Financial TimesAmericas' Fastest-Growing 2025
Certified PartnerOdoo Gold · Adobe Gold Solution · Shopify Premier partnershipsOdoo Gold · Adobe Gold · Shopify Premier partnerSee all
One business dayA senior architect replies to every inquiry · +1 (631) 897-7276Architect reply within one business dayLet's talkCall us
Inc. Regionals 2025Fastest-growing private companies in AmericaInc. Regionals 2025 honoreeSee awards
Americas' Fastest-Growing Companies 2025Recognized by the Financial TimesFT Fastest-Growing Companies 2025See the rankingRanking
4.8 / 5 on Clutch265 verified client reviews4.8/5 on Clutch · 265 reviewsRead the reviews
4.8 · 265 Clutch reviews
US United States
June 19, 202315 min readUpdated on October 25, 2026

The AI shopping experience: AI does the research, your store has to confirm it

· Chief Executive Officer
The AI shopping experience: AI does the research, your store has to confirm it

A shopper asks an AI assistant for a carry-on that fits under a regional jet seat. It compares six bags, flags two with bad wheel reviews and sends her to your product page. She doesn't browse. She checks the size, the price and the return policy, then buys or leaves in under a minute.

That scene is the AI shopping experience most retailers already have, whether they planned for it or not. The comparing happens upstream, in Google's AI Mode, ChatGPT or Amazon's Rufus, and your store's job shifts from persuading to confirming. Our position is that much of the spending on AI in online shopping lands at the wrong end. Retailers buy a chat widget when the outcome is mostly decided by product data and by that last minute on the product page.

This piece makes that case for ecommerce, digital and customer experience leaders at US retailers and brands. It follows the path a shopper now takes, says where generative AI earns its place, and ends with the order we'd fix things in.

What counts as an AI shopping experience

An AI shopping experience is any part of buying online where software interprets what the shopper means and answers, instead of offering a fixed menu. That covers conversational product search, AI shopping assistants, personalized recommendations, visual and voice search, virtual try-on, agents that track prices or check out, and support chatbots after the order.

For years, AI in ecommerce worked out of sight. Machine learning models trained on historical data reordered a carousel or scored a payment, and the shopper never spoke to them. Now shoppers talk to AI directly, in their own words, and often on someone else's platform. AI has moved from the stockroom to the front door.

Most of that front door runs on generative AI: large language models and image models that write answers, summaries and pictures. Older models predict ("this shopper will probably like these boots"). Generative models produce ("a paragraph comparing these three boots for wet weather"). Our guide to generative AI vs machine learning covers the difference in depth. It matters later, when you decide what to fund.

The research now happens in someone else's chat window

The biggest change isn't on your website. Adobe Analytics reported that traffic to US retail sites from generative AI tools rose 693.4% year over year over the 2025 holiday season, November 1 to December 31. Adobe is careful to add that the base of users is still modest. The same report puts US online holiday spend at a record $257.8 billion. AI referrals are a fast-growing slice of a very large pie, not the pie itself.

Salesforce's holiday data points the same way from another angle. In January 2026 it reported that AI and agents drove 20% of retail sales over the 2025 holidays, or $262 billion globally. Shoppers referred from AI search tools converted 9 times more often than those referred from social media. That's Salesforce's own model of AI influence, not a count of sales AI alone caused. The direction holds anyway: shoppers who arrive through AI tend to arrive ready to buy.

That changes what your product page is for. A shopper coming from an assistant has usually compared you with two or three alternatives already. They land deep, with one narrow question. If the first screen doesn't answer it, the assistant has already shown them somewhere else that might.

What the assistant knows about you comes from product feeds, structured data and the text on your pages. Google says its Shopping Graph holds more than 50 billion product listings, with 2 billion updated every hour. That graph feeds shopping results in AI Mode and the Gemini app. ChatGPT reads merchant feeds. Shopify's help center says eligible products from stores selling to US customers can appear there. In that setup, ChatGPT acts as a "discovery-focused referrer." Rufus answers from Amazon's own catalog, reviews and listings.

So if your titles, sizes, materials and stock levels are vague or stale, the assistant skips you or describes you wrong. Our piece on agent-readable catalogs goes further into the feed work.

On your own site, shoppers ask instead of filter

The same habit follows shoppers onto retail sites. They type a need instead of a keyword, asking for "a rug that hides dog hair" rather than "rug, polypropylene, 5x8" in a filter. It's the question they'd put to a store associate, and the largest retailers have built assistants to answer it.

