AI in Retail: What 19 Major Retailers Actually Deployed

The biggest retailers in America have the least to say

Of the 19 retailers checked, 7 make no verifiable AI claim at all. Costco mentions artificial intelligence exactly twice in its annual report and both are risk factors. TJX calls its markdown engine specialized computer systems. Aldi, Publix, Wegmans and Trader Joe's have not made one primary-source AI claim between them, across thousands of stores.

Costco's own filing concedes that competitors have faster adoption of artificial intelligence than it does. That is a company choosing not to, in writing.

The loudest AI stories quietly stopped being updated

Target announced a staff chatbot chainwide in 2024 and a generative gift finder for the 2024 holidays. Neither appears in anything Target has published in 2026, including its own June 2026 fact sheet listing where its AI actually lives. Best Buy's 2024 support assistant was never confirmed live and its announcement page still reads expected to launch late summer 2024.

None of that proves those tools are dead. It proves nobody said - and on a page built from primary sources, an unrepeated announcement is not a deployment.

The most expensive AI story in retail is a write-off

Kroger closed three robotic fulfillment warehouses, canceled a fourth site before it opened, and recognized $2,497 million in impairment and related charges - roughly $2.5 billion - including about $350 million paid to Ocado just to terminate the arrangement. Its own filing says it identified opportunities to optimize the network by closing facilities which had not met operational or financial expectations.

Walmart Policy commitment

Sparky and pricing commitments. In a September 25, 2026 letter, Walmart's chief executive committed that Walmart will not set different prices based on who you are, and that the information shoppers share with Sparky will not be used to raise a price or hide lower-priced options that meet their needs.

A written commitment, not a metric. The letter is signed by John Furner, President and CEO, and sits on Walmart's corporate site under Everyday Affordability. Its bold line is 'We price the product, not the person.' It also says Walmart's people will continue to oversee pricing and that the company will monitor and test its technology against these commitments. It extends the July 1, 2026 pricing page cited elsewhere on this page. What is new is that the shopping assistant is named inside the pricing pledge.

Source: Walmart, letter from our CEO · Sep 25, 2026

Costco No verified AI

Nothing. Costco's 2026 filings contain no AI disclosure at all.

The strongest verified silence on this page, and it got quieter in 2026. The Q4 FY2026 earnings release of September 24, 2026 also contains no mention of AI or artificial intelligence. Costco's Q3 FY2026 quarterly report mentions artificial intelligence zero times - the only technology reference in the document is the boilerplate phrase investments in technology, inside the forward-looking-statements paragraph. Across Costco's entire filing history the phrase artificial intelligence appears in exactly two documents, both annual reports, both in Risk Factors, and in neither case as a claim to be using it. The one substantive thing Costco has ever said on the subject concedes the opposite - that some competitors have greater technology capabilities, including the faster adoption of artificial intelligence, than it does.

Source: Costco Q4 FY2026 earnings release, Form 8-K exhibit (SEC) · Sep 24, 2026

Target Shipping

Photo Search, AI Review Insights, Buy Again and Continue Shopping. Four AI features inside Target's own app rather than inside somebody else's - image search, review summarization, reordering and session resumption.

This is the entry that moves Target off the page's earlier read that its consumer AI was mostly Target showing up inside Google, Microsoft and OpenAI products. On September 8, 2026 Target named four of its own, with launch timing attached to each: Photo Search launched in August and lets a guest photograph a lamp or a throw pillow and find similar items rather than describing them in words; AI Review Insights launched in June and groups review text into themes such as suction power, durability, weight and ease of use; Buy Again surfaces frequently purchased items and launched last year; Continue Shopping reconnects a guest with recently viewed products and launched last fall. It is also the closest thing on this page to a controlled experiment in how a retailer talks about results, which is why it is worth quoting rather than paraphrasing. Three of the four carry an outcome and none carries a number. Review Insights: early results show the feature is helping drive conversion and add-to-cart activity while reducing decision fatigue. Buy Again: has helped drive strong year-over-year growth in purchases. Continue Shopping: has driven strong guest engagement and more add-to-cart activity. Photo Search, the newest, carries no outcome claim at all. Read the shape rather than a verdict: a retailer that publishes specific launch months and specific feature names, then describes every result in adjectives, is telling you the features are real and the measurement is either early or private. Set against that, the standard Target's own new AI executive set for himself. On August 11, 2026 Target announced that Chandhu Nair would join on August 24 as senior vice president and its first chief AI officer, with Purvi Shah moving to senior vice president of UX. Asked what a coordinated approach makes possible, Nair answered: the measure of success will not be how much AI we deploy, it will be the difference it makes for Target's growth, the guest experience and our team's ability to do their best work. He was two weeks into the job when the feature article published. Hold his standard next to the four adjectives above and draw your own conclusion.

