Your product content now has two audiences: Humans and machines

How AI became a buying audience, not just a tool

Adobe Analytics tracks over 1 trillion visits to U.S. retail websites, and its data indicates that traffic from generative AI sources grew 393% year over year in the first quarter of 2026. While the growth is noteworthy, a more valuable insight lies in understanding how these visitors behave after they arrive.

According to Adobe’s behavioral data, shoppers who come from AI sources tend to browse more, stay longer, and leave less frequently than visitors from traditional channels. The pattern suggests that these AI-driven visitors arrive already informed and focused on their shopping experience.

• 393% year-over-year growth in AI-sourced traffic to U.S. retail sites, Q1 2026
• 48% longer time on site for AI-referred shoppers compared to non-AI traffic
• 27% lower bounce rate for shoppers arriving from generative AI sources

Numbers like these transform what an AI assistant means for your business. Unlike a search engine that displays your content and leaves the judgment to users scrolling through the results, an AI assistant evaluates your content, compares it against competing products, and ultimately decides whether your listing appears in the response. In other words, the machine has evolved from a distribution channel to an evaluator that precedes human judgment.

BCG’s consumer research points in the same direction. GenAI assistants and chat tools ranked as the second most influential touchpoint among consumers who used them during their purchase journeys, and they were identified as the single most influential touchpoint for daily users. This level of influence operates upstream of your conversion metrics. If a product is not presented by the AI, it is unlikely to be considered by the consumer at all.

What human buyers still need from product content

Only 29% of consumers currently base their purchase decisions on AI recommendations, according to the EY Future Consumer Index. In contrast, 45% of consumers still discover new products through in-store displays, compared to just 17% who find them through online recommendations. 

When we consider these numbers alongside adoption data from earlier research, a clearer picture emerges. While machines increasingly influence which products enter the conversation, consumers ultimately decide which products they buy, and these two decisions rely on different factors.

A Gartner survey of U.S. consumers conducted in January 2026 provides further insight, revealing that people are more open to using AI tools for discovery and research than for tools that make purchasing decisions on their behalf. 

Put simply, consumers are willing to accept a machine’s shortlist of options but will apply their own judgment to make the final decision. This judgment considers aspects a shortlist cannot convey, such as how the product fits their context, whether the product description resonates with their situation, and whether the brand inspires enough confidence for them to commit.

Therefore, different audiences engage with the same content in various ways, and the table below illustrates where their requirements diverge.

Notice that nothing in the left column conflicts with anything in the right. Humans and machines reward different qualities in your product content, but they never reward opposite ones, and that distinction carries the rest of this article.

What machines read when they recommend a product

Researchers at Luiss University in Rome conducted three experiments to determine whose product recommendations consumers are more likely to follow, comparing an AI source with a human expert. The first study involved 485 participants from the U.S. and U.K. 

Published in Electronic Commerce Research (Mazzù, Andria, Baccelloni, and De Angelis, 2025), the studies revealed that consumers viewed AI as more transparent and credible than human experts when it recommended search products, meaning items whose attributes can be assessed before purchase. The perception led to a greater intention to follow the AI’s recommendations. However, no such advantage was seen for experience products, which can only be evaluated after use. 

This pattern is significant, especially since most product content in commerce, and nearly all in B2B, pertains to search products. Specifications, dimensions, compatibility, and compliance are all examples of verifiable information that the study’s participants trusted AI to evaluate.

Transparency and credibility are qualities AI can demonstrate only when your content provides something to verify. Before an assistant can confidently recommend your product, it must find the following:

• Stated fit and constraints: An assistant responding to the question “Will this work for X?” needs a written answer, as it cannot infer information that isn’t explicitly stated.

• Complete attributes in a machine-readable format: Missing or loosely defined specifications are perceived as gaps, and competitors will fill them by clearly articulating their information.

• Consistent facts across all sources checked: If an assistant encounters different weights or conflicting compatibility claims, it cannot trust either source.