Assistant What it does for shoppers Source
Amazon Rufus Answers product questions, compares items, tracks prices and can buy from other stores through Amazon's "Buy For Me" feature Amazon Q4 2025 results
Walmart Sparky Helps customers search for items, sums up reviews and suggests products for an occasion Walmart, June 2025
Google AI Mode Conversational product search, comparison tables and price tracking across retailers Google, November 2025

Amazon offers the clearest numbers. In its Q4 2025 earnings report, it said Rufus was used by more than 300 million customers in 2025. It also said Rufus helped deliver nearly $12 billion in incremental annualized sales. Read that carefully. It's Amazon's own measure, from a company with Amazon's catalog depth and review volume, and a chat widget on a mid-size store won't produce anything like it. The useful lesson is narrower: conversational search works when it sits on top of deep, accurate product data and real reviews.

Recommendations, the oldest form of on-site AI, carry a warning of their own. They use purchase history and browsing behavior to put relevant products in front of each shopper. Done well, they feel like a store that remembers you. Done badly, they push. A 2025 Gartner survey of 1,464 buyers and consumers found that personalized marketing created a negative experience for 53% of customers. Those customers were 3.2 times more likely to regret a purchase, and Gartner tied it to shoppers feeling overwhelmed or rushed. The answer isn't less personalization. It's personalization built on what customers tell you (sizes, preferences, past returns) instead of guesses that feel creepy.

Try-on and the virtual shopping experience

Fit is the most physical gap in online shopping. Will the jacket hang right? Will the sofa crowd the room?

Google launched virtual try-on in the US in July 2025, letting shoppers upload a photo to see apparel on themselves. When it announced the tool in May 2025, Google said it covers billions of apparel listings. For retailers, the virtual shopping experience takes three forms. Photo-based try-on handles apparel and accessories. AR room placement uses the phone camera for furniture and decor. Visual search lets a shopper upload a photo and find similar products when they can't describe what they want.

Each one removes a different doubt: whether a style suits, whether it fits the space, what the thing is even called. Try-on images are still AI estimates, not fitting rooms. Keep size charts and real-customer photos beside them, or your returns will report the gap later.

Checkout: agents want to buy, shoppers want the final say

This is where AI in online shopping moved most in 2026, and where the hype ran ahead of the shoppers.

In late 2025, ChatGPT's Instant Checkout let US users buy from some merchants inside the chat. Walmart announced in October 2025 that it would sell through it. Then in March 2026, OpenAI changed course. It refocused ChatGPT on product discovery and let merchants use their own checkout. Shopify's help center now puts it plainly: ChatGPT users complete their purchase on the merchant's own online store checkout.

Google went the other way. In November 2025 it introduced agentic checkout. When a tracked item hits the shopper's price, Google can buy it on the merchant's site with Google Pay, once the shopper confirms. It runs on the Universal Commerce Protocol, an open standard that keeps the merchant as the merchant of record. In September 2026, Google said UCP already enables direct checkout for hundreds of thousands of brands and retailers.

Shoppers explain the split better than the platforms do. In a May 2026 Gartner survey of 322 US consumers, only 11% were willing to let AI make purchase decisions, even for low-stakes items like personal care. More were open to AI narrowing their choices: 31% for household supplies and 28% for personal electronics. The same release cites a separate Gartner survey from late 2025. In it, 54% of US shoppers who used AI said they had to double-check the accuracy of everything it told them.

So the working model is simple: AI researches and the shopper decides. That puts the weight back on the moment the AI hands the shopper to you, and that moment still fails for old reasons. Baymard Institute puts the average documented cart abandonment rate at 70.22%, across 50 studies. Among US shoppers, the top reason (40%) is extra costs like shipping, tax and fees that show up too late. Forced account creation (18%) and long checkouts (17%) follow. An assistant can deliver a ready buyer to your product page. A surprise shipping fee still sends them back.

Support is where trust breaks fastest

Chatbots for routine questions were where most online retailers started with AI. It's still where interactions go wrong quickest, and Gartner's 2026 customer service research is blunt about it.

In an August 2026 release, 50% of customers said GenAI makes their interactions easier. Yet 87% said it's essential to offer a way to reach a human agent. A September 2026 release found only 27% would try a chatbot again after a bad experience. Only 7% had used a chatbot or digital assistant in their most recent service interaction. And in a July 2026 release, customers were about 3 times more likely to take a service issue to tools like ChatGPT or Gemini than to a company's own chatbot.