Source: Target newsroom, AI-powered shopping features · Sep 8, 2026

Walmart Shipping

Sparky. A shopping assistant reached through an Ask Sparky button in the Walmart app, which answers questions and can act on your behalf.

The most useful thing Walmart has said about Sparky is a comparison, not a count. On its Q2 FY27 earnings call in August 2026 its chief executive reported that the number of customers using Sparky is up 70% from last year, and that the customers and members who use Sparky for shopping spend 40% more per order than others who don't. Read that second one carefully, and note what it is not: Walmart reports the gap, it does not claim Sparky caused the bigger basket, and people who opt into an AI assistant are unlikely to be an average sample. Walmart still publishes no absolute Sparky user number, so there is no denominator here either.

Source: Walmart Q2 FY27 earnings call transcript · Aug 20, 2026

Walmart Shipping

Supply-chain automation. Automated distribution and fulfillment centers that pick and sort with far fewer hands.

On its Q2 FY27 earnings call in August 2026, Walmart's chief financial officer said 3,100 of its US stores are now served with some level of automated freight, and that it is processing over 50% of its eCommerce fulfillment volume through automated facilities. The quarter before, Walmart's Q1 FY27 summary put the automated share of eCommerce fulfillment center volume at about 50%. Read the two together carefully: over half against about half is roughly flat across the two quarters, and the wording changed between them, so they may not measure exactly the same thing. The 3,100 stores figure is the new number, and it is the one to watch next quarter.

Source: Walmart Q2 FY27 earnings call transcript · Aug 20, 2026

Target Shipping

AI inside somebody else's app. Target's current consumer AI is mostly Target showing up inside Google, Microsoft and OpenAI products rather than inside Target's own.

Target reports AI-driven traffic up 2,000% year on year in a single quarter, and claims to be the first mass retailer live across three AI platforms - Google Search and AI Mode and Gemini, Microsoft Copilot, and a Target app inside ChatGPT. Take the 2,000% for what it is: a percentage with no starting number attached, which is what companies publish when the base is small. What the fact sheet leaves out matters as much. Store Companion, the staff chatbot Target announced chainwide in 2024, is not mentioned. Neither is the Bullseye Gift Finder. Both were dropped from this page in August 2026 because no Target source published in 2026 confirms either still exists. What survives in the filing is deliberately vague: Target says it expanded the use of AI-powered tools to simplify work for store and headquarters teams, and names nothing.

Source: Target fact sheet, conversational AI · Jun 18, 2026

Walmart Shipping

Digital shelf labels. Electronic price tags that replace the paper ticket on the shelf edge.

Roughly 2,300 US locations, with Walmart saying chain-wide rollout is expected within the next year. Walmart does not call this AI. Separately, in July 2026 it published a standing pricing policy that names and disavows the practices people fear: it says it does not use surveillance pricing or surge pricing, and that DSLs change how we show price, not how we determine price.

Source: Walmart newsroom, digital shelf labels · Mar 2, 2026

Amazon Changed

Amazon Go and Amazon Fresh. Amazon's own checkout-free and AI-forward grocery store banners.

The biggest walk-back in retail AI, and Amazon stated it plainly on its own newsroom: it had not yet created a truly distinctive customer experience with the right economic model needed for large-scale expansion, and had made the difficult decision to close the stores. The technology outlived the stores built for it - Just Walk Out now runs in over 360 third-party locations across five countries, and in more than 40 North American fulfillment-center breakrooms.

Source: Amazon newsroom, store closures · Jan 27, 2026

Aldi US No verified AI

Nothing. No headline in Aldi US's corporate news archive between 2020 and 2026 mentions AI.

Roughly 60 headlines were read across six years. The only tech-adjacent items are a delivery partnership and curbside pickup. Aldi's own description of how it operates is the most on-brand sentence in this research: the rule around here is if it isn't necessary, we don't do it. Period.

Source: Aldi US newsroom · Aug 18, 2026

Trader Joe's No verified AI

Nothing, and not by accident. There is no online ordering, no loyalty program, no app scanning and no checkout technology.

Trader Joe's entire public site index is recipes, stories, announcements and search. There are no cart, checkout, account, loyalty or app URLs at all - and no product pages or store-locator pages either. In a category racing to automate, one of the most successful grocers in the country has declined to participate in any of it.

