What are the top startups in the AI shopping market?

Which companies actually count as AI shopping startups?

We count AI shopping startups as companies that directly change how people discover, compare, choose or buy products, which leaves us with a much smaller market than the broader “AI for retail” category.

That includes consumer products such as Daydream, Dupe, Onton and Phia, where people deliberately use AI to find something to buy. It also includes companies such as Constructor, Remark and Envive, whose technology sits inside retailer websites and guides shoppers toward products. Lily AI belongs one layer underneath them: it improves the product data that search engines, retailer systems and AI agents use when deciding what to recommend.

We exclude warehouse AI, demand forecasting, ad-generation tools and generic customer-service bots unless they materially affect the buying decision.

The difficult part is comparing companies that operate at different points in the journey. Dupe says 20 million-plus shoppers have used its product. Constructor processes hundreds of billions of shopping interactions inside retailers. Remark works with more than 85 brands. Those numbers describe three different kinds of scale.

So our ranking gives more weight to repeated usage, measurable commercial impact, retailer adoption and proprietary shopping data than to funding or a large cumulative signup number. That changes the leaderboard considerably.

Is AI shopping really changing what people buy yet?

AI shopping is already changing purchase behavior, and the strongest evidence now comes from conversion data rather than chatbot usage.

Adobe Digital Insights tracked AI-referred retail traffic from late 2024 through early 2026. At the beginning of that period, shoppers arriving from AI assistants converted considerably worse than normal visitors. By March 2026, the relationship had flipped: AI-referred visitors converted 42% better than non-AI traffic. Adobe also found that 80% of consumers who already use AI for shopping were using it more frequently, while 79% said an AI assistant made them more confident about a purchase.

Amazon gives us another, much larger test. Rufus, since renamed Alexa for Shopping, was used by more than 300 million customers during 2025. Amazon said those users were around 60% more likely to complete a purchase and attributed nearly $12 billion in incremental annualized sales to the assistant.

We would be careful with the $12 billion figure because Amazon controls the attribution model. The behavioral pattern is harder to dismiss. Adobe sees higher conversion across the open web while Amazon sees higher purchase completion inside its own marketplace.

The clearest shift currently happens before checkout. People are increasingly asking AI to narrow the consideration set, compare products and answer specific questions before deciding what to buy. Fully autonomous purchasing is developing more slowly.

Can AI shopping startups survive Google, Amazon and ChatGPT?

Generic AI shopping assistants have a bad competitive position today because Google, Amazon and ChatGPT already own the distribution, product data or shopping history that a horizontal assistant needs.

Google's advantage has become enormous. Its Shopping Graph now contains more than 60 billion product listings, up from the 50 billion-plus figure Google was citing earlier in the year. Google says people already shop across its services more than a billion times per day, and Gemini now connects conversational shopping with that product graph, price data, inventory and Universal Cart.

Amazon has something Google cannot fully reproduce: years of individual purchase history tied directly to a marketplace. Alexa for Shopping can recommend, compare, track prices and increasingly act for the shopper. Amazon can also measure whether the recommendation ends in a purchase.

ChatGPT attacks from distribution. OpenAI says hundreds of millions of people use ChatGPT to find products, and its shopping research experience can now compare products, work through constraints and personalize recommendations using conversation context. OpenAI has also made a revealing strategic choice: after experimenting aggressively with native checkout, the company said its first Instant Checkout model did not give merchants enough flexibility and shifted more attention toward product discovery.

A startup therefore needs a narrower advantage than “we built an AI that helps you shop.” Daydream understands fashion intent across thousands of brands. Onton specializes in visually complex discovery. Remark owns conversations derived from human product expertise. Constructor learns from retailer shopping behavior.

Those positions can survive even if the general shopping interface increasingly belongs to the big platforms. A horizontal shopping chatbot with little proprietary data has a much harder road.

If you want more recent data on this point, please see our latest AI shopping market report.

Why are fashion and furniture producing so many AI shopping startups?

Fashion and furniture are producing some of the strongest AI shopping startups because traditional ecommerce search handles these categories unusually badly.

