How to prepare product data for Agentic AI
We recently published a guide on how to prepare your product data for Agentic AI to ensure your products are correctly and widely represented in prompt answers. Now a full article on the topic is published in the March-April 2026 Harvard Business Review.
What are the main insights from the article?
- Stats about online shopping behaviours show that online shoppers are relying more and more on AI to find and buy products:
> 66% of 18-24 years olds,
> 51% of 25-34 year olds,
> 42% of 35-44 year olds,
> 31% of 45-54 year olds
ask AI models for brand, product and service recommendations (YouGov Study) - Every major AI company is developing agents in anticipation of mainstream adoption. Think OpenAI + Stripe/PayPal + Shopify = automated customer journey.
- There are three interaction modes emerging between brands and consumers:
> Brand agents engage directly with human customers to help them explore products, make decisions and access services (e.g. check inventory, schedule test rides, estimate re-sale value, answer leasing questions)
> Consumer agents act on behalf of the potential customer by filling out forms or completing purchases.
> AI agents interact autonomously on both sides of the transaction without direct human involvement - Brands must evaluate which aspects of traditional customer relationships to preserve vs. evolve:
> Decide whether you need an AI Agent:
Do your customers want to interact with an agent? This is often more likely in contexts with low stakes, routine decisions, and predictable outcomes; and often less likely during high-stakes decisions and personally meaningful purchases, such as hobbyists whose personal identity is associated with the purchase.
> Get Customers to Use your Agent:
Focus on capabilities that personal agents cannot easily replicate, such as deep, proprietary product knowledge. The goal should be a consultative, personalized experience. Pair this with a human-in-the-loop model to create trust.
> Make other agents choose your brand:
Develop seamless integration points with the broader AI ecosystem.
What are the key take aways for component suppliers and bike brands?
- If you are selling D2C and/or need to make sure distributors, retailers and online platforms are successfully selling online, ask yourself: How do we adapt our communications strategy when our primary audience may not be human?
- Most consumers do not complete the purchase via AI yet, and I doubt they will buy a couple-thousand-euro-eBike through ChatGPT in the future. However, for lower basket values such as replacement components, they might. And even if not, they definitely use LLMs for prepurchase research. And that is where you definitely want to show up.
- Find out how you can ensure that AI shopping agents select and correctly represent your brand and products. This requires ongoing learning, experimentation, and adaptation. Examples from the article shows that marketing and brand managers continuously test visibility by prompting across different models and recording brand and product performance, as well as using reasoning models (e.g. Perplexity’s R1 model) to understand the process of selecting products and brands after a prompt.
Where can you start?
Start building the foundation in two ways:
- Understand how your customers prompt, as rewording can significantly change a model’s response. This includes and ongoing process of:
-> testing how product information performs across different prompt variations
-> monitoring, through search logs and customer service interactions, the actual phrasing that consumers use.
Use that as the foundation for refining and optimising marketing content. - Get your product data in order. Ensure clear, harmonised and consistent product descriptions, attributes, pricing, and policies across all online representations (your website, your catalogue, your dealers websites, your Shopify store) that make it easy for AI models to read and understand the data.
Clean product data starts way before the marketing department. With a harmonised data structure, single source of truth that connects product development and purchasing with commercial teams, product data can flow across your organisation without a lot of manual work, hundreds of spreadsheets and back and forth to find out what the correct piece of data is.
Companies that rely on fragmented spreadsheets and inconsistent product data risk disappearing from this new AI-driven commerce. Our free guide shows you the first steps in the right direction.
Master your product data with the NOCA portal
Ensure everyone uses the same, consistent, and up-to-date product data across your organisation and the data is synchronised across marketing and sales channels to create less manual work and more revenue.
COMPANY