Unfortunately, much retail content fails to meet these requirements before verification even starts. An Adobe analysis of various U.S. retail websites found that a significant portion of their pages is unreadable to LLMs. Adobe now benchmarks this issue with a dedicated AI visibility checker. 

Unreadable pages reflect a familiar problem in a new guise: in many organizations, the product data already exists, but AI can’t read it. Understanding what clean, structured data entails and why AI search relies on it is a critical area of focus.

Do you need separate content for AI and for people?

Google directly addressed this question in its official guidance on generative AI features in Search. The response contrasts with much of the current advice available. Instead of establishing a new discipline, the guidance clearly distinguishes what enhances visibility from what isn’t worth your time. Here’s what they say is worth doing and skipping:

This discipline has developed its own vocabulary. There is an ongoing debate about whether AEO is the new SEO for product discovery, or if GEO should take priority in your content strategy. 

However, Google’s stance is that optimizing for generative AI search is still fundamentally about optimizing for search, as the generative features rely on the same core Search systems as everything else. Notably, Google’s recommendations don’t feature any machine-specific tasks.

Regardless, the practical takeaway for your product content remains unchanged. You should not create separate versions for humans and machines. Instead, maintain a single set of product content that presents accurate, consistent information, crafted to persuade both people and machines. 

The real challenge lies not in the writing itself, but in the operations behind it, where most teams feel the most pressure.

How do you make product content that works for both audiences?

Examine where your product information is stored to gain clarity on the work ahead. Engineering specifications are found in PLM systems, operational details in ERP systems, and customer-facing content is stored elsewhere. 

Meanwhile, channel formats and buyer expectations are constantly evolving, putting pressure on businesses as agentic commerce takes hold in B2B transactions. The following steps create a cohesive flow of product data that both audiences can trust: 

1. Consolidate your product information before refining your product story.

Both audiences rely on the same foundational data, so the first task is to gather complete and consistent information in a single, governed location. A PIM platform with a flexible data model, such as Inriver, can ingest product data from your existing sources as-is, without the cleanup or complex ETL projects that often hinder these initiatives. The data becomes quickly usable and ready for AI applications.

 2. Document the answers explicitly

Machines cannot infer the meaning of content that is not clearly stated, so include important details such as fit, constraints, and qualifications directly in the content instead of leaving them in a product manager’s mind. Address the questions a buyer would typically ask, such as compatibility, dimensions, materials, and compliance. Also, ensure images are optimized for AI visibility.

3. Implement verification between your AI workflows and your channels

Enrichment at scale now relies on AI-assisted workflows, making verification even more critical. Inriver adds a layer of verification and validation to every product-data-based workflow, ensuring that the output is checked against trusted product data before it reaches any channels. This ensures that the information machines process about your products remains accurate as volume increases.

4. Deliver consistent information across all platforms

If a revised specification is updated on your product page but not on your syndicated feed, it may satisfy the human reader but mislead the machine. Since machines process information first, syndicate updates to every channel and AI surface simultaneously to prevent inconsistency.

5. Focus on the human element after streamlining the facts

Once product information flows reliably, the presentation becomes the key differentiator. Earlier comparisons have shown that confidence, context, and imagery influence human buyers. Teams can then shift their efforts from reformatting and firefighting to enhancing the content that truly persuades.

Follow this sequence, as each step builds on the previous one. Establish reliable facts first, governance second, and focus on presentation last.

Is your product content ready for both humans and machines?

Answering that question honestly is more challenging than it seems. You understand what your content means, but an assistant reading it without prior knowledge does not. To start, pull up your best-selling product page and read it as if you were a stranger searching for specific facts. Identify any areas where the information is lacking; these are your starting points for improvement.

You’re not alone in discovering these gaps. The gap between how teams evaluate their product data and how that data actually performs is significant across the industry, as highlighted by Inriver’s Product Data Paradox Index.

Since channels and models will continue to evolve even after you address today’s gaps, it’s essential to establish a feedback loop for your product content, rather than treating it as a one-time fix. The quickest way to see this in practice is to schedule a demo with an Inriver expert, where they show how product data becomes AI-ready without an ETL project, using the sources you already have.