A 23-year-old in Echo Park wants new running shoes. She doesn’t open a browser. She doesn’t walk into a store on Sunset. She types one sentence into an assistant: best running shoe for me. And the most famous athletic brand on the planet doesn’t make the top three.
That isn’t a hypothetical we made up to scare you. It’s a test case cited this month in a Fast Company op-ed by Nir Wegrzyn, CEO of the brand consultancy BrandOpus. Ask an AI for a running shoe for a 23-year-old, he writes, and Nike didn’t land in the top three. Brooks, per testing reported in Harvard Business Review, surfaced more reliably. Ask for a cream for sensitive skin and Dove, a brand with a decades-long claim on gentle care, didn’t appear at all. The assistant associated it with affordability instead.
Nike. Dove. Two of the best-funded brand machines in history, and the robot shrugged.
Wegrzyn’s line for this is clean: “AI does not feel your brand.” We’d put it less politely. All the warm, fuzzy, emotionally resonant shit you spent a decade building with humans doesn’t transfer to a machine by osmosis. The agent isn’t nostalgic. It didn’t cry at your Super Bowl spot. It reads what’s there, weighs what’s consistent, and moves on.
And it’s about to do a lot of the shopping.
Morgan Stanley estimates that agent-influenced shopping could account for $190 billion to $385 billion of U.S. e-commerce by 2030, somewhere between 10% and 20% of the market. That’s not a pilot program. That’s a new customer, with a lot of money, who has no idea who your founder is and doesn’t care.
So what does this customer actually read?
According to the Fast Company piece, it reads consistency. Wegrzyn points to research from Charlie Oscar, reported by WARC, that looked at 30 brands across six categories and attributed 63% of LLM visibility to long-term brand equity. Not this quarter’s campaign. Not the trend you jumped on in March. Years of saying the same thing, the same way, with the same assets. Volvo means safety because it has meant safety for decades. New Balance reads as authentic. Lululemon reads as wellness. The machine is a ruthless auditor of whether your story holds together over time.
Which brings us back to Nike, and the reason Wegrzyn gives for its stumble. “Just do it.” “Believe in something.” “Why do it?” Each line is great. Together, to a machine trying to summarize what Nike stands for, they read like three different brands sharing a swoosh. Humans fill the gaps with feeling. Agents don’t have feelings to fill them with.
Now here’s the part that should make every marketing director squirm, because it isn’t about the brand at all.
It’s about people.
The same week, Social Media Today reported on research LinkedIn’s AI search marketing lead, Robert Yanik, cited from Meltwater. Accounts with more than 10,000 followers, together with company pages, made up 65% of LinkedIn citations in AI chatbot answers. Yanik’s framing: “AI models like ChatGPT, Gemini, and Copilot are built to reward credibility.” His advice, roughly: pick three or four themes you want to own, get credible employees posting about them, and find one or two creators whose audience overlaps with your buyers.
The article was careful to say the data is correlational, and we’ll be careful too. Nobody outside these companies knows exactly how each model picks its sources. But the direction is hard to miss. The machines lean on named humans with track records. The faceless brand account posting a stock photo and “Happy Monday, team!” isn’t building anything an agent will cite.
Put those two findings next to each other and you get an uncomfortable picture. The agent trusts brands that are consistent, and it trusts people who are credible. A lot of companies are neither. They rebrand every three years because a new CMO wants a legacy. They run campaigns that contradict the last campaign. Their smartest people, the partner who actually knows how the work gets done, the head chef, the lead clinician, never publish a word.
Try a simple test. If an assistant were asked tomorrow to recommend a firm like yours, which of your people could it name? Which article, which post, which interview would it pull from? For most mid-sized businesses the answer is nobody and nothing. The website says “passionate team.” The LinkedIn page reposts an award. The real expertise lives in someone’s head and a few client emails. That’s invisible to a human scrolling, and it’s invisible to a machine for exactly the same reason.
This is where someone in the meeting says, “So we need an AI strategy for this.”
No. You need the strategy you were supposed to have all along.
Every principle that makes a brand legible to an agent is something brand people have been begging for since before the internet. Own a small number of distinctive assets and stop redesigning them. Wegrzyn argues for at least three. Heinz kept its keystone logo while it refreshed everything around it; that’s the move. Tell one clear, relevant story and repeat it until you’re sick of it, because the moment you’re sick of it is roughly when the market starts to remember it. Put your real experts in public, under their own names, saying specific things.
None of that is new. What’s new is the penalty for ignoring it. For years, a messy brand could still win with a big enough media budget, because humans are forgiving and attention can be bought. An agent doesn’t care how much you spent on the flight. It cares whether the signal is coherent. You can’t outspend incoherence anymore. You can only fix it.
This is also where a lot of agency logic quietly dies. The industry has made a very good living off constant reinvention: new platform, new campaign, new tagline, new look. Some of that is necessary. Plenty of it is just motion that bills well. When the customer is a machine scoring you on consistency, reinvention for its own sake is a self-inflicted wound. The best creative work for the next five years may be the discipline to not change the thing that’s working, and to make every new piece carry the same core.
There’s a culture point buried in here too, and it’s the one we find most interesting. Think about the places in Los Angeles that people never stop talking about. The taco stand with the same hand-painted sign for thirty years. The record store that’s had the same clerk, same opinions, same layout since your older brother shopped there. They didn’t grow by reinventing themselves. They grew by being unmistakably themselves, every single day, until the city couldn’t describe its own neighborhood without them. That’s exactly the kind of signal a machine can read. Turns out the old-school way of building a name is the most future-proof thing you can do.
So the job is not to “optimize for AI.” The job is to become impossible to summarize wrong.
Three assets you never touch. One story you never abandon. Real people with real names who say true, specific things in public. Do that, and the agent describes you the way you would. Skip it, and the agent describes you the way it guesses.
And it guesses fast, with total confidence, for millions of people at once.
You used to be able to charm your way into the cart. The new shopper isn’t buying the charm. It’s reading the receipts.