A support bot gets one chance. If it reads a damaged-delivery complaint as a tracking request, the shopper gives up and probably won't try the bot again.

What works is narrow. The bot handles order status, returns and policy questions it can answer from real data. Because it knows the order, it answers about that order instead of reciting an FAQ. When it isn't confident, it passes the case to a person with the context attached, so the customer never repeats themselves. Design a conversational AI support flow around that handoff, not around deflection. The most underused version is proactive: a message that a delivery will be late, sent before the customer asks, beats the fastest chatbot reply.

Generative AI in ecommerce: where it earns its keep

Generative AI in ecommerce means models that create content (text, images, answers) inside the buying and selling process. It does four kinds of work: it writes, summarizes, converses and generates images. These use cases are already running at large retailers and platforms, each with an example you can check:

Use case What it does Live example Check before you scale
Product content Drafts titles, descriptions and attributes from a few inputs Amazon's listing tools turn a seller's short description into listing content for review Every spec, size and claim
Catalog enrichment Rewrites technical names and descriptions in shopper language Adobe Commerce catalog enrichment (beta), with an editable preview Review before you apply
Review summaries Condenses customer feedback into a short paragraph on sentiment Amazon review highlights, built only from verified-purchase reviews Real reviews only; keep the negatives
Shopping assistants Answers natural language queries and compares options Amazon calls Rufus a "generative AI-powered" assistant; Walmart has Sparky Grounding, so it can't invent specs
Imagery and try-on Generates backgrounds, variations and try-on previews Google's virtual try-on; Shopify media generation Real photos alongside
Support replies Drafts answers from store policies Shopify Inbox suggests instant answers and replies A visible route to a person
Merchant copilots Answers "what sold last week" and drafts copy for ecommerce teams Shopify Sidekick writes ShopifyQL queries, weekly summaries and product copy The numbers behind the summary

None of these is a moonshot. Each takes a task your team already does by hand and produces a first draft. The value comes from the review step you design around it, not from the model.

The verified generative AI for retail examples share a pattern, once you set aside the ones that only live in vendor slide decks. Amazon uses generative AI on both sides of its marketplace: Rufus and review highlights for shoppers, listing tools for sellers. All of it rests on a deep catalog and verified reviews. Platforms now ship the same capabilities as standard. Shopify says many of its AI tools come "at no additional cost," though availability varies by plan, device and language. On Adobe Commerce as a Cloud Service, the Product Visuals option includes access to Adobe Firefly and Adobe Express. When every retailer can reach the same models, the advantage sits in the data those models read.

Not everything sold as retail AI is generative, and the difference matters for budgets. Demand forecasting, dynamic pricing, fraud scoring and customer segmentation are mostly predictive machine learning. They still belong in the same plan, because they decide whether the item an assistant recommends is in stock, fairly priced and shipped on time. Put guardrails on pricing, since shoppers comparing through assistants notice erratic swings. Tune fraud models so good customers aren't declined the moment an AI hands them over.

Building the business case

Start where generative AI replaces manual work you can measure:

  1. Catalog content. Count the hours spent writing and fixing product descriptions, and the share of SKUs without accurate ones.
  2. Site search. Measure queries that return zero or poor results, and how those sessions convert.
  3. Support. Find the customer inquiries a policy-grounded draft could answer. Track resolution and satisfaction, not deflection.

Pick one, set a baseline and pilot it on a single category. A pilot without a before-and-after number won't survive the next budget review. For the content tools themselves, our roundup of AI product description generators compares the options.

Switch on, buy or build

Once you know the use case, decide where the capability comes from. Most retailers end up using all three routes for different jobs.

Option Best when Watch out for
Switch on what your platform ships The job is generic: product copy, image edits, support drafts Features and limits vary by plan
Buy a specialist app One job matters a lot (search, reviews, support) and speed counts Another vendor holding customer data
Build a custom layer Answers depend on your own data: complex catalogs, B2B pricing, returns rules You own the guardrails and upkeep

On Shopify, our guide to what Shopify's AI features do goes feature by feature. On Adobe Commerce, our Adobe Commerce generative AI guide separates what's live from what's still in beta.

Whichever route you pick, the plumbing looks the same. You need clean, structured product data behind APIs your tools can read. You need clear terms with your model provider on where prompts and customer data go, and whether they're kept. The model has to be grounded in your catalog, orders and policies, not its training data. Keep a test set of 50–100 real customer questions with known answers and rerun it after every change. And log and review anything a customer will read.