Source: Trader Joe's site index · Aug 18, 2026

Wegmans No verified AI

Barcode scanning in the Wegmans app. Scanning items with your phone while you shop.

The app's FAQ today describes barcode scanning only for adding items to a list or saving them for later. No shop-and-scan or self-checkout function appears anywhere in its twelve entries. Stated carefully: this documents absence today rather than a dated removal, because Wegmans published no announcement either way.

Source: Wegmans app FAQ · Aug 18, 2026

Kroger Changed

Ocado automated fulfillment network. Robotic warehouses built to fill online grocery orders without people walking the aisles.

The most expensive documented failure in retail automation, and Kroger writes it down in its own annual report rather than burying it. It closed facilities in Pleasant Prairie, Frederick and Groveland in January 2026 and canceled a fourth site in Charlotte, recognizing impairment and related charges of $2,497 million - roughly $2.5 billion - which included a cash termination payment to Ocado of about $350 million. Kroger's own sentence does the work: it identified opportunities to optimize its automated fulfillment network by closing facilities which had not met operational or financial expectations. Both the corporate phrasing and the admission sit in the same line.

Source: Kroger FY2025 annual report (SEC) · Mar 31, 2026

Publix No verified AI

Nothing shipped. Across roughly 1,400 stores, the only substantive mention says the company assesses the use of artificial intelligence.

Publix mentions AI four times in the filing, and only one spells the term out in full. That sentence says the company continues to implement strategic business and technology initiatives, and assesses the use of artificial intelligence - a statement about considering it, not a claim to be using it. Publix's Discount Finder is described by Publix as powered by a third-party pharmacy pricing service with no AI claim attached, so it should not be counted as one.

Source: Publix FY2025 annual report (SEC) · Mar 2, 2026

Kroger Shipping

Sage. An employee app rather than a shopping assistant: one place for store staff to check a shift schedule, request time off, set availability and view pay stubs.

The clearest figure any retailer on this page has attached to a named AI product, and it is about staff rather than shoppers. Kroger's vice president of human resources described Sage on a panel with IBM at the National Retail Federation's Big Show in January 2026. The company began workshopping and developing it in early 2024 and, five months later, the platform was capable of 30 basic use cases drawn from data Kroger already had. In his words: today, Sage has over 95 use cases and is actively used by up to 150,000 associates. Store leaders get real-time labor data as well as shift changes, pending punches and time-off approvals, and employees can ask Sage direct questions about company policies and training completion. His summary of why it works is worth keeping: functional, tangible, real-time, 24/7. One distinction that gets lost constantly - this is not the Gemini shopping assistant Kroger announced at the same conference. They are separate products discussed at separate panels.

Source: Grocery Dive, reported from the NRF show floor · Jan 13, 2026

Lowe's Shipping

Mylow and Mylow Companion. A customer home-improvement advisor built with OpenAI, plus an associate version on store handhelds.

The rare AI deployment with a real usage number attached, and Lowe's put it in an SEC-filed document rather than a press release: Mylow reached 5 million associate questions as of its Q1 2026 results, across a fleet of 1,761 stores. The newer figure comes from the Lowe's-hosted Q2 call transcript of August 19, 2026, where CEO Marvin Ellison says Mylow has supported over 25 million questions from customers and associates, so that count mixes shoppers with employees. On the same call he says the conversion rate for online customers who use Mylow is triple that of customers who do not use the tool. That compares shoppers who chose to open the assistant with shoppers who did not, and the transcript gives no underlying rates, so it is not a measure of what the assistant added. The 5 million is the narrower associate-only count. Lowe's also claimed at launch in May 2025 to be the first retailer to implement this kind of technology at scale - treat that as the company's own characterization from 2025, not as a current fact, because no 2026 source repeats it.

Source: Lowe's Q2 2026 earnings call transcript (Lowe's-hosted) · Aug 19, 2026

Home Depot Shipping

Magic Apron. A proprietary suite of generative AI tools that answers how-to and which-product questions.

Magic Apron outgrew the product page it started on. Home Depot now describes it as a conversational assistant across its digital platforms, plus a store version that knows which aisle you are standing in. On the August 2026 call an executive gave the first usage figure the company has ever offered: we're getting millions of questions per month now. Note what that is and is not - it is a volume of questions, not a count of people, and Home Depot gave an order of magnitude rather than a number. Worth knowing where it runs: the models are Google Cloud's Gemini, so Home Depot is the application layer on somebody else's model.

Source: Home Depot Q2 FY2026 earnings call transcript · Aug 18, 2026

IKEA Shipping

Billie. A service bot that handles simpler customer inquiries before a person picks up.

