AI Already Made the Shortlist. Did Your Brand Make It?

The number buried in Shopify's latest commerce data is not the one everyone repeated. Yes, sessions arriving from AI tools like ChatGPT, Perplexity and Gemini grew more than eightfold year over year on Shopify stores. The detail that should change how you think about your storefront is quieter. Those shoppers convert at a 50% higher rate than organic visitors, and spend about 14% more per order. Orders referred from AI grew nearly thirteenfold. The machine is not just sending traffic. It is sending pre-sold, higher-value buyers, and it is deciding who gets them.

Shopify's VP of Product, Mani Fazeli, described this as the customer journey condensing. That is the gentle version. What actually happens is that the research, the comparison and most of the persuasion now take place inside a conversation you never see. By the time someone reaches your product page from ChatGPT, the deliberation is finished. They are not browsing. They are collecting.

Why does the AI shopper spend more?

Because the hard part is already done. A person who asks an assistant for the best linen bedding under two hundred pounds, or a non-alcoholic aperitif that actually tastes like something, has outsourced the shortlisting. The model has weighed the options, dismissed the weak ones and arrived at a recommendation with the confidence of a knowledgeable friend. That is a fundamentally different visitor from someone who clicked a paid ad out of mild curiosity. They arrive warm, narrowed and ready, which is exactly why they convert higher and basket bigger.

For the brands we work with in ecommerce, hospitality and lifestyle, this is the first genuinely good news in two years of doom about zero-click search. AI was supposed to sit between you and the customer and keep them for itself. Instead, for the brands it chooses, it behaves like the best salesperson you never hired. The catch is in those three words. For the brands it chooses.

Legibility is the floor, not the edge

The DTC newsletter's advice this week was practical: audit your product schema, clean your feeds, make sure your catalogue is syndicating to the surfaces these assistants pull from. Do all of it. A store the machine cannot read is a store the machine cannot recommend, and structured, accurate product data is now table stakes rather than a technical nicety.

But legibility only gets you into the room. It makes you eligible to be considered, not certain to be picked. Plenty of brands will tidy their schema this quarter and still never surface in a single recommendation, because being readable and being recommendable are two different jobs. The first is engineering. The second is positioning.

What the machine recommends, and what it quietly skips

A language model recommends what it can describe with confidence. Ask it for the best anything and it reaches for the brands it can characterise in a sentence: the one known for a specific material, a specific founder story, a specific and repeatedly stated point of view. Brands with a sharp, consistent identity give the model something to hold. Brands that hedge, that sound like every competitor, that have no claim they make again and again, get averaged into the beige middle and skipped. The model is not being unfair. It simply cannot single out a brand it cannot summarise.

This rewards exactly the thing good brand-building always rewarded, only now the audience includes a reader that never forgets and never skims. Consistent messaging across your site, your reviews, your press and your third-party mentions is no longer just brand hygiene. It is the training data the machine uses to decide whether you are a confident recommendation or a vague maybe. The brands that win the AI shelf will be the ones that were already clear about who they are for and why they exist.

The algorithm used to decide who got seen. Now it decides who gets recommended. Those are not the same job, and most brands are still optimising for the first.

Sparked by DTC's newsletter.