The Agent Will See You Now: Marketing When AI Does the Shopping
Key Stats
20% of Cyber Week 2025 orders were influenced by AI and agents.
AI could complete up to 25% of online transactions by 2030.
23% of Americans used AI to help make a purchase in the past month.
AI has started to buy things, not just compare them. Here is what changes when an agent sits between your brand and the buyer, what the data already shows, and how to stay on a shortlist a machine builds.
For thirty years, digital marketing rested on one quiet assumption: a human would eventually decide. They might arrive by search, by ad, by a friend’s link, but at some point a person with eyes and a wallet landed on a page you controlled. That assumption is breaking.
The thing arriving at your site is increasingly not a person deciding. It is a person who has handed the deciding to software. They asked an assistant for the best mid-range CRM for a 40-person team, and the assistant did the looking. It read the reviews, compared the specs, and came back with a short answer and maybe a link. This is not a forecast. Salesforce found AI and agents influenced 20% of all orders during Cyber Week 2025, driving $67 billion in sales, and Morgan Stanley estimates roughly 23% of Americans bought something with AI’s help in the past month.
Here is what makes this hard for a marketing leader. You cannot charm an agent. You cannot retarget it. It feels no urgency, ignores your limited-time offer, and will not be moved by a clever headline. It reads, it weighs, it decides. Most of the marketing you built was designed for a person in a hurry, not a machine doing patient research on that person’s behalf.
The pain shows up first in your dashboard. Sessions look flat while the deals that close are warmer and faster, because something is filtering your traffic before it reaches you. Then it becomes a control problem. The agent summarized your category for the buyer, and you have no idea whether you were in the summary, how you were described, or who got named instead.
So the shift is easy to state and hard to absorb. You used to optimize to be chosen. Now you have to optimize to be legible. An agent does not read your brand. It reads your data: your structured information, third-party reviews, comparison content, and the consensus about you across the web. It rewards what it can parse cleanly and skips what it has to guess at.
The agent already shops. It won’t buy yet.
of Cyber Week 2025 orders were AI-influenced
Salesforce, 2025of Americans bought via AI last month
Morgan Stanley, 2025still won’t let an agent buy on its own
Bain, 2025higher conversion from AI-referred visits
Adobe, 2026The money is not speculative
Juniper Research projects agentic commerce transaction value climbing from about $8 billion in 2026 to $1.5 trillion by 2030. McKinsey puts US agentic retail near $1 trillion by then. Bain expects AI to touch most online shopping in some form, completing up to a quarter of transactions. The forecasts disagree on the exact number, and it does not matter much. One transaction in four or five is not a niche.
Trust is the brake on that curve, and it matters as much as the engine. Bain finds about half of US shoppers still will not let an agent complete a purchase on its own. Salesforce reports that even among people already shopping with AI, only 48% are open to letting it buy for them. So the honest read: fully hands-off purchasing is still early and still distrusted, while agent-assisted research, where the human keeps the final click but lets the machine gather, is already mainstream. Build for the assisted journey now and you are positioned for the autonomous one when trust rises.
The agent concentrates your demand
The traffic that does arrive through AI behaves differently, and better. Adobe found AI-referred shoppers converting 42% higher than other traffic, spending 48% longer on site, and worth 37% more in revenue per visit. Salesforce measured AI-referred traffic converting at eight times the rate of social. Fewer visitors reach you this way than through old organic search, but the ones who do are further along and more likely to buy. The agent is not killing your demand. It is concentrating it, doing the filtering you used to do with landing pages and nurture emails, and handing you the survivors.
There is a danger in that, and it is where careless brands lose. Left to its defaults, an agent optimizes for what it can measure cleanly: price, specs, ratings. If all it can extract about you is a price and a feature list, it will reduce you to a price and a feature list, then compare you to everyone else on those flattened terms. That is commoditization by algorithm, and commodities compete on price, a race no growing company wins. It bites first in spec-driven categories. Morgan Stanley finds AI purchases are already most common in groceries and everyday goods, where the cheapest match wins.
The defense is to make your real differentiation legible. If what sets you apart is reliability, or fit for a particular kind of buyer, that has to exist in extractable, corroborated form: in your content, in third-party reviews, in the consensus the agent reads. Most brands are not there. Adobe found that roughly a quarter of the content on retailers’ own homepages is not readable by AI models at all. The comparison happens with or without you. Legibility decides whether it happens on your terms or on price alone.
Fragile trust is a reason to move now. When a buyer finally lets an agent buy for them, the agent reaches for the option that is best documented, best corroborated, and most clearly differentiated. That position is built over years, not bought in a quarter. Salesforce already found retailers running their own branded agents growing sales 32% faster than those without. The window is open because most competitors still treat the agent as a curiosity rather than a customer.
How to start today
Audit what the agent already says about you. Pose your five highest-value buyer questions to ChatGPT, Perplexity, and Google’s AI Mode, and note whether you are named, how you are described, and who gets recommended instead. The screen you cannot see is the one you are being judged on.
Make your real differentiation machine-readable. Fix the structure, schema, specs, and pricing logic on your top product and comparison pages, and write each page to answer one buyer question with the proof right behind it. Adobe found a quarter of retailer content is not readable by AI at all; give the agent more than a price and a feature list.
Seed the third-party sources the agent trusts. Get accurate, current information into the review sites, directories, and comparison content that LLMs cite. A stale price on a site you do not own can quietly drop you from the shortlist.
Instrument it and watch the right numbers. Split AI referrals out of organic in GA4, then track your share of agent recommendations, how that traffic converts, and whether both improve as you fix legibility. Volume will look small; quality is the signal that defends the budget.