Billie is still live and now runs on generative AI, confirmed by IKEA's own chief data and analytics officer in May 2026: if you have a question for IKEA, you will see Billie popping up in the corner, and that is based on generative AI and the latest models. What IKEA has not done is update the numbers. Its last published performance figures are dated 2021 to 2023 - roughly 47% of inquiries received resolved, about 3.2 million interactions, nearly EUR 13 million saved, and 8,500 call-center co-workers reskilled into roles such as remote interior design. Ingka's FY2025 annual summary does not mention Billie at all. Treat those figures as three years old, because they are.

Source: Ingka Group newsroom, chief data and analytics officer interview · May 13, 2026

Home Depot Shipping

Sidekick. An associate app that prompts staff about which shelves need attention.

Home Depot's annual report says Sidekick gives its US store associates that prompt, but publishes no store count for it against the 2,359 stores it operated at the end of fiscal 2025. Its on-shelf availability improvements are described qualitatively in the filing rather than quantified, so treat any specific basis-point figure you see elsewhere with care.

Source: Home Depot FY2025 annual report (SEC) · Mar 18, 2026

TJX No verified AI

Nothing, across 5,214 stores and approximately 377,000 associates. Its markdown and allocation engine is called specialized computer systems.

The cleanest no in retail. Artificial intelligence appears four times in the annual report and all four sit in Risk Factors, two of them as confessions. Machine learning appears zero times. TJX states the risk in its own words: we could be at a competitive disadvantage if, over time, our competitors are more effective than we are in their use and integration of rapidly evolving technologies, including artificial intelligence.

Source: TJX FY2026 annual report (SEC) · Mar 31, 2026

Gap Inc. Announced

Conversational discovery at Old Navy and Banana Republic, Alta Daily, Daydream. Conversational AI shopping tools where a shopper describes what they want in plain language.

Announced October 5, 2026 on Gap Inc.'s 30th anniversary of gap.com. Old Navy and Banana Republic are each adding a conversational discovery tool. The Gap brand is working with an app called Alta Daily, which turns style inspiration into curated looks and lets shoppers try them on an avatar, and Daydream, an outside AI fashion discovery platform, lets shoppers describe what they need in everyday language. Retail Dive describes Alta Daily and Daydream as third-party tools. Gap's release gives no launch dates, no user numbers and no sales figures, so this is an announcement with nothing to measure yet.

Source: Gap Inc. newsroom · Oct 5, 2026

Sephora Shipping

AI Beauty Chat. A conversational assistant that builds routines and recommends products.

Live on Sephora's own site as a round-the-clock assistant, though Sephora does not name the vendor behind it. Two older AI-branded features have quietly gone from the website: Color IQ has no URLs in Sephora's live site index and is not among its bookable in-store services, and Virtual Artist redirects to the homepage. Its ChatGPT storefront is described by Sephora itself as piloting.

Source: Sephora AI Beauty Chat · Aug 18, 2026

Macy's Announced

AI in shopping and operations. Stated plans for product discovery, agentic commerce and personalization, plus internal use in customer service, supply chain and coding.

Macy's does describe real intent in its own words - applying artificial intelligence to how customers discover, shop and engage with our brands through intelligent shopping assistance, agentic commerce and hyper-personalised experience - and separately says we use AI as a tool. What is missing is any named product, any named vendor and any number. Its most recent quarterly release credits store remodels rather than AI for its results.

Source: Macy's FY2025 annual report (SEC) · Mar 27, 2026

Ulta Beauty Changed

GLAMlab. Virtual try-on that puts makeup on your camera image or photo.

The feature is live. What changed is the claim, not the product: in its March 2026 annual report Ulta describes these as engaging experiences including virtual try-on and analysis capabilities, and neither artificial intelligence nor augmented reality appears anywhere in that filing - though both appear in every one of its annual reports from 2018 through 2024. The feature survived; the AI attribution did not.

Source: Ulta FY2025 annual report (SEC) · Mar 26, 2026

Best Buy Shipping

Agentic commerce partnerships. Best Buy's catalog showing up inside ChatGPT, Google's shopping protocol and a third-party checkout tool.

The interesting part is what Best Buy quietly stopped talking about. It announced a Google Cloud customer-support assistant in April 2024 as expected to launch late summer 2024, never confirmed it live, edited that page in January 2026 and left the 2024 future tense sitting there untouched. The words Google, Gemini, assistant and chatbot appear nowhere in its FY2026 annual report. What Best Buy did instead was point outward - partnering with OpenAI to surface its catalog in ChatGPT, supporting Google's Universal Commerce Protocol, and becoming the first retail partner to launch native checkout with a platform called Wizard. None of that got a press release. It surfaced on an earnings call and in one paragraph of the filing.

