AI Shopping Agents in Fashion: What Changes and What Doesn't

Published: 05.02.2026

This is part two of my learnings from building Alle, a consumer AI product in fashion. Unlike the first part, where our learnings were earned from first-hand execution, this one comes from research and domain-driven thinking (shaped by our past experience in commerce), around an opportunity we identified for a pivot. So consider this a disclaimer before taking the implications of this post at face value.

This post centers on shopping agents (also thematically referred to as agentic commerce/agentic shopping). These are intelligent software systems capable of taking autonomous actions on behalf of users, based on their instructions, across the commerce transaction life cycle. We were trying to build a mental model of what this technology now enables and, by extension, the changes it could bring to the broader commerce landscape.

image

A few caveats before we go ahead:

  • This post assumes readers are already familiar with the classic commerce funnel of Need (in case of intent) → Discovery (intent/impulse) → Consideration → Transaction, along with the broader landscape of horizontal discovery platforms, horizontal/vertical commerce marketplaces, and D2C brands.
  • While the research focuses on consumer fashion, my sense is that most insights (though not all) should be transferable across other consumer lifestyle categories.

I will structure this around L1 questions I was either asked or personally curious about, and then go deeper into each of them.

Question 1: In an AI-first world, could horizontal discovery platforms (ChatGPT, Gemini etc.) achieve stronger product-market fit for fashion shopping than vertical full-stack platforms (Myntra, Meesho etc.)?

To answer this, I found it helpful to first understand consumer behaviour across the discovery-to-transaction funnel today.

‣

Where does fashion discovery for shopping happen today?

  1. Commerce destinations

  • These include fashion marketplaces (like Myntra, Amazon Fashion, Meesho, etc.) and popular brand stores (like H&M, Zara, Savana, etc.).

  • Their primary proposition is wide, diverse, and trendy selection, relevant pricing, domain-relevant UX for discovery, and most importantly, transaction trust built through systems for quality control, fulfillment experience, return policies, and overall reliability.

  • Most consumers default to these destinations in two key moments:

  • In moments of need for active discovery. This journey often starts well in advance of the actual purchase, with consumers opening apps multiple times to browse actively, wishlist items, and finally transact when the need is closer or when they have money.

  • In moments of active leisure for passive discovery (though the overall market size for passive discovery on social platforms is much larger).

Fashion shopping motivations that drive app opens are an interesting topic on their own. For a more nuanced and holistic perspective, I highly recommend reading this.

  1. Entertainment destinations

  • These are relatively few in number, with the market largely concentrated across Instagram, TikTok, and YouTube.

  • This discovery funnel typically starts with boredom-alleviating behaviour, where entertainment destinations have strong product-market fit today. As consumers spend time on these apps, and through their exceptional personalisation systems, they end up discovering creator content or ads that can lead to wishlisting or impulsive purchases.

  • Depending on the quality of algorithmic personalisation, users may discover products aligned with an active need (for eg: Meta’s Advantage+ Catalog Ads) or a passive one (for eg: relevant casual office outfits that they might choose to buy now or later, since such needs are ongoing).

  • Shopping may happen as a result, but it is more likely to be impulsive than intentional. The user is not coming in with a specific need. They are browsing out of curiosity or for entertainment, and if something catches their eye, they could end up buying shortly after discovery (anywhere from the same day to a few days later).

  • While this funnel’s discovery originates on Instagram, users are eventually redirected to commerce destinations. These destinations benefit from passive discovery, as they now have the user’s attention and can surface adjacent product inventory as well.

  1. Horizontal/Social Search destinations

  • These are very few in number, with the primary players being Google (and consequently Google Shopping) and Instagram/TikTok search. The most recent entrants here are general purpose AI assistants, primarily ChatGPT, Gemini, Claude, Perplexity etc.

  • Their core value proposition rests on two things:

  • the widest selection of products (in the case of Google Shopping) or inspiration (in the case of Instagram)

  • search-based discovery

This is enabled by great technology, and aggregation across all commerce destinations, including marketplaces, brands, and independent labels.

  • Consumers typically turn to these platforms when their primary commerce destination(s) are unable to fully satisfy their shopping need.

‣

Why do commerce destinations continue to dominate fashion discovery?