A shopper looking for a phone can often specify storage, screen size and price. Fashion language is messier. Someone might want “a loose cream dress for a wedding that feels expensive but not formal.” Furniture brings its own problems around shape, material, room style and visual similarity.

Daydream was built around exactly that fashion-language gap. Its catalog now covers roughly three million products from more than 10,000 brands across 325 retailers. Onton started in furniture, where shoppers can move between natural-language searches, product images and generated visual scenes. Dupe approaches the same problem from similarity: give it an expensive sofa, dress or beauty product and ask for something that looks close enough at a lower price.

Phia has built another fashion-specific route through price comparison, resale and alternative products.

General models will keep getting better at understanding these requests, so language understanding alone will not protect these companies. The more interesting asset is the behavioral trail behind the query: which two dresses shoppers really consider substitutes, which visual similarities lead to clicks, which recommendations get rejected and which attributes actually end in a purchase.

That kind of vertical shopping data gets better as people use the product. Fashion and furniture simply give startups more opportunities to collect it.

Which consumer AI shopping startups are people actually using?

Onton, Daydream, Dupe and Phia currently have the strongest public evidence of real consumer usage, although Onton's monthly-active-user figure is much more useful than the cumulative numbers disclosed by several competitors.

Dupe's current website says more than 20 million shoppers have used the service. That is the biggest disclosed consumer number in this group, but it is cumulative. We cannot read it as 20 million active users.

Onton's figure is cleaner. The company told TechCrunch that monthly active users grew from roughly 50,000 to more than two million. That is a roughly 40-fold expansion, and monthly activity gives us a much better picture of current usage.

Daydream recently passed 1.5 million shoppers. Phia has also reported more than 1.5 million users, alongside 9,600 brand partners. Both have reached genuine consumer scale in a short period, although Phia's monetization data now deserves much more scrutiny than its user count.

Startup	Strongest current usage evidence	What we can safely conclude
Onton	2M+ monthly active users, up from about 50K	Strongest disclosed active-usage signal
Dupe	20M+ cumulative shoppers	Very large reach, but active retention is unclear
Daydream	1.5M+ shoppers	Meaningful fashion audience with growing retailer distribution
Phia	1.5M+ users	Real consumer adoption despite doubts around sales attribution

Is Daydream the strongest consumer AI shopping startup right now?

Daydream is currently our strongest consumer-first AI shopping startup because it has started turning a 1.5 million-shopper fashion app into technology that retailers themselves want to install.

The original Daydream product already had an attractive position. Instead of asking shoppers to navigate filters, it lets them describe what they want naturally and searches a catalog covering millions of fashion products.

The more interesting move came recently with Powered by Daydream. STAUD, Alice + Olivia, Couper, Cult Mia and Hampden went live with Daydream's technology on their own websites. More than 25 additional brands and retailers signed up, including Anine Bing, Mansur Gavriel, ba&sh, Sandro, Maje and MESHKI.

So Daydream already has more than 30 live or committed retailer relationships in the first disclosed group. The business looks different now. The company can keep learning from consumers on Daydream while also selling its search and discovery technology closer to the transaction.

We still lack the conversion lift, recurring revenue and customer-retention numbers that Constructor can show. Daydream therefore remains a bet on where the market is going rather than the most proven business in the sector.

Among startups trying to build a genuine consumer shopping destination, however, Daydream has the most convincing position today.

If you want more recent data on this point, please see our latest AI shopping market report.

Are Dupe and Onton getting more traction than their funding suggests?

Dupe and Onton are two of the most capital-efficient consumer challengers in AI shopping, and Onton's growth is especially difficult to ignore.

Onton went from roughly 50,000 monthly active users to more than two million before announcing a $7.5 million financing round. Even without pretending those figures reveal revenue or retention, reaching that level of monthly usage with a relatively small company is unusual.

The product also appears to be used for more than one quick search. When Onton disclosed the funding, it said more than 20% of users were active weekly, while some high-intent shoppers were generating more than 100 searches or images per month.