Source: Best Buy FY2026 annual report (SEC) · Mar 18, 2026

CVS No verified AI

Nothing in the retail business. In its business description, artificial intelligence appears twice, both in health-segment information-systems sections.

Zero mentions appear in the pharmacy and consumer wellness segment. CVS does run conversational AI in its health insurance business, but that is not the drugstore, and the distinction matters because the two get merged constantly. CVS also explicitly does not attribute its store closures to automation, citing store density and corporate workforce decisions instead.

Source: CVS Health FY2025 annual report (SEC) · Feb 10, 2026

The assistant is showing you a catalog somebody had to join

You ask ChatGPT or Gemini for a product and get a tidy shortlist with a buy button.

The retailers on that shortlist got there by integrating. Target's own June 2026 fact sheet names exactly where it did this - Google Search and AI Mode and Gemini, Microsoft Copilot, and a Target app inside ChatGPT. Best Buy told investors it partnered with OpenAI to surface its catalog in ChatGPT and supports Google's shopping protocol. Nike shipped a cart that lives inside Gemini rather than on nike.com. None of that is sponsored placement, and none of it is neutral either: a merchant that has not done the integration work is not in the running.

Practical read: the shortlist is not the market. Ask the assistant which retailers it cannot see, and check one price directly with a retailer it did not name before you buy.

Source: Target fact sheet, conversational AI · Jun 18, 2026

Maryland now requires a disclosure when an algorithm or your personal data set the price

Starting October 1, 2026, in Maryland, a price set that way has to say so.

Maryland Chapter 154 (House Bill 895, the Protection From Predatory Pricing Act) takes effect October 1, 2026. The required statement reads: THIS PRICE WAS SET BY AN ALGORITHM OR BY USING YOUR PERSONAL DATA. One section covers food retailers of at least 15,000 square feet that sell tax-exempt food, and third-party food delivery providers; a second covers other merchants, with exclusions including insurance-regulated entities and financial institutions. A violator has 45 days after notice to cure.

Practical read: this is a statute now in force, and the disclosure line is the part a shopper can look for at checkout. Enforcement and private-suit provisions differ by section, so read the chapter itself before relying on this summary.

Source: Maryland General Assembly, Chapter 154 · Oct 1, 2026

Seattle voted to stop grocers using personal data to set prices

Nothing changes at the shelf today. What changed is that a city now has a written rule about it.

Seattle's mayor announced on September 23, 2026 that the city adopted a Fair and Transparent Pricing policy, which she describes as first-in-the-nation for a city. It covers large online and brick-and-mortar grocery retailers and bars using sensitive personal information, including browsing history, social media activity and chatbot conversations, to set the prices customers pay. Traditional coupons and transparent discounts for groups such as seniors or veterans remain allowed, and prices must be clearly posted and available to all shoppers.

Practical read: the chatbot line matters most for this page. A conversation with a shopping assistant is named as data that cannot set your price. This entry rests on the mayor's release; the ordinance number and effective date were not published in the sources read, so neither is stated here.

Source: Office of the Mayor, City of Seattle · Sep 23, 2026

A regulator wrote down what a personal price has to tell you

Nothing yet, on the shelf. What changed is that there is now a written federal test you can hold a checkout screen against.

On August 19, 2026 the FTC proposed an enforcement policy statement on personalized pricing, which it defines as using personal data to set prices according to what a company believes an individual consumer is willing to spend. The Commission vote to publish it for comment was 2-0. It says up front what it cannot do: Congress has not given the Commission the authority to prohibit personalized pricing outright, but the Commission intends to enforce aggressively against any practices associated with personalized pricing that violate Section 5 of the FTC Act. So the remedy on offer is disclosure, and the statement is specific about what a real one looks like - clear and conspicuous, and including the fact that the price is personalized, the basis of that personalization, and the type of data used. It then names the failure mode in one line: telling a consumer only that he is being shown a specially selected price would likely be misleading because it omits important information.

Practical read: the eight-page statement gives seven plain-English examples, and three of them are ordinary shopping - a grocery chain charging a delivery customer more for milk based on data showing several children live in the household, a retailer charging more for a home-security camera system based on court filings indicating the customer was recently the victim of a crime, and a retailer charging more on its website based on data revealing the shopper is standing inside one of its own stores or parking lots while browsing. That is the useful part for a shopper, because it converts a dread you cannot act on into a test you can apply at a checkout screen. Read the limits honestly too: this is a proposal open for public comment, not a rule, the Commission says it confers no rights and binds nobody, and it declines for now to take a position on whether any personalized pricing is unfair even when fully disclosed. Comments were due September 18, 2026 and the FTC has since extended the deadline by seven days.