  • In the context of the original question, the core insight is that despite horizontal search destinations (like Google/Google Shopping) existing for many years, the primary discovery surfaces are still commerce destinations (though in the recent years, they have ceded some ground to Instagram/TikTok, especially for passive browsing use cases, and also due to rising disposable incomes making casualwear purchases more impulse-driven rather than intent-led).
  • More specifically, every consumer has a “mental short-list” of a few apps they go to when starting discovery. For eg: for some people this could be Myntra, Savana, and H&M, which they open whenever they want have a need, or simply want to browse fashion during active leisure moments.
  • One would assume that the meta-aggregation Google Shopping has across the entire universe of fashion inventory should earn it the first right of passage. But that’s not the case. This is primarily because large selection cannot offset the most important commerce variable of trust.
  • Trust in commerce is not a commodity and is expensive to earn. Most consumers will experiment with buying directly from new brands only a few times a year, and the majority of purchases will continue to happen where trust already exists. In that sense, trust is scarce. Historically, trust has been best solved by commerce companies because they do the complex and intensive work of ensuring quality, reliable fulfilment, timely customer support, and strong return experiences, rather than acting as mere product classifieds. Because doing this well demands scarce operational and technological rigour, trust ends up concentrating among a few large marketplaces and brands.
  • Starting fashion discovery on Google is inefficient from this lens. Imagine searching for a “casual white top” on Google and seeing results from brands you don’t trust, in a UX that isn’t purpose-built for fashion. The overall value equation ends up being significantly weaker than discovering directly on a commerce destination.
  • That said, Google (and to a lesser extent Instagram search) still has utility when a consumer is unable to discover something relevant on their primary commerce apps. In those cases, the value proposition of widest selection, despite a sub-optimal discovery experience, wins over everything else.
  • This means in a steady state, a commerce discovery funnel exists across both commerce destinations and horizontal search platforms, with commerce holding the primary and larger discovery mindspace for consumers.
  • And now, enters AI. It’s already evident that it can change consumer expectations and behaviours around how people shop for fashion. This brings us to the next question.

‣

What changes does AI enable in fashion shopping?

  • It’s useful to first think about the impact AI is having across different stages of the shopping funnel before diving into what shopping agents will change.

Layer

What AI enables

Discovery

• Natural language queries (“what to pair with this blue sleeveless top?”, “show me outfits for a date night”) • Multimodal image queries (a user uploads an image of a brown skirt they saw on Instagram and asks for something similar in white) • Personal style memory (LLMs can build a richer understanding of a user’s taste, making personalisation significantly stronger) • Content-first feeds (image generation models can create inspirational outfit images similar to Pinterest or Instagram, making catalog images more useful than today’s model photoshoot feeds)

Consideration

• Better personalisation at the discovery layer naturally makes consideration more efficient • Personal stylist capabilities (users can ask for help choosing between shortlisted options) • Smarter size selection (richer qualitative recommendations beyond simple size charts)

Transaction

• Agentic checkout that navigates to PDPs, compares prices, autofills addresses, and completes payment • Agentic post-transaction systems that coordinate returns and customer support

  • But, notice that AI shopping agents (including those built into ChatGPT/Gemini), still do not solve for trust. Trust in commerce extends beyond information and interfaces into the physical world - product quality, delivery, returns, and real-world accountability. At best, right now AI can improve convenience across the funnel: enhancing discovery and consideration before purchase, reducing friction during checkout and price comparison, and simplifying post-purchase actions like returns or customer support.
  • While these improvements are valuable and non-trivial to build, they are not exclusive. Both new horizontal discovery platforms and existing commerce destinations - marketplaces and brands - can offer the same capabilities. Crucially, none of these features directly create trust; they mainly reduce the friction around trust by increasing proportion of trustworthy results.
  • This brings us to the implication for how the market landscape changes as a result of agentic capabilities.

‣

Why commerce companies will continue to enjoy stronger PMF (or become even stronger)?

  • This suggests that even if ChatGPT or Gemini offer agentic shopping to consumers, this alone will not give them any lasting advantage over commerce destinations. The core technology behind these features will not be proprietary. As models improve, similar capabilities will be accessible to everyone else through both closed-source and open-source solutions.

Because of this, ChatGPT and Gemini will not be meaningfully better than existing marketplaces at being the primary discovery solution in the consumer’s mind. This situation is similar to the past, when commerce marketplaces/brands remained the primary discovery surface even though Google (Google Shopping) already existed.

  • In fact, my prediction is that commerce destinations will have stronger PMF against new horizontal discovery destinations (like ChatGPT and Gemini) than they ever had against their historical precedents. This is because beyond trust, commerce companies make strategic business and product decisions that are next to impossible for horizontals to structurally compete with. Some of these are inherent to the category, while others are the result of deliberate, defensive choices made by marketplaces and brands to protect their positioning as the primary discovery surface for the category against the horizontal.

  • Inability to index full inventory

Major marketplaces restrict access to large portions of their inventory (I’m not sure of the exact figure, but I’ve heard it ranges from 40-70% of catalog breadth) when syndicating to horizontal platforms. This is done to give consumers a reason to begin discovery within their own apps.

The result is partial selection coverage on horizontals, which weakens the promise of universal selection, erodes the value of discovery, and ultimately reduces trust in recommendations.

  • Lack of real-time stock and pricing

Fashion inventory and pricing change rapidly due to sales, coupons, and frequent SKU-level stock updates. Scraping this data reliably is both brittle and expensive. While Google Shopping allows marketplaces and brands to index their inventory through APIs, it still does not achieve full catalog coverage, as noted earlier.

  • No purpose-built browse mode

Fashion shopping is inherently serendipitous. Infinite scroll and swipe-based feeds encourage exploration and habit formation. Horizontals lack a native browse mode optimised for this behavior, making it harder to become a high-frequency discovery destination.

