Garbage in, garbage out: If AI search doesn't understand your product, it won't recommend it

This phrase is almost as old as computers themselves: "Garbage in, garbage out." And yet, it has arguably never been more relevant than it is today.

Because the way customers discover, evaluate, and purchase products is undergoing a fundamental shift. Google is no longer the only place where a product search begins. Users are asking AI systems like ChatGPT or Microsoft Copilot for recommendations, having them compare products, and increasingly expecting concrete answers rather than a list of search results.

And the next step is already on the horizon: Agentic Commerce. In the future, it won't just be humans searching for products. AI agents will conduct research, compare products, verify requirements, and—with the appropriate authorization—even prepare purchase decisions or execute transactions.

This changes a fundamental question in e-commerce:

No longer just: How do I get my product on page 1?

But increasingly:

Can a machine understand my product, compare it with others, and identify it as a suitable option?

This is exactly where product data quality becomes a strategic priority.

From SEO to GEO: Products must not only be found, but understood

For years, SEO has revolved around one central question: How can my content be found by a search engine and ranked as high as possible?

With Generative Engine Optimization (GEO) comes a new dimension. Generative systems must not only find information but also interpret, categorize, and connect it to generate a concrete answer. This is especially true for products.

Imagine a user who doesn't just search for "office chair," but asks:‍

"Which ergonomic office chair is suitable for someone who is 1.90 m tall, works from home eight hours a day, and has a budget of 300 euros maximum?"

This is no longer a classic keyword search. An AI system must evaluate and compare product information: Is the chair suitable for people who are 1.90 m tall? What is the maximum weight capacity? Is the lumbar support adjustable? Are the armrests adjustable? What seating positions can be set?

The quality of the answer therefore depends significantly on what information is available about the products—and how complete, clear, and structured that information is.

This shifts the importance of product data: ‍

Products must not only be discoverable. They must be machine-understandable.

The problem: Good products do not automatically have good product data

That sounds trivial at first. In practice, it is one of the biggest challenges in digital commerce.

Companies often have enormous amounts of product information. The problem is not necessarily that there is no data. The problem is that it is distributed across many sources, structured differently, and sometimes incomplete.

Information comes, for example, from supplier files, Excel spreadsheets, PDFs, manufacturer portals, websites, catalogs, or existing systems. One supplier describes a product in their own way, while a marketplace requires a completely different structure. And sometimes, the very information that is crucial for the purchasing decision is missing.

Let's take a really good ergonomic office chair. It features adjustable lumbar support, 3D armrests, adjustable seat tilt, a weight capacity of up to 120 kilograms, and is suitable for people up to 190 centimeters tall.

However, the supplier only provides:

"Office Chair 3000. Black. Mesh. 17.4 kg. Modern office chair with ergonomic design."

The product may be excellent. For an AI system, however, it is still difficult to assess its actual suitability.

Now, a customer asks their AI shopping assistant:

"I'm looking for an ergonomic office chair for someone 1.90 m tall who works from home for eight hours a day. It should have an adjustable lumbar support and armrests, and cost no more than 300 euros."

The AI finds two products. Product A is described simply as a "modern office chair with ergonomic design." Product B, on the other hand, contains specific information regarding height, weight capacity, lumbar support, armrests, and other relevant features.

Which product is the agent more likely to identify as a suitable option?

Product B.

And here is the crucial point: Product A might actually be the better product.

The AI just can't reliably recognize it as such.

The problem isn't that Product A is bad. The problem is that its product data says too little about its actual features and suitability. If information isn't reliably included in the available product data, it is difficult to incorporate into a robust product evaluation.

Better product data doesn't guarantee a better ranking or an AI recommendation. However, it does create an essential prerequisite:

A product can only be reliably found, understood, compared, and recommended if the relevant information is present, consistent, and machine-readable.

Product data quality thus becomes a component of digital discoverability and decision-making capability.

