Retailers have spent years optimizing for what shoppers do on their sites. But a new consumer survey shows that the most important phases of the buying journey, including the final decision-making, are happening before prospective customers even get there. Eight in 10 shoppers under 44 now use LLMs as their primary shopping vector. And most say they have already made a decision, based on what AI tells them, before they visit a retailers’ website. 

In The new discovery engine: How consumers are using AI to find, trust and choose brands, and what’s at risk for those they never see, Rithum surveyed 1,046 online shoppers in the U.S. and U.K. and found that large language models, or LLMs, are trusted partners in how people research products, compare options and decide where to buy. Among AI-active shoppers, more than 90% use LLMs for product research and comparison, and 53% use them to choose the retailer. And one in five have already bought from a brand they'd never heard of—simply because AI recommended it.

Discovery is starting before the site visit

Retailers have long focused on shaping shoppers’ impressions. But AI is leveling the playing field in ways that make brand recognition and hard-won loyalty less reliable than ever. According to Rithum's survey, 13% of shoppers say they're more likely to switch retailers or products based on AI suggestions. And the shoppers most likely to act on those recommendations without a second look are the high-income, high-intent buyers that retailers compete hardest to win. 

By the time one of those shoppes land on a product page, LLMs have already done most of the work. Shoppers are using them to decide which products deserve a click, which retailers are worth visiting and which brands are worth their time. That makes the product page less of a discovery surface and more of a place where a choice that has already started to form gets confirmed or lost.

This is already common among shoppers with strong spending power and years of buying ahead of them. Over the last three months, AI shopping adoption reached 80% among shoppers ages 18 to 27 and 80% among shoppers ages 28 to 43. Among households earning $100,000 to $150,000, it reached 84%, according to the survey.

New brands are getting more room

Once AI starts doing more of the filtering, familiar names no longer have the field to themselves. In the survey, 19% of shoppers said they now buy from brands or products they had not heard about before thanks to LLMs, and 13% said they are more likely to switch retailers or products after using an LLM. That creates an opening for newer brands and raises the pressure on assortments that once relied more heavily on recognition alone.

The survey suggests shoppers care less about brand familiarity in that moment than whether the recommendation makes sense. Nearly half said a clear explanation of why a product or brand was chosen would do the most to increase trust in an AI recommendation. When the answer feels specific and grounded, shoppers are more likely to follow it.

Product information is now part of the pitch

That brings the conversation back to the work that retail teams already know well. AI builds recommendations from product information: materials, dimensions, compatibility, intended use, price and availability. When those details are complete and consistent, LLMs have more to work with. When product details are thin, stale, or contradictory, the recommendation gets weaker (or doesn’t get made).

A known brand still has value, but shoppers responded strongest to recommendations that feel well explained; not a name they recognize. Nearly half said a clear explanation of why a product or brand was chosen would do the most to increase trust in an AI recommendation. In an LLM-driven shopping experience, the brands most likely to win teh sale aren’t those that are familiar, but those that give AI the most details to work with.

Trust still rises and falls on the basics

In the survey, 67% of shoppers said pricing is the most important detail for AI to get right in a shopping recommendation. When an LLM gets product information wrong, 58% said trust in the product or brand drops, and 16% said they leave the purchase altogether.

Retailers also get less time to recover once a recommendation takes hold. Thirty-two percent say they spend less time browsing other sites after using LLM tools. When they do verify what they see, only 5% start with a retailer or brand website. Most go first to search engines, online reviews, or friends and family.

The advantage goes to retailers that make products easy to understand

The answer here has less to do with chasing the next AI tactic than with getting the fundamentals right. Retailers still need accurate pricing, current availability and product details complete enough for AI to describe an item clearly and for shoppers to trust what they see. That is the throughline in every customers’ response to how they use LLMs, and it is only going to become more important as AI takes on a larger role in product discovery.