The rise of commerce driven by artificial intelligence is transforming how products are discovered, evaluated and purchased.

Increasingly, consumers are turning to generative AI tools to ask complex, conversational questions and are expecting precise, trustworthy answers in return. In fact, nearly one in four consumers have already used a generative AI tool to search for a product instead of a traditional search engine.

This shift exposes a longstanding issue that supply chain and data leaders can no longer ignore: the fragility of product data governance. For years, product data has been fragmented, inconsistent and unverified, and often sourced from multiple vendors, legacy systems and manual inputs. 

In traditional e-commerce environments, these gaps were manageable. But in AI-powered systems, they become amplified.

For years, brands could engineer their way to the top of search results: more ad spend, smarter SEO, better marketplace tactics. AI-driven discovery doesn't work that way. These systems don't reward the brand with the biggest budget; LLMs are optimized for relevance in their output – and that relevance comes from answers grounded in deep, trustworthy data. If, as far as the LLMs are concerned, your product information is incomplete, inconsistent or unverified, no amount of spend compensates for it. Visibility is now earned through data integrity.

We see examples of these new failure modes every day: LLMs recommending products that have been out of stock for years (leading to frustrating 404s for the shopper) and mixing up marketing claims and reviews between SKUs (leading to loss of shopper trust), to name a few.

AI generates recommendations, comparisons and summaries based on the data it ingests. When that data is incomplete, outdated or inaccurate, the outputs can be misleading or outright wrong. Large language models often attempt to fill in the gaps when data is missing, and when they guess wrong, the result is misinformation that damages consumer trust and brand credibility alike.

The rise of fake and AI-generated reviews makes this harder. In 2024 alone, Amazon blocked or removed more than 275 million fake reviews and spent over $500 million — hiring 8,000 employees — to combat them. 

With 75% of consumers already concerned about fake reviews, unverified data becomes a reputational issue. Research shows that humans and AI systems alike struggle to distinguish real reviews from fabricated ones, and the problem compounds as AI generates more of them at scale. More than half (54%) of consumers believe brands themselves bear primary responsibility for solving fraudulent content online — a clear signal that trust is a brand imperative. Without effective validation mechanisms, unreliable data corrupts the AI ecosystem quietly and quickly.

This is where real-world, validated data becomes essential. In an AI-driven landscape, authenticity is a foundational data signal. Verified, experience-based content like authenticated customer reviews gives AI systems a continuous stream of grounded, real-world insight. These signals capture nuances that structured product specs miss: how a product actually performs, where it falls short, and what edge cases matter to real customers.

Many brands are scrambling to "AI-proof" their pages without realizing the answer is already in front of them. The reviews and user-generated content (UGC) they've spent years collecting are precisely what AI systems are looking for. Research from Columbia and Yale found that products with both high ratings and high review volumes receive the strongest preference boost from AI agents. The content brands have already built is becoming a direct input into AI-driven product discovery.

Many retailers and brands have begun experimenting with on-site AI-powered shopping assistants. These are designed to offer the convenience of AI-driven shopping, within the walled garden of a retailer or brand site. Of course, the assistants are only as good as the system architecture behind them; it’s not sufficient to simply wrap a frontier LLM in UI and call it a day. To be effective, these assistants must ground their answers in rich, granular product data – including the trust signals (like ratings and reviews) that LLMs rely on when making product recommendations. This type of   retrieval-augmented generation (RAG) system is only as effective as the data it retrieves. If the underlying sources lack integrity, the outputs reflect those same weaknesses. Strengthening product data governance must be a prerequisite for successful AI adoption. 

The era of hacking the algorithm is over. The era of owning the truth has begun. For data leaders, that means three things: 

• Verify at the source. Product information and customer feedback must be authenticated before they enter the ecosystem. 
• Break down silos so AI systems get a unified, machine-readable view of the full product experience. 
• Treat data enrichment as an ongoing discipline, not a one-time project.

None of this requires starting from scratch. The data most brands need to compete in an AI-driven world already exists, and it's sitting in years of customer reviews, product feedback, and authenticated UGC. The work now is making sure it's clean, connected and trusted enough for AI systems to use it. That's a data-governance problem.

Nick Shiftan is chief technology officer of Bazaarvoice.