Agentic Commerce: When machines don't just find products, but select them

With Agentic Commerce, this trend shifts even further.

Today, humans make many decisions themselves. They search, open product pages, compare features, check reviews, and then decide which product to buy.

An AI agent could increasingly take over this process.

A user could, for example, state: "Find me an office chair for up to 300 euros, suitable for a height of 1.90 m, with adjustable lumbar support and armrests. It should be delivered within five days and have at least a four-year warranty."

An agent must derive concrete criteria from this and evaluate products based on those criteria.

A general product description is not enough for this. The agent needs the most precise answers possible to questions such as: What height does the manufacturer recommend? What is the weight capacity? Which features are actually included? Which variant has which properties? How long is the warranty? Is the product currently in stock? When can it be delivered?

This fundamentally changes the role of product data.

In traditional commerce, product data describes a product. In agentic commerce, it increasingly provides the basis for decision-making that an agent can use to evaluate and select products.

AI-ready product data does not simply mean more content

This is where a common misunderstanding arises. AI-ready product data does not mean producing as much text as possible.

It is much more about structured, consistent, and reliable product knowledge.

For an AI agent, a specific attribute can be more valuable than an additional paragraph of marketing copy. Whether an office chair is suitable for people up to 190 centimeters tall, whether its lumbar support is adjustable, or whether it can support 120 kilograms are pieces of information that can be directly incorporated into a product selection.

This makes it clear:

‍Product data quality is not just a question of data volume. It is a question of usability.

Important quality dimensions are therefore completeness, accuracy, consistency, clarity, structure, and timeliness. An AI system must be able to rely on the fact that relevant attributes are present, values are correct, units are unambiguous, and information is not contradictory across different systems and channels.

This is not about building a separate database for every new technology. On the contrary: the central product database must be consistent and flexible enough to serve different commerce applications.

An online shop requires different information and structures than a marketplace. A product comparison requires different data than a catalog. And an AI agent may require particularly precise attributes, units, and relationships between products and variants.

The central challenge is therefore:

‍How can a consistent and reliable product model be created from heterogeneous raw data that can be used for various commerce applications?

Why a modern PIM is becoming more important

This is exactly where a modern PIM system comes into play.

Today, a PIM should do far more than just store and manage product data centrally. It should be able to ingest product information from various sources, verify data quality and completeness, identify missing or conflicting information, and use AI to enrich and structure data.

Equally important is the ability to prepare a consistent data foundation for various shops, marketplaces, and commerce platforms. Since every channel has its own categories, attributes, and requirements, product data must be flexibly mapped to different structures without requiring a manual data project for every new output.

The future of Product Information Management therefore does not lie in simply storing data centrally.

It lies in turning heterogeneous information into reliable, structured, and reusable product knowledge.

Product Data Quality is becoming the foundation of digital product visibility

SEO, GEO, and Agentic Commerce may seem like three different topics at first glance. In reality, they are increasingly built on the same foundation: high-quality product information.

The requirements are constantly evolving.

For SEO the goal is to make products discoverable and understandable for search engines. For GEO the focus is increasingly on ensuring that generative systems can understand, categorize, and include products in relevant answers. And in Agentic Commerce product information must be so reliable and structured that AI agents can evaluate and select products based on specific requirements.

With each of these developments, the importance of Product Data Quality increases.

This does not mean that Product Data Quality alone determines visibility or revenue. Search algorithms, brand awareness, price, availability, reviews, and many other factors continue to play an important role.

But one thing is becoming increasingly critical:

If machines are to understand products and make decisions based on product information, the quality of that information must be spot on.

The next generation of commerce will therefore not only be shaped by better search algorithms and more powerful AI models.

It will also be shaped by the quality of the product knowledge depend on, which these systems operate on.

Product Data Quality is thus shifting from an operational data management task to a strategic prerequisite for digital discoverability, AI discovery, and agentic commerce.

‍