Product Discovery Is Breaking; and AI Is About to Make That Impossible to Ignore
Product discovery is quietly becoming one of the biggest constraints on growth in ecommerce and marketplaces. At the same time, the way people start shopping is shifting dramatically: AI-driven search, assistants, and early “shopping agents” are beginning to sit between customers and brands. If your discovery stack was built for keyword matching and simple collaborative filtering, this creates a widening performance gap.
This post explores how consumer discovery behavior is changing, where legacy recommendation systems break down, why AI-powered search and autonomous agents make these weaknesses more visible, what capabilities companies need to stay competitive, and how NavOut helps teams bridge the gap.
1. Product discovery has moved from browsing to asking
For nearly two decades, online shopping followed a predictable pattern: enter a keyword, scan a results grid, adjust filters, compare tabs, and decide. The emerging pattern is different. Consumers increasingly begin with a question or an instruction, and AI systems interpret intent and compress the decision space into a few highly relevant options.
AI-powered search is becoming the new “front door”
McKinsey’s 2025 analysis shows that around half of global consumers now prefer AI-powered search tools such as ChatGPT, Gemini, and AI Overviews to understand products and compare alternatives [1]. AI search is projected to influence more than $750 billion in U.S. consumer spending by 2028, and brands that fail to adapt may see a 20–50% decline in organic visibility as AI answer surfaces replace traditional results pages [1].
These systems don’t simply route people to links—they generate summarized, personalized, ranked answers.
Generative AI is increasingly mediating shopping decisions
Bain’s research finds that 30-45% of U.S. consumers already use generative AI to research or compare products, and a growing portion now begin holiday shopping on AI platforms rather than retailer sites [2].
Google has accelerated this transition. Its 2025 AI shopping rollout enables users to describe what they want conversationally, explore AI-curated comparisons drawn from tens of billions of product listings, and rely on AI to call stores, check inventory, and even make purchases when conditions are met [3],[5]. Meanwhile, Google reports that visual and AI-driven discovery surfaces now support billions of shopping moments per day and increasingly function as the bridge from browsing to purchase [4].
For an expanding share of consumers, the journey starts with: “Show me the best option.” That is not what legacy recommendation systems were built to understand.
2. Where legacy discovery stacks break
While most ecommerce and marketplace leaders feel the symptoms of poor discovery (think: high abandonment, low relevance, narrow personalization etc.), the underlying structural causes run deeper.
One major limitation is that legacy systems rely on shallow representations of products. Items are treated as collections of attributes, brief descriptions, and sparse behavioral data. Yet the factors that define customer preference—why a jacket performs well in wet conditions, which earbuds fit certain ear shapes, why a laptop suits design work—exist in unstructured text, imagery, reviews, and user-generated content. MIT Sloan research shows that this unstructured expression reflects true customer needs far more accurately than traditional analytics can capture [6].
This mismatch leads to what users often feel instinctively: “These results don’t understand me.”
Another systemic weakness is the handling of new and long-tail products. Collaborative filtering depends on historical behavior. Modern catalogs depend on niche, emerging, or specialized inventory; precisely the items with limited behavioral data. These products remain buried unless a user is unusually persistent.
A final issue is overreliance on static business rules. Many companies lean on hand-tuned boosts, rigid segments, and manual merchandising to simulate relevance. But shopping behavior is now dynamic and distributed, spanning social surfaces, marketplaces, AI search, and real-time triggers. Google’s analysis emphasizes that “ambient shopping” requires systems capable of adapting to rapidly evolving data environments [4]. Static rules simply cannot keep pace.
Together, these constraints create friction, shallow discovery, and unrealized revenue.
3. Why AI search and agents will expose these weaknesses
Retailers once offset weak discovery with heavy merchandising, discounts, or paid acquisition. AI is eroding those buffers.
McKinsey notes that AI-generated answer surfaces increasingly mediate consumer journeys, and brand websites constitute only a small portion of what these models read and summarize [1]. Products with weak descriptive signals or inconsistent data are less likely to appear in AI-generated responses.
Agentic AI accelerates this reality. Bain reports widespread retailer experimentation but slow maturity; meanwhile, consumer adoption of agentic behaviors (automated comparisons, stock checking, price monitoring, autonomous checkout) is already growing [2]. Separate analysis shows that over two-thirds of retailers have at least partially deployed agentic AI internally, even if unevenly [7].
Google’s new autonomous shopping capabilities illustrate what this looks like operationally: AI agents can now call stores, confirm inventory, navigate alternative options, and complete transactions, all without the user manually browsing [5].
In this environment, your internal discovery quality governs:
How clearly AI systems can interpret your catalog
How confidently agents surface your products
Whether you are recommended, or excluded, when a user asks for help
Weak internal discovery translates directly into weak external visibility.
4. What companies need to compete in this new landscape
Competing in a world where humans and AI systems jointly mediate shopping requires modern discovery infrastructure.
Companies need far richer representations of products. They need products that are structured, multi-modal, and are expressive of functional and aesthetic qualities. Companies will need deeper models of user behavior that capture both long-term preferences and in-the-moment intent. They need to interpret behaviors beyond simple clicks.
Companies also need to implement continuous learning loops. Discovery systems must adapt as inventory, trends, and user intent shift, rather than relying on quarterly or manual updates.
Finally, companies require interfaces designed for AI ecosystems: machine-readable product intelligence, retrieval and ranking APIs, documentation that external AI systems can interpret, and a strategy for GEO (GenAI Engine Optimization). According to McKinsey, GEO is becoming essential for brands that want to remain visible in AI-driven environments [1].
5. How NavOut helps companies bridge the gap
NavOut provides the intelligence layer modern discovery requires. It ingests diverse product data (structured and unstructured) and creates rich, coherent representations that reflect both the functional and contextual meaning of items. It models users through long-term patterns and real-time signals, enabling experiences that adapt as intent evolves.
NavOut exposes this intelligence through retrieval and ranking APIs designed not only for your UI but also for AI systems and agents that require structured, high-quality inputs. And critically, it integrates into existing storefront and commerce infrastructure without requiring a major rebuild.
The result is a dual benefit: smoother, more relevant discovery for users right now; and clearer, more trustworthy signals for the AI systems that increasingly shape demand.
6. Where teams should focus next
If you lead product, data, growth, or category operations, your next step is understanding how your current discovery system performs under the demands of modern AI behavior. How well is your catalog represented? How often do users hit dead ends? How visible are your products in AI-driven discovery channels? How quickly can your system adapt to new signals?
Teams that invest early in discovery intelligence will be the ones best positioned to thrive as AI reshapes shopping. NavOut exists to help with that transition.
Citations
Zidarescu, A. (2025). New Front Door to the Internet: Winning in the Age of AI Search (summary of McKinsey & Company research).
Bain & Company (2025). Agentic AI in Retail: How Autonomous Shopping Is Redefining the Customer Journey. Link
Sato, M. (2025). Google will let users call stores, browse products, and check out using AI. The Verge.
Scott, S. (2025). Retail never stands still. Here’s how marketers can keep pace. Think with Google.
Google (2025). AI Shopping and “Let Google Call” product announcements.
Hauser, J. R., Li, Z., & Mao, C. (2022). Artificial Intelligence and User-Generated Data are Transforming How Firms Understand Customer Needs. MIT Sloan.
Hale, C. (2025). Over two-thirds of retailers have already partially deployed AI agents. TechRadar Pro.
