"Premium", "eco-friendly", "clinically proven" and "built to last" give an AI assistant almost nothing it can repeat, compare or defend. When a shopper asks for the best linen duvet cover for hot sleepers, the answer is built from facts: weave, weight, care, size range, price, delivery, returns. Product pages that state those facts plainly are the ones an assistant can describe accurately.

This matters more on Shopify than it used to. Shopify's Catalog API, which feeds product data to AI channels, marks fields such as the description, top features, technical specifications and unique selling point as inferred by machine learning, with accuracy that depends on the product data you provide. An agent may be reading a machine's summary of your page. What you write is the raw material for that summary.

Why do adjectives disappear from AI answers?

An assistant has to justify its answer, so it leans on statements it can check and compare across sources. An adjective without a boundary cannot be compared: "premium" means nothing next to a competitor's "100% long-staple cotton, 300 thread count". Specific facts survive the trip into an answer; unsupported praise tends to be dropped or softened.

A May 2026 study makes the point with numbers. Researchers ran 252,000 controlled comparisons across six AI models, using product review pages with the brand names removed, and changed one content factor at a time. Relevance to the question and position mattered most.

An explicit price and a recent date helped consistently. Completeness and trust cues, such as backing claims with tests or certifications, added smaller gains, while formatting-only edits made little difference.

What does a claim an AI can repeat look like?

A repeatable claim names what it is about, gives a number or a named standard, and points to where the proof lives. You do not need to cite sources on every line of a product page, but every claim should be one you could back up if a customer, a retailer or a regulator asked tomorrow.

The figures in the table are placeholders to show the shape of a good claim, not facts about any product. If you cannot fill in the right-hand column for a claim, the honest move is to say less.

What changed for green claims in the EU on 27 September 2026?

From 27 September 2026, EU member states apply Directive (EU) 2024/825. It bans generic environmental claims, such as "environmentally friendly", "eco-friendly", "green" or "biodegradable", unless the trader can demonstrate recognised excellent environmental performance relevant to the claim. It also bans offset-based "climate neutral" style claims about a product, and sustainability labels not based on a certification scheme or set by public authorities.

Among the practices it adds to the EU's blacklist of unfair commercial practices, four concern environmental claims directly:

• Displaying a sustainability label that is not based on a certification scheme or established by public authorities.
• Making a generic environmental claim without recognised excellent environmental performance relevant to the claim.
• Making an environmental claim about the entire product or business when it only concerns one aspect.
• Claiming, on the basis of greenhouse gas offsetting, that a product has a neutral, reduced or positive impact on the environment.

This is a summary, not legal advice; if you sell into the EU, have your claims reviewed against your member states' national rules. For GEO, the useful observation is that the same precision regulators now require is what AI assistants can repeat. "Outer made from 60% recycled polyester (GRS certified)" is the kind of specific, checkable claim both are looking for; "eco-friendly jacket" is neither.

How do you sharpen product pages without changing your brand voice?

Keep the voice in the story and put the facts where a shopper and a machine both expect them. Most established brands do not need a rewrite of their tone; they need a consistent facts block on every product page and the discipline to keep it identical across the store, feeds and marketplaces.

1. Keep your opening copy. The paragraph that sounds like your brand can stay. Add the facts beneath it rather than replacing it.
2. Add a facts block. Materials, dimensions, weight, fit notes, care, origin and certifications in a short list or table, in text rather than inside images.
3. Name the standard or the source. A certification by its name and scope, a test by its method, a study by who ran it and how many people took part.
4. Price and date. Keep the price visible in text, and update the page date only when the facts genuinely change.
5. Make it match everywhere. The same material percentages and sizes on your product page, your Google & YouTube channel data, marketplaces and retailers' listings.

A 20-product audit you can run this week

Take the 20 products that earn you the most and read each page as a skeptical shopper who is comparing you with a named competitor. Every time you have to guess, you have found a fact an assistant also has to guess. Fix those first, then work down the catalog in order of revenue.

1. Could a stranger compare this product with a competitor's using only your page?
2. Is every material, measurement and care instruction in text?
3. Does every environmental or health claim have a named basis, or should it go?
4. Are sizing and fit guidance specific enough to choose a size without asking you?
5. Do the price, stock and delivery promise match what your feeds say?

The pages around your products matter too: size guides, care guides and returns pages are often what AI answers cite, and the practical guide to getting recommended covers the rest of the path.

FAQ

Should I remove every adjective from my product pages?

No. Adjectives carry your brand's voice and help people feel the product, and there is no reason to lose that. The fix is to pair each descriptive phrase with the facts behind it, so an assistant has something concrete to repeat.

Do I need to cite sources on a product page?

Not for ordinary specifications like dimensions or materials. For claims that go beyond the product itself, such as clinical results, environmental benefits or certifications, name the standard, the lab or the certifier, and link to a page with the details. That is what makes the claim checkable.

Does the EU green-claims ban affect stores outside the EU?

It applies to commercial practices aimed at consumers in the EU, so a Shopify brand selling into EU markets needs its claims to comply there. Brands elsewhere are not bound by it, but specific, provable claims are still the ones AI assistants can repeat with confidence.

What if a claim is true but hard to prove publicly?

Say what you can show. "Made in our own workshop in Porto" is provable; "the most sustainable workshop in Europe" is not. A smaller claim that an assistant can repeat accurately is worth more than a larger one it has to hedge.

Sources

• Shopify.dev, Catalog API: Search (inferred description, top features, technical specifications, unique selling point)
• Vishwakarma, Kumar and Jamidar, What Gets Cited: Competitive GEO in AI Answer Engines (arXiv, May 2026)
• Directive (EU) 2024/825, Empowering Consumers for the Green Transition (applies from 27 September 2026)
• Ireland, Department of Enterprise, Tourism and Employment, Empowering Consumers for the Green Transition