Dupe has taken a simpler route. Its current site says 20 million-plus shoppers have used the service, and the product now spans furniture, fashion and beauty. The company also has a native presence inside ChatGPT, which is an interesting distribution strategy when shoppers increasingly begin product research there.

We still know less about Dupe's active-user retention and economics than we would like. The 20 million figure tells us reach, while Onton's monthly figure tells us ongoing activity.

That gives Onton a slight edge in our ranking. Dupe may have touched many more shoppers, but Onton currently gives us better evidence that people keep coming back.

Can Phia still rank near the top after the affiliate scandal?

Phia still belongs among the leading consumer AI shopping startups, but the affiliate-attribution scandal pushes the company well down our overall ranking.

The consumer traction remains real. Phia has reported more than 1.5 million users, 9,600 retail brand partners and $43.5 million in funding. Its combination of fashion comparison, resale options, price tracking and personalized discovery has clearly found an audience.

The problem sits in the commercial evidence. Investigations found that Phia's browser extension had attributed some transactions to itself even when Phia had not generated the shopper's original referral. Impact.com subsequently reprocessed affected actions from the period it was reviewing, reassigning sales to the correct publishers or refunding advertisers. Other affiliate networks also investigated or reconciled transactions.

Phia says the problematic behavior has been fixed and has offered reversals for affected transactions. More recent reporting has kept the controversy alive by questioning how early people inside the company knew about the attribution behavior.

For our ranking, we do not need to settle every disagreement around intent. Sales attributed through the affected affiliate system cannot carry the same evidentiary weight as they did before the controversy.

The 1.5 million-user figure still tells us Phia built something people wanted to use. We simply have much less confidence in what the company's historical sales-attribution numbers tell us about incremental value to retailers.

If you want more recent data on this point, please see our latest AI shopping market report.

Is Constructor the company to beat in AI shopping?

Constructor is the company to beat in AI shopping today because no other independent startup we found combines comparable shopping volume, customer retention and repeated proof that retailers make more money after installing the product.

Constructor reported 322 billion product-discovery interactions during FY26, up 266% over two years, while its customer base increased 82%. Gross revenue retention finished the year at 96%. Its current website now says the platform powers more than 400 billion requests annually and shows average client retention of 98.5% over the previous three years.

The merchant results make the case stronger. Constructor says Belk has generated $35 million in additional revenue through its search optimization program, with the retailer's AI Shopping Agent converting at more than twice the rate of standard search. Petco recorded a 13% ecommerce conversion lift in another published case study.

Vendor case studies always deserve some skepticism. What makes Constructor different is recurrence. We see the same pattern across multiple retailers, products and years, alongside high customer retention. That is much stronger than one impressive pilot.

Constructor is also moving outside the old search box. Its platform now includes AI Shopping Agent, Product Insights Agent, Merchant Intelligence Agent, retail media, product-data enrichment and tools for supplying information to external answer engines such as ChatGPT.

That breadth gives Constructor a credible path even if more product discovery happens away from retailer websites.

Company	Strongest business evidence today	Main weakness
Constructor	322B FY26 interactions, +82% customers, 96% gross revenue retention	Big platforms increasingly control offsite discovery
Remark	85+ brand partners and $60M+ claimed brand revenue lift	Much smaller installed base
Envive	Large measured conversion lifts across several brands	Less public retention and revenue-scale data
Lily AI	Controlled lifts across Google, Meta and onsite search	Operates mostly below the consumer interface

Can Remark's human-expert model really become a moat?

Remark has one of the more credible data moats in AI shopping because its models learn from real product experts rather than relying only on the same public information available to every general-purpose LLM.

Remark built a network of tens of thousands of people with specific product knowledge, from stylists and chefs to athletes and skincare experts. Their interactions with shoppers become training material for AI personas that can answer the same kinds of questions later.

The commercial trajectory is becoming easier to see. At its Series A, Remark said it worked with more than 60 brands, had grown revenue fourfold in a year, retained 100% of customers and reached nearly 130% net dollar retention. The company's current site now lists more than 85 brand partners, 500,000-plus guided conversations and more than $60 million in claimed brand revenue lift.