Source: Federal Trade Commission · Aug 19, 2026

Digital price tags are not surge pricing

The paper price ticket has been replaced by a small screen that can change instantly.

Walmart has roughly 2,300 US locations converted and expects chain-wide rollout within the next year. It published a standing policy in July 2026 naming and disavowing the practice people fear: DSLs change how we show price, not how we determine price, and it explicitly names surveillance pricing and surge pricing as capabilities it does not use.

Practical read: the label is a display, not a pricing engine. The question worth asking any retailer is not do you have digital tags, it is will two shoppers standing side by side ever see different prices for the same item.

Source: Walmart, pricing approach · Jul 1, 2026

A retailer put AI behind the shelf, measured it, and took it back out

Whether the thing you came in for is actually in stock, and who counted it.

Starbucks deployed an AI inventory system supplied by NomadGo that used computer vision to count beverage components such as milk and syrups, so stores would stop running out of things. About nine months later it was withdrawn, and the company confirmed it on the record twice. To Restaurant Dive: it has moved to a single, consistent process across all inventory counts, and this approach supports accuracy and product availability in our coffeehouses. To Fortune, a spokesperson: we test ideas in our coffeehouses, listen closely to partner feedback, and make changes to deliver a better, more consistent experience. Starbucks also gave reporters images of its own employees' comments on the removal. One of them reads: thanks for discontinuing Automatic Counting, the thought behind it was great, but the execution was proving difficult. Carl Addison, a shift supervisor of nine years in Shoreline, Washington, told Fortune the app required stores to rearrange back-of-house storage and then miscounted anyway - it started off not particularly accurate and got less accurate over time.

Practical read: this is the clearest public example of a company checking its own AI and acting on the answer, and the check got settled by the people doing the counting rather than by anyone in a boardroom. Read the scope honestly too. Starbucks withdrew this one system, not AI generally, and neither source published a store count, so any store figure you see attached to this story did not come from the companies or the reporting.

Source: Fortune · May 28, 2026

The Senate asked the questions you would want asked

Concern that electronic shelf labels plus cameras could add up to personalized prices.

Two US senators wrote to Kroger in August 2024 with exactly eleven numbered questions, including whether it would use customer data to institute personalized pricing on its shelf-label platforms, whether customers could opt out of that data collection, and how it would ensure it did not discriminate on protected characteristics such as race, gender or age when offering personalized prices.

Practical read: those eleven questions are a ready-made checklist you can put to any retailer. Note also that this was a letter seeking answers, not an enforcement action, so do not read it as a finding of wrongdoing.

Source: US Senate letter to Kroger · Aug 5, 2024

The checkout-free stores are closing

Walk-out-without-paying stores that felt like the future five years ago.

Amazon is closing its Amazon Go and Amazon Fresh stores and converting locations to Whole Foods, stating it had not yet created a truly distinctive customer experience with the right economic model needed for large-scale expansion. The technology survives as something Amazon sells to others - over 360 third-party locations across five countries - and in more than 40 of its own fulfillment-center breakrooms.

Practical read: where you will actually meet this now is a stadium concourse or an airport, not your weekly grocery run. Amazon states that Just Walk Out does not collect biometric information. Its separate palm-reader product is a different technology and is not covered by that statement - this page previously described how it works, and that description was removed in August 2026 because the only source for it was a 2023 page Amazon has not updated since.

Source: Amazon newsroom, store closures · Jan 27, 2026

Where you live decides what rights you have

The same camera, with completely different rules depending on the state.

Portland, Oregon bans private entities from using face recognition in places of public accommodation outright, with a private right of action attached. Illinois provides substantial per-violation damages and its supreme court has held that no actual injury is required. In most of the rest of the United States there is no requirement that anyone tell you at all.

Practical read: if a store accuses you based on a camera match, ask in writing whether a biometric system generated the alert and request the record. That is exactly the kind of notice regulators have made enforceable elsewhere.

Source: Portland City Code 34.10, facial recognition · Aug 18, 2026

One retailer was banned from facial recognition, and the findings are ugly

Cameras in the ceiling with no sign explaining what they do.

