AI shopping does not start with ChatGPT, but with your product data.

"We have a Product Information Management system, so we are AI-ready" is like saying, "We have an ERP, so our finance is perfect."

Tooling alone says nothing about your data quality.

In practice, we see the same pitfalls recurring with retailers in fashion and electronics:

• Attributes are inconsistent ("cotton", "100 per cent cotton", or simply empty)
• Important context is hidden in marketing copy instead of structured fields
• Every brand or supplier uses its own taxonomy
• Variant structure is chaotic (colour and size are mixed up)
• There are no use-case attributes such as occasion, activity, or climate
• Price, stock, and images are not always current
• Data is pushed blindly to all channels without optimisation
• No one is ultimately responsible for data quality (governance is missing)

Fashion case: invisible during AI discovery

A fashion retailer had tens of thousands of T-shirts in their Product Information Management system. When prompted for a "shirt smart enough for under a suit but not too warm for summer," their products barely appeared. Why? There was no attribute for "occasion," "layering suitability," or "breathability." The AI simply had no evidence.

Electronics case: missing context for AI

An electronics store with 25,000 products was almost entirely missed for AI questions like "compact monitor for working from home with lots of daylight." Attributes like "anti-glare," "small space," and "working from home" were hidden in descriptions or missing completely. The AI could not make the match.

The result in both cases: you are online, but you are invisible in AI-driven discovery: exactly the problem that high-quality product data solves.

The 5 non-negotiables for LLM-ready product data

AI does not ask for more data, but for better structured and contextual data. These are the five absolute must-haves:

1. Consistent taxonomy: Clear categories and values, aligned with Google's product data specification for Merchant Center.
2. Complete, normalised attributes: Everything explicitly recorded in schema.org/Product according to Google's structured data guidelines, nothing hidden in text.
3. Context and use-case attributes: Occasion, situation, combinations, intent. This is the biggest blind spot and the real differentiator.
4. Correct product structure: Clear parent-variant relationships and links.
5. Current and distribution-ready data: Price, stock, and media always up to date.

What is most underestimated? Point 3: context attributes. In the past, this was just "marketing talk." For AI, it is the key to visibility.

How do you measure if your data is AI-ready?

• Data Quality Score (basic hygiene)
• Context Coverage Score (what percentage of your products has real use-case info?)
• Matchability Test (test with realistic AI prompts and see if your products return)

Retailers often score 80 to 90 per cent on basic data, but lower than 20 per cent on context. That is where the gap lies.

"We'll solve it with AI on top" is an illusion

You hear it more and more: "We'll just throw an LLM at it: it will fix our messy data." Wrong.

That is like putting a smart translator on top of an incomplete dictionary. You get beautiful sentences, but the meaning is incorrect. Research from OpenAI into hallucinations and the OpenAI GPT-4 capabilities documentation show that LLMs generate structurally incorrect facts when they lack sufficient grounding.

Consequences:

• Hallucinations: AI invents features like "breathable" or "suitable for office" when it is not stated anywhere
• Incorrect customer expectations leading to higher returns
• Your product is not selected because a competitor has provided context
• You lose control over how your brand is presented

In short: AI does not make bad product data better: it makes bad product data scalable. Real product data starts at the source, not at the model.

How ConnectingTheDots takes a different approach

Most PIM providers say: "We manage product data." We say: "We ensure product data becomes understandable, for humans and for AI."

We do that with one cohesive chain:

1. Import & Onboarding: We retrieve data at the source (suppliers, Excel, APIs, systems) and normalise the chaos before it enters the Product Information Management system. 90 per cent of problems originate here.
2. PIM with strong Data Model & Structure: We don't just build storage, but a data model with business rules, taxonomy, and governance.
3. Controlled AI enrichment: AI adds context based on hard attributes, never ad hoc.

The result? One consistent source of truth ready for your online store, Google Merchant Center, marketplaces, and future AI feeds. No loose tools, no hallucinations, just scalable and interpretable product data.

How does PIM contribute to the accuracy of AI shopping advice?

A PIM system centralises and validates all product information, giving AI models access to a single, consistent source of truth. This prevents AI from using incorrect or contradictory data, resulting in more accurate and reliable advice for the consumer.

Can the PIM system integrate existing product data?

Yes, the Import & Onboarding component of ConnectingTheDots is specifically designed to import, standardise, and enrich product data from various source systems. This process ensures all product information becomes usable within the PIM system and for AI applications.

How does a PIM system prepare my business for the Digital Product Passport?

A PIM system structures and manages a wide range of product information, including origin, materials, and sustainability characteristics. This forms the necessary data infrastructure for creating and managing Digital Product Passports, which are essential for future EU regulations.

What are the consequences of poor product data for AI shopping?

Poor product data leads to inaccurate recommendations, incorrect product information, and a frustrating customer experience in AI-driven shopping. This can result in lost revenue, a damaged brand image, and unnecessary returns.

Why is investing in PIM more important than investing directly in AI for e-commerce?

Without a robust PIM foundation, AI operates on unreliable data, which negates its effectiveness. Investing in PIM establishes the foundation for accurate, complete, and consistent product data, which is essential to harness the full potential of AI applications in e-commerce.

1. Conduct a Context Coverage Audit: Test 20 to 50 realistic AI prompts and see how often your products return and why they don't.
2. Fix taxonomy and variant structure: Make this watertight. This is the foundation.
3. Add context attributes per product family: Start with occasion, situation, and combinations. Use AI to scale this smartly, not manually.
4. Optimise for channels: Specifically prepare your data for Google Merchant Center structured data and feed specifications.
5. Build in governance: Ensure someone owns data quality and that updates happen structurally.

Each step brings your data closer to LLM-ready status.

AI shopping is no longer hype. The shift toward AI-driven discovery is already visible and accelerating. Those who treat their product data as a strategic asset (structured, context-rich, and understandable) will win disproportionately in the new customer journey. Those who wait risk becoming invisible before the customer even reaches their site.

Do you want to know how your product data scores on Context Coverage and AI-matchability? Feel free to contact us for a quick, no-obligation audit. We will help you move from messy catalogue data to LLM-ready decision-making information.

In short: this is not a technology issue but a data issue. Those who get their product data in order build a sustainable competitive advantage in the new AI-driven customer journey.