Recent customer examples make the usage more concrete. Pepper Home attributes more than $2 million in revenue to Remark. MirrorMate reports a 32% conversion rate among shoppers using the experience, while Travelon has attributed more than $229,000 in revenue to it.

Remark has also widened the product beyond product advice. Recent launches include checkout directly from chat and a compatibility API designed for products where fitment matters.

We would still rank Constructor above Remark because Constructor operates at a completely different scale. Remark's advantage is more specific: if shoppers keep asking questions that require judgment rather than simple retrieval, a proprietary archive of expert reasoning could become extremely valuable.

If you want more recent data on this point, please see our latest AI shopping market report.

Is Envive actually making shoppers buy more?

Envive currently has some of the strongest conversion experiments in AI shopping, with several retailer deployments showing that conversational guidance can move people from hesitation to purchase.

At Spanx, Envive says its AI agent increased conversion by more than 100%, producing $3.8 million in annualized incremental revenue and a 38-times return on spend. The rollout itself is revealing: Spanx initially exposed the system to 15% of traffic during its biggest shopping weekend, then increased exposure to 90% after seeing the early results.

Supergoop gives us a second test in a different category. Envive reports an 11.5% conversion increase, 5,947 additional monthly orders and $5.35 million in annualized incremental revenue while average order value held at $75.

Clove provides a third example around footwear, where sizing questions create obvious purchase friction. Shoppers exposed to Envive recorded a 22.7% lift in checkout conversion.

We would not average those three percentages because the products, shoppers and experimental designs differ. We can still learn something from the pattern. Envive has now produced material conversion changes in shapewear, skincare and footwear rather than relying on one unusually successful customer.

The remaining gap is business scale. We have better visibility into Envive's customer-level ROI than into its overall revenue, customer retention or total volume. That keeps Envive below Constructor and Remark for now.

Could Lily AI's product data layer be more valuable than another shopping chatbot?

Lily AI has become more interesting as AI shopping has grown because every shopping agent now depends on understanding product data that was originally written for humans, databases and old search engines.

Consider a retailer selling a dress internally tagged as “women's woven midi.” A shopper may ask Gemini or ChatGPT for “something romantic for a summer wedding that doesn't feel too formal.” If the product record contains no language around occasion, silhouette, aesthetic or fit, the model has less evidence that the dress belongs in the answer.

Lily AI enriches that underlying product record. That puts the company underneath Google Shopping, Meta, onsite search and external AI assistants rather than tying its business to one shopping interface.

The measurement has also improved recently. Lily says it ran more than 1,000 controlled experiments before publishing its latest benchmark. In a matched-spend Google Shopping test with a 28-day holdout, Lily-enriched product language increased revenue by 28% versus the control. A separate Meta Advantage+ test produced a 21.4% ROAS lift, while an onsite-search experiment showed a 28.3% increase in revenue.

Those are much more useful numbers than an attribution dashboard saying AI “touched” a sale. Each test had a control group.

Marks & Spencer is also using Lily AI to enrich product data at scale, adding another large retailer to a customer group that has included Bloomingdale's, J.Crew, Tapestry and Bombas.

Lily AI will probably never look like the winner of the AI shopping race from a consumer's perspective. Strategically, we think its position has improved because the number of AI surfaces trying to read retailer catalogs keeps multiplying.

Will AI checkout become the main AI shopping opportunity?

AI checkout is becoming real, but product discovery currently looks like the more valuable battleground for independent AI shopping startups.

OpenAI's own strategy is revealing. The company launched Instant Checkout aggressively in 2025, then later said the first version did not give merchants the flexibility it wanted. OpenAI shifted more attention toward richer product discovery while letting merchants keep more control over checkout. Some eligible in-chat checkout experiences still exist, but discovery is clearly doing more of the strategic work.

