AI product discovery: how AI is changing the way shoppers find products

A shopper who needs a rain jacket for a week of hiking in Scotland used to type “waterproof jacket” into your search bar and start scrolling. Now she might ask ChatGPT first, get a few suggestions, and land on your product page with an opinion already forming. In July 2026, 42% of US consumers said they’d used AI assistants for online shopping, up from 39% in March, according to Adobe. The same survey found that 66% of consumers visit a brand’s website directly to verify an AI recommendation.

That second number is the one to plan around. AI now does part of the discovery work, some of it on sites you don’t control, and a lot of shoppers still come to your site to make the final call. This guide maps the four places AI shapes product discovery, what each one needs from you, and where to put your effort first.

What AI product discovery means for ecommerce

AI product discovery is the use of AI to help shoppers find products that fit what they need. It happens in four places: your site search, your product recommendations, your own AI shopping assistant, and third-party AI agents like ChatGPT, Gemini, and Perplexity. In each one, AI works out what the shopper means and narrows a catalog down to a few good options.

If you search the term, you’ll also find a lot of product management content. In that world, “product discovery” means deciding what to build next, and that’s a different topic from this one. For the broader ecommerce definition, including the parts that don’t involve AI, see our glossary entry on ecommerce product discovery.

The four places AI now shapes discovery

A single shopper can touch all four before buying, so it helps to see them side by side.

AI-powered site search

Keyword search matches the words a shopper types to the words in your product data. When someone types “jacket for a wet hiking trip” and none of your titles contain those words, a keyword engine struggles. AI-powered search interprets the intent behind the query and maps it to the right products, and plenty of sites still have trouble with it. In Baymard Institute’s 2026 search benchmark, 43% of sites had issues with use-case searches like “party dress,” and 44% had issues with compatibility searches like “Dell laptop adapter.”

Shoppers who search are telling you what they want in their own words, so a failed search loses someone who was ready to buy. Most of the fixes live in your product data, and the site search best practices checklist walks through them.

AI product recommendations

Recommendation engines predict what a shopper might want from signals like browsing behavior, past purchases, and what similar shoppers bought. They’re good at “customers also bought” and “complete the look,” and they work without the shopper having to do anything.

Their limit is that they only know what the shopper has done so far, and a first-time visitor who’s looked at two products gives the engine very little to go on. Website personalization helps more with returning visitors. For new ones, it’s worth comparing recommendation widgets with AI sales agents.

On-site AI shopping assistants

An AI shopping assistant is a conversation on your own site. The shopper describes what they need, and the assistant asks follow-up questions and answers from your catalog and policies. When it’s connected to customer data, it can also take a returning customer’s past orders into account.

This is the one surface where AI can ask a question back. That matters most for products with fit, sizing, or compatibility questions, where the right answer depends on details the shopper hasn’t mentioned yet. If you’re evaluating one, the AI shopping assistant guide explains how assistants work and how they differ from chatbots.

Third-party AI shopping agents (ChatGPT, Gemini, Perplexity)

The surface changing fastest is the one you don’t own. Shoppers ask ChatGPT, Gemini, or Perplexity for product suggestions and compare options there before visiting a store. Salesforce reported that use of agentic search as the first step in the shopping journey grew 200% year over year, and that the rate of shoppers discovering products through brand-owned properties fell 7% between August 2025 and May 2026.

The shoppers who do click through from these tools tend to be ready to buy. In August 2026, AI-referred visitors to US retail sites converted at a rate 48% higher than non-AI traffic and generated 43% more revenue per visit, according to Adobe. Adobe also expects AI traffic to retail sites to grow 130% year over year this holiday season.

You can’t control what an agent says about you, but you can shape what it has to work with. That starts with preparing your product feed for AI agents and optimizing for AI search. For what happens to the rest of the journey, see how AI shopping agents are changing the ecommerce funnel.

What changes for shoppers

Across all four surfaces, two shopper habits keep showing up, and both affect how you plan.

They describe situations. A shopper asks for a gift for a dad who fishes, or boots for a wedding in a barn. That kind of request works well in a chat window, and as the Baymard numbers show, it still trips up a lot of search bars.

They double-check. Trust in AI shopping help is still unsettled. Gartner found that 54% of shoppers who used generative AI for a recent purchase said they had to double-check all the information it gave them, and willingness to let AI make the purchase decision topped out at 11%, even in low-stakes categories like household supplies. So AI often hands you a shopper who has a shortlist and some open questions, and your site has to answer them.

What brands need to get right

All four surfaces run on the same groundwork.

Complete, accurate product data

AI can only recommend what it can read. Fit, materials, dimensions, compatibility, and intended use need to be in your catalog and your feed in a structured form. When they’re missing, search, recommendations, your assistant, and outside agents are all left guessing.

Content that answers real shopper questions

Sizing guides, care instructions, comparison pages, and clear policies give AI something accurate to pull from. Your customer service conversations show you what’s missing, because the questions shoppers ask your team every week are the questions your site should already answer.

A conversation with a path to a person

Some purchases need a back-and-forth: a technical product, a gift, an expensive item, or a fit question the size chart doesn’t settle. An on-site assistant can handle a lot of those. When it can’t, it should hand the shopper to a person with the conversation attached, so they don’t have to start over.

Where to start

You don’t need all four surfaces working perfectly to see a difference. Start where your shoppers are getting stuck. The signals below usually point to the right first fix.

If several of these apply, start with product data. Every other surface depends on it, so fixing it helps all four.

How to measure AI product discovery

Pick a baseline before you change anything, then watch a small set of numbers for each surface you work on. Most of them already live in your analytics or service tools.

• Site search: the share of searches that return zero results, the share of shoppers who leave right after searching, and conversion rate for sessions that include a search.

• Recommendations: click-through on recommendation slots and the share of orders that include a recommended product, split by new and returning visitors.

• On-site assistant: conversion rate for shoppers who chat compared with those who don’t, the share of product questions the AI resolves on its own, and how often it hands off to a person.

• Third-party agents: sessions from AI referral sources like chatgpt.com, perplexity.ai, and gemini.google.com, plus how those sessions convert compared with the rest of your traffic.

Then look at pre-purchase questions coming into your service team. If the same questions keep showing up after you’ve improved your content and your assistant, that tells you where the next gap is.

The Gladly view: discovery should be a conversation

Search and recommendations predict what a shopper wants from a query or a click history, and for a lot of shopping, that’s enough. When a shopper isn’t sure what they need, a conversation can ask them.

That’s the idea behind Gladly Agentic Commerce. Gladly’s AI shopping assistant pulls product information from your website and draws on each customer’s conversation and purchase history, so a returning customer doesn’t get treated like a stranger. When a purchase calls for a person, it hands the shopper to your team with full context. And because Gladly is also a customer service platform, the conversation that helped someone choose a jacket is still there when they ask about their order or want to exchange it for a different size.

Balsam Hill learned why that matters. It had a separate shopping bot next to its service chat, and customers kept asking questions in the wrong one. After moving product questions into the same Gladly conversation that handles orders, nearly one in five shoppers who viewed a product in chat added one to their cart in the first 30 days, and resolution rose from 80% to 91%.

At KÜHL, AI took over repetitive questions so the team had time for fit advice and upsells, and revenue per call went up 120%.

Gladly works alongside your site search and recommendation engine and covers the moments when a shopper needs to talk a purchase through.