generated answers. Search is now one of the largest uses of ChatGPT, Puntoni notes, much of it about the marketplace. Google has pushed the same way, and hard: in what it called the biggest change to Search in 25 years, it now leads with AI-synthesized answers rather than a list of links. The links still exist, but they are no longer the point. The shift is changing behavior even among the people who study it. “I scroll down to that first,” says Bradlow. “You can give me all the paid stuff, but I’m going straight to the AI.”

That creates a problem for brands. The answers are good enough that “nobody's clicking on anything,” Puntoni says, and when the clicks stop, so does the traffic the open web is built on. No one has yet worked out how to make money from an answer that keeps the reader in place, a prospect he calls “killing the internet.” For marketers, the question moves from whether a page ranks to whether the brand appears inside the answer at all.

At the other end sits the fear that gives brand managers pause: that a shopping agent will treat every purchase as a spreadsheet, choosing on price, attributes, and availability, and strip away whatever premium a brand once commanded. Puntoni thinks that fear is real in some categories and overblown in others. A strong brand can be a “moat,” he suggests, not because it survives the agent’s scrutiny but because it keeps the shopper from inviting that scrutiny at all. A customer with a real preference simply tells the agent to buy the brand. Even in a category as commoditized as bottled water, brand still moves share, as Liquid Death’s rise shows.

Wilson adds the point most marketers will not have considered: brand reaches the agent indirectly through the language the models are trained on. Reviews, earned media, and consumer sentiment all feed the machine. Dove is talked about more, and more warmly, because of its real-beauty campaigns, which makes it likelier to surface when an agent goes looking.

That mechanism, Bradlow notes, runs in a loop. “Stronger brands take more space in the lexicon of language and video,” he says, “so they’re represented more in the models, which means they're represented more in what gets shown.” Being talked about gets a brand surfaced, and being surfaced gets it talked about more. The loop governs both ends of the funnel, and it turns on the oldest question in marketing: how much, and how well, people are talking about you.

From Artist to Art Critic

Run through every one of these shifts and the same requirement comes up: someone has to judge whether the output is any good. Puntoni frames it as a change in the job. “You’re moving from being an artist to being an art critic,” he says, “and it’s very hard to be an art critic.” The technology is engineered to sound plausible, so a reader’s guard has to stay up against output that is fluent, confident, and occasionally wrong. But a “human in the loop” offers little protection, he warns, when “the human is asleep in the loop.” The real question is how the work will be verified (and by who, and how) when the model is right 90 percent of the time.

The danger is sharpest for those least able to catch it: the novice who, in Wilson’s words, cannot “parse out the slop from the signal.” The marketing expert’s job is shifting toward verification, which raises a harder question: how do you keep producing experts when the lower rungs of the work are the first to be automated?

The Customer May Not Be a Person

Which brings us back to the beginning. Puntoni expects a quiet but profound split in two words marketers have always used interchangeably. The consumer, the person who wants the thing, stays human. The customer, the one who actually searches, compares, and buys, may increasingly be “an AI system, an agentic system powered by large language models.” Companies will have to optimize not for a human scanning a page but for a machine choosing on a person’s behalf. His hope is that the payoff reaches past efficiency, toward experiences that create real value and that no one could have offered before.

That is the through-line of the whole conversation. AI raises the floor for everyone. What it cannot supply is the judgment about where to aim, the taste to tell good from merely fluent, and the brand that means something to a person even when a machine is doing the buying. Those, for now, remain stubbornly human.

---

Share This

Subscribe to the Wharton@Work RSS Feed