The FTC found that Rite Aid's system falsely flagged consumers as matching someone previously identified as a shoplifter; that employees followed consumers around stores, searched them, ordered them to leave, called the police and publicly accused them, sometimes in front of friends or family; and that it was more likely to generate false positives in stores located in plurality-Black and Asian communities.

Practical read: the resulting order is the useful part. It requires notifying consumers when their biometric information is enrolled and when action is taken against them. That is a reasonable standard to expect of any retailer using this.

Source: FTC, Rite Aid facial recognition order · Dec 19, 2023

A company owns what its chatbot promises

A support bot tells you a policy that turns out not to exist.

A tribunal rejected an airline's argument that its chatbot was responsible for itself, calling it a remarkable submission, and held that it makes no difference whether the information comes from a static page or a chatbot. The company was held to what its bot said and ordered to pay.

Practical read: screenshot what a retail chatbot tells you before you act on it. That screenshot is the whole ballgame if the promise turns out to be wrong.

Source: Moffatt v. Air Canada, 2024 BCCRT 149 · Feb 14, 2024

Your returns are scored by a company you have never heard of

You return something with a receipt and are told the return is warned or denied, with a phone number for a company you do not recognize.

The retailer sent your transaction to a third party that links it to your ID or payment card, builds a return history and returns an approve, warn or deny recommendation in real time. The federal consumer bureau lists that company among consumer reporting companies, a category that under the Fair Credit Reporting Act carries a duty to investigate disputes free of charge.

Practical read: this is the most actionable item on this page. The scoring company's own consumer site says its online portal takes requests for your return-activity report up to 60 days after the transaction and asks for the transaction ID printed on the slip you were handed, so do it before you cool off. The federal consumer bureau separately says the company provides one free report on request. Note the two sides give different advice about disputes: the company tells you to take errors to the retailer, because the retailer supplied the data, while the bureau says you may dispute with the reporting company and the retailer both. Frame any challenge as a Fair Credit Reporting Act dispute rather than a customer-service complaint.

Source: Consumer Financial Protection Bureau, consumer reporting companies · Mar 23, 2023

The AI label exists, but it was not built for you to see

A product photo that looks slightly too perfect.

Google's merchant rules require that all images created using generative AI carry metadata indicating the image was AI-generated, using a specific industry tag. That is a machine-readable marker in the file, not a visible label on the tile you are looking at. Nothing in the policy requires a retailer to tell you on the page.

Practical read: you can read it anyway. Long-press or right-click a product image and ask Google Lens or Gemini whether it was made with AI - that checks the embedded marker the retailer is required to include but not required to show you.

Source: Google Merchant Center, AI-generated imagery policy · Aug 18, 2026

The biggest names have the least to say.

Costco, TJX, Publix, Aldi and Trader Joe's operate tens of thousands of stores between them and make no AI claim at all. Scale does not predict adoption.

Announced is not shipped.

Look for the denominator, and treat its absence as an answer in itself. Best Buy announced a support assistant in 2024, edited the page in 2026, and left the words expected to launch late summer 2024 sitting there untouched.

The best proof is a walk-back.

A company that tells you what it removed is more credible about what it kept. Kroger, Amazon and Starbucks all did this in public, in writing. Starbucks is the one where the thing removed was the AI itself rather than a warehouse or a store format, and where the staff who used it were quoted saying why.

A quiet company is not a stalled one.

Target announced a staff chatbot chainwide in 2024 and has published nothing confirming it since. That absence is the finding - but it is not proof the tool is gone, only proof nobody has said.

Selling AI is not using AI.

Best Buy sells AI laptops. Amazon licenses its checkout technology to other retailers. Neither tells you whether AI runs their own stores.

Your leverage is at the returns desk.

Of everything on this page, the right to request your own return-activity file is the one with a form, a deadline and a federal statute behind it.

Which major retailers actually use AI?

Walmart, Lowe's, Home Depot, Kroger and IKEA have the most substantiated deployments, and only some of them publish a number you can hold them to. Kroger's staff app Sage is described by its own vice president of human resources as having over 95 use cases and being actively used by up to 150,000 associates - the clearest figure on this page, and an employee-facing one rather than a shopper-facing one. Lowe's reported its associate assistant reaching 5 million questions across 1,761 stores, in an SEC-filed document. Home Depot says Magic Apron now takes millions of questions a month. Walmart reports the number of customers using Sparky up 70% year on year but has never published an absolute count. IKEA published the savings from its service bot and the number of staff it reskilled, which almost no retailer does, though those figures now date from 2023.

Does Costco use AI?