What to fix first, in order

Work from the data out, not from the chatbot in. One warning before the list. If your site is slow, your shipping costs surprise people or your photos are poor, AI will route more shoppers to the same problems faster. And if you sell 40 products that shoppers already understand, good filters and a clear size guide will beat a conversational assistant. They'll cost less to run, too.

  1. Product data. Write titles and attributes for a question, not a filter. Put dimensions, materials, compatibility and care in plain text, and keep stock and prices current in your feeds. This one change helps every AI surface at once: Google, ChatGPT, your site search and your recommendations. Generative AI won't rescue a thin catalog. Ask a model to describe a product it knows nothing about and it will write something confident anyway.
  2. Product pages that answer fast. Shoppers from AI assistants land deep. Put price, delivery date, total cost, return policy and size guidance where they can see it without scrolling. Add a short FAQ to your high-return products.
  3. Conversational search where the catalog is complex. If shoppers struggle with filters (parts, apparel fit, home goods by room), a natural-language search bar helps more visitors than a chat widget. Everyone uses search; few open a chat. Ground it on your catalog and reviews so it can't invent specs.
  4. A narrow support bot, handoff first. Start with the 5–10 questions behind most of your tickets. Offer "talk to a person" from the first message, and measure resolution and satisfaction.
  5. Personalization from stated preferences. Use sizes, stated preferences, purchase history and past support conversations. If you run a customer data platform, check that those preferences actually reach it. Test pressure-heavy prompts before you scale them; Gartner's regret data is reason enough.
  6. The AI channel, tracked like any other. Tag AI referral traffic separately in your analytics and watch its conversion, average order value and return rate. You can't improve a channel you can't see.

If we were starting on your store, step 1 would come before anything else: run your feed against the questions shoppers actually ask and see which answers come back wrong. Our AI and machine learning team can run that audit and carry steps 2–4. Our generative AI development team builds the custom layer when platform tools run out. As an Adobe Gold Solution Partner specialized in Adobe Commerce and a Shopify Premier Partner, we can do that work inside the platform you already run.

The rules don't change because a bot said it

AI doesn't change your obligations. It adds new ways to break them. California's CCPA gives residents the right to know, delete, correct and opt out of the sale or sharing of their personal information. That reaches chat transcripts and photos uploaded for try-on. The FTC's rule on fake reviews took effect in October 2024. It covers reviews that misrepresent being from someone who doesn't exist, AI-generated ones included. If you summarize reviews with AI, summarize real ones and keep the negatives.

Accuracy carries liability too, because generative AI writes fluent text whether or not it's true. In Moffatt v. Air Canada (February 2024), a Canadian tribunal held the airline responsible for its chatbot's wrong advice on bereavement fares. The tribunal said the airline "is responsible for all the information on its website," whether it comes from a static page or a chatbot. A retailer whose bot invents a refund rule should expect the same question. Ground answers in your own policy pages and product data, log what the bot says, and have a person check AI-written descriptions and AI-edited images before they go live.

Some purchases need people outright. High-consideration items, custom orders and anything touching health or safety call for someone who can say "it depends" and mean it.

FAQ

Will AI shopping agents replace online stores?

Not in the near term. A 2026 Gartner survey found only 11% of US consumers were willing to let AI make purchase decisions. OpenAI also moved ChatGPT purchases back to merchants' own checkouts in March 2026. Agents are becoming a major source of discovery, but stores still handle the final decision, payment and fulfillment.

What are some examples of generative AI in retail?

Amazon uses generative AI for review highlights, the Rufus shopping assistant and seller listing tools. Walmart's Sparky sums up reviews and suggests products, and Google's virtual try-on generates apparel previews from a shopper's photo. Shopify and Adobe Commerce now ship generative features for product copy, images and catalog enrichment.

How is AI changing the online customer experience?

AI is moving the research stage of shopping off retail sites and into assistants, so shoppers arrive later in their decision with narrower questions. On-site, it replaces filters with plain-language search and adds visual tools like try-on. After purchase, it handles routine support, though Gartner's 2026 research shows customers expect an easy path to a human.

Back to the carry-on

The carry-on shopper didn't need persuading. She needed the size, the real price and the return policy, fast. Get that minute right and the assistants will keep sending you the next shopper.

If you're weighing where AI belongs in your store, our ecommerce team can start with your catalog. Tell us what you sell.

ShareLinkedInXEmail

Related reading

Putting AI into production?

Explore AI and ML services