  • Weak pricing levers

Marketplaces actively shape effective prices through loyalty programs, payment offers, timed sales, and exclusive discounts. Horizontals typically lack access to these levers, which means commerce apps often deliver better final prices to consumers.

  • Limited control over post-purchase operations

Trust in fashion is reinforced after checkout through smooth returns, instant refunds, and efficient customer support. Horizontals don’t have any control over these trust levers.

  • I expect all of these to be positively impacted by AI, to varying degrees of improvement, which will further strengthen the positioning of a commerce destination, especially marketplaces, against horizontals.

‣

So why are ChatGPT/Gemini still pushing out agentic capabilities for fashion shopping?

  • Primarily because they sit on a discovery funnel for user’s shopping intent. Remember how we discussed that, despite users having primary commerce apps for discovery, there are moments where those apps fail to surface relevant products. This creates room for a secondary discovery surface, which has historically been Google and will increasingly become ChatGPT/Gemini. (A reasonable estimate could be ~10–20% of fashion discovery instances annually where primary commerce apps fail to show relevant products.)

At the scale of users ChatGPT has, combined with the recurring nature of fashion discovery, the monetisation opportunity is still massive enough to pursue.

  • When an agent can finish a purchase, it’s not only more convenient for the user, it shifts value from clicks and impressions to the transaction itself. That could improve attribution and raise revenue per query versus CPC or CPI models. It’s still unclear whether this will really happen. It’s possible the market equilibrium settles into something very similar to today’s CPC/CPI ad models, with similar margins and economics. Brands probably won’t give advertisers higher margins just because agents are doing the work, and they also won’t want to raise prices for consumers.
  • One important point to note is that executing a transaction autonomously is not the same as monetising it. Brands or marketplaces still need to agree to share margin with the agent. I’ve heard a few people argue that business partnerships aren’t necessary since agents can transact through browser use. While that may technically work, the need for monetisation and, at the very least, transaction speed efficiency will ultimately necessitate partnerships.

TL;DR: In today’s fashion shopping funnel, discovery primarily happens on trusted commerce destinations, with entertainment platforms driving passive inspiration and horizontals like Google stepping in only when core apps fail. Despite AI dramatically improving discovery, personalisation, and transaction convenience, it does not solve the most important variable in commerce: trust around quality, fulfilment, returns, and accountability. Because trust is operational and hard-earned, it continues to concentrate among large marketplaces and brands. While horizontal AI platforms like ChatGPT and Gemini can offer agentic shopping, these capabilities are non-exclusive and structurally constrained by limited inventory access, weaker pricing levers, and lack of post-purchase control. As a result, commerce destinations are likely to retain, and even strengthen, their PMF for fashion discovery in an AI-first world.

Question 2: Can a new marketplace opportunity exist on the back of shopping agents?

  • Based on the predictions implied by the previous section, I don’t believe a new marketplace opportunity exists where the core differentiator is a transaction experience replaced by shopping agents. Historically, consumers have gravitated toward destinations that solve for trust, selection, price, and reliability, while spillover discovery tends to flow to horizontal platforms.
  • Even in the past, we’ve never really seen the emergence of vertical discovery-only businesses. The archetypes that exist today fall into one of two buckets: horizontal discovery (Google, ChatGPT, Gemini) and commerce platforms (Amazon, Myntra, etc.).
  • This raises a key question: what unique problems can a new agent-first marketplace actually solve?

As discussed earlier, trust, selection, and pricing matter far more in commerce than reducing friction in discovery and checkout, which are the core strengths of shopping agents. Both marketplaces and horizontal discovery platforms will build these agent-driven capabilities over time. That leaves an open question of what additional value a new agent-first marketplace can deliver, and whether that value is strong enough to be defensible.

Question 3: Are there any new consumer opportunities possible due to shopping agents?

  • So far, I’m unable to imagine a truly new consumer-facing opportunity built purely due to the advent of AI shopping agents.

  • That said, I do think there could be an opportunity to convert existing or new discovery surfaces (that are outside of horizontal assistants and commerce platforms), into experiences that offer agentic transaction capabilities.

  • These surfaces could include social recommendations, expert suggestions, and similar channels that already exist today. In many ways, this resembles building a modern affiliate network, but with stronger transaction attribution, lower margin leakage, and greater convenience for the consumer.

  • Executing this would likely involve partnering with commerce destinations, indexing their inventory, and building the technology rails to handle transactions and post-transaction experiences.

But one could argue that Google and OpenAI will also have a strong right to play here, since they could offer similar capabilities directly through their LLM APIs (like they’ve done with web search, tool use, etc.)

These are my current mental models around agentic shopping in fashion, and I suspect similar patterns will emerge across beauty, home, and travel categories. As with any evolving field, these learnings are imperfect, and I’m certain that the future will differ in many subtle (or even structural) ways from what I’m projecting here. If you have a different read on the facts or implications, I’d genuinely appreciate your thoughts and feedback.