There is no evidence that it does, and its 2026 filings are quieter than its 2025 ones. Costco's most recent quarterly report mentions artificial intelligence zero times. Across its entire filing history the phrase appears in exactly two documents, both annual reports, both in Risk Factors. Costco's own filing concedes that some competitors have greater technology capabilities, including the faster adoption of artificial intelligence, than it does.

Do digital price tags mean stores will charge me more at busy times?

Not at the retailers that have addressed it directly. Walmart published a policy in July 2026 stating that digital shelf labels change how we show price, not how we determine price, and explicitly naming surveillance pricing and surge pricing as things it does not do. An electronic label is a display, not a pricing engine. On September 25, 2026 Walmart's chief executive added a written commitment that information shared with its Sparky assistant will not be used to raise a price or hide lower-priced options. The open question is personalized pricing, which two US senators formally put to Kroger in 2024.

Why is self-checkout disappearing from some stores?

Several large chains have pulled self-checkout back and put staff at the front end instead, and the reasons they give publicly are shrink and customer experience rather than AI. This page used to carry a detailed account of one chain that used computer vision to measure the problem before deciding to remove it. That account was removed in August 2026 because the only source for it was a third-party transcript, and no primary company document published since restates it. The retreat is real; the AI attribution is not something we can currently source.

What happened to Amazon Go?

Amazon is closing its Amazon Go and Amazon Fresh stores and converting locations to Whole Foods, stating it had not yet created a truly distinctive customer experience with the right economic model needed for large-scale expansion. The checkout-free technology itself survives, in over 360 third-party locations such as stadiums and airports, and in more than 40 Amazon fulfillment-center breakrooms.

Can a store use facial recognition on me?

In most of the United States, yes, and with no requirement to tell you. Portland, Oregon is a notable exception, banning private use in places of public accommodation outright, and Illinois provides substantial per-violation damages with no requirement to show actual injury. The FTC has brought one major retail case, against Rite Aid, which resulted in a five-year ban.

Why was my return denied when I had a receipt?

Many large retailers send return requests to a third-party scoring service that links the transaction to your ID or payment card and returns an approve, warn or deny recommendation. You can request your own return-activity report. That company's consumer site says its online portal accepts requests up to 60 days after the transaction and asks for the transaction ID printed on the slip you were handed. The federal consumer bureau lists the company among consumer reporting companies, says it provides one free report on request, and notes that under the Fair Credit Reporting Act a reporting company must investigate disputes free of charge.

Is the retailer responsible for what its chatbot tells me?

A Canadian tribunal said yes, in a decision that has been widely cited since. It rejected the airline's argument as a remarkable submission and held that it makes no difference whether the information comes from a static page or a chatbot. Screenshot what a support bot tells you before acting on it.

Filings first, press releases last

Most retailers here are public companies. A press release announces the pilot; the annual report and the earnings-call question-and-answer six months later are where a company says what actually happened. Most of the strongest findings on this page came from filings that nobody writing about retail AI reads.

No retailer claim here rests on a document older than January 2026

Every retailer entry carries the publication or filing date of the document behind it, and that date is 1 January 2026 or later. Where a claim could only be supported by a 2025 or older source, it was removed rather than left standing with an old citation - even when the claim was a good one. The one deliberate exception is law: the consumer-rights section below cites statutes, regulator orders and court rulings by the date they were enacted or decided, because a rule still in force is not out of date. Those are the only pre-2026 dates on this page.

That rule cost this page real material, including the single best counter-narrative story it had. If a fact matters, somebody will have restated it in a filing this year. If nobody has restated it in eighteen months, that silence is its own finding, and pretending otherwise is how a reference page rots.

Absence is evidence, when you check properly

A No verified AI label here does not mean nobody looked. It means the company's own annual report and newsroom were searched and the only mentions were risk-factor boilerplate - language every public company now carries whether or not it uses anything. That distinction is the whole exercise.

Using AI and selling AI are different things

Best Buy sells AI laptops. Amazon licenses its checkout technology to other companies. Walmart sells data insights to its suppliers. None of that tells you whether AI is running their own stores, and conflating the two is the most common error in coverage of this topic.

Hunt for the denominator

Deployed at 400 stores means little until you know the chain has 2,000. Where a company published the denominator it is quoted here. Where it did not, that is stated too, because a missing denominator is usually the story.

What is deliberately not here

No hypothetical examples and no vendor marketing statistics presented as independent research. Several widely repeated claims were cut because no primary source could be verified, including reports of AI systems falsely accusing shoppers of theft - the cases found either could not be traced to a court filing or turned out not to involve AI at all. Saying so is more useful than repeating it.