Google is pushing further into transaction infrastructure through Universal Commerce Protocol, Google Pay and Universal Cart. Its new cart can follow products across Search, Gemini, YouTube and other Google surfaces while tracking prices, stock and compatibility.

Large platforms have obvious advantages here. They already handle identity, payments, merchant relationships and fraud. Standard protocols also make the mechanical act of ordering easier for many agents to perform.

The harder problem sits one step earlier: deciding which product deserves to be bought. Constructor has retailer behavior. Daydream has fashion searches and preference conversations. Onton sees visual discovery behavior. Remark captures expert-guided decisions. Lily AI structures the product information agents consume.

For startups, those decision layers currently look more defensible than owning the final checkout button.

So what are the top AI shopping startups today?

Constructor is the top AI shopping startup today, while Daydream is our strongest consumer-first bet and Onton is the most impressive smaller challenger.

Constructor takes first place because it has already built a large, sticky commercial business. Hundreds of billions of shopping interactions, 82% customer growth in its latest fiscal year and 96% gross revenue retention give us much more evidence than a typical AI commerce startup can provide.

Daydream ranks second. The recent move from a 1.5 million-shopper destination into retailer infrastructure makes the company more interesting than it was even a few weeks ago. Daydream can now learn from both its own consumer audience and shoppers using its technology directly on brand websites.

Onton moves ahead of Dupe in our ranking because two million-plus monthly active users tells us more than a cumulative reach number. Going from roughly 50,000 to more than two million monthly users is one of the clearest breakout trajectories we found.

Remark comes fourth. Its scale remains below Constructor, but the public brand count has moved from 60-plus around its Series A to more than 85 today, while its human-expert training model gives the company one of the more distinctive proprietary datasets in the market.

Dupe ranks fifth. Twenty million-plus cumulative shoppers make the consumer traction impossible to ignore, and its simple “find me something similar for less” behavior gives it strong purchase intent. Better active-user and monetization data could move Dupe higher.

Lily AI takes sixth because product intelligence is becoming more valuable as shopping fragments across Google, ChatGPT, retailer sites and other AI agents. Its recent controlled experiments also give us unusually clean evidence that better product data can change revenue.

Envive ranks seventh. We like the repeated conversion results across Spanx, Supergoop and Clove, but we still know much less about Envive's overall scale and retention than we do about Constructor or Remark.

Phia ranks eighth. Its consumer growth still puts it in the conversation, but the affiliate-attribution controversy removes one of the strongest pieces of evidence previously supporting a higher position. Phia can climb again if user retention stays strong and cleaner merchant economics emerge.

Rank	Startup	Why we rank it here now	What could change our mind
1	Constructor	Best combination of scale, retention and measured retailer impact	Offsite AI discovery weakens its retailer-site position
2	Daydream	Strong consumer traction plus a fast-emerging B2B fashion-search business	Retailer pilots fail to produce meaningful conversion or recurring revenue
3	Onton	2M+ monthly users and unusually fast growth from a small base	Engagement falls as the company expands beyond its strongest categories
4	Remark	85+ brands, growing commercial evidence and proprietary expert data	General models reproduce enough product expertise to shrink the advantage
5	Dupe	20M+ cumulative shoppers and a very clear high-intent use case	Active usage or affiliate economics prove much weaker than reach suggests
6	Lily AI	Strong controlled tests and a valuable product-intelligence position beneath AI agents	Platforms solve catalog enrichment themselves
7	Envive	Repeated conversion lifts across several large consumer brands	Needs stronger company-wide scale and retention evidence
8	Phia	1.5M+ users and major fashion-shopping reach	Cleaner attribution and merchant data are needed to rebuild confidence

If you want more recent data on this point, please see our latest AI shopping market report.

Who is the author of this content?

NEW MARKET PITCH TEAM

We track new markets so founders and investors can move faster

We build living "market pitch" documents for emerging markets: AI, synthetic biology, new proteins, and more. Instead of outdated PDFs or hallucinated LLM answers, our clients get a clean, visual, always-updated view of what's really happening: key players, deals, regulations, and signals that matter. Learn more about us.