Pretty product pages? Great. But they need to give AI agents the info they need.

I watched an AI Overview name a competitor for a query a client should have owned. Same category. Similar price band. Weaker photography. The Overview quoted fibre content, country of make, care instructions, and a stockist region from the other brand’s product page. Our client’s page had a mood paragraph, three lifestyle shots, and a size chart.

The product was fine. The page gave the model almost nothing it could defend.

That is the gap I keep finding on established fashion and homewares brands: the offline business is solid, the photoshoot is expensive, and the product page still reads like a lookbook caption. Humans will forgive that when they already love the brand. AI Overviews and ChatGPT answers will not. They assemble sources they can stand behind. Plain facts travel. Atmosphere stays on the site.

I wrote earlier about the site-level pages I open first when a brand wants answer engines to stop skipping them. About, proof, commercial clarity, consistency. This piece goes one level down: the product and collection pages where ecommerce brands either become citable or stay decorative.

What an Overview is doing when it skips you

An Overview is not browsing for inspiration. It is building a short answer a stranger will trust without clicking ten tabs. For a product query, that usually means materials, origin, who it suits, how it ships, what makes one option different from another, and whether the seller looks like a real business.

If your page leads with “elevated essentials for the modern wardrobe,” the model has a feeling and no facts. If a quieter competitor states “100% European linen, made in Portugal, ships Australia-wide in 3–5 business days, designed for wholesale and retail,” the model has lines it can reuse. Citation follows the reusable line.

This is why your next customer might not be a person in the first pass of research. A model, an assistant, or a buyer’s AI tab may filter the field before a human ever lands on your homepage. Product data is the filter.

The facts I put on a money product page

When I rewrite a hero product for citation readiness, I focus on answers a careful buyer, or a model acting for one, would ask out loud. Keyword stuffing is irrelevant if those answers are missing.

What it is. Fibre, fill, finish, dimensions, weight where it matters, scent notes for wellness, ingredients for food and body. Write the catalogue version first. Then you can add the brand voice around it.

Who makes it and where. Factory region, studio location, “designed in / made in” if those differ. Vague “ethically made” without place is weaker than a plain country and a short process note you would defend to a wholesale buyer.

Who it is for. Size range, fit notes, room type, use case, wholesale vs retail path. Lifestyle brands hide this under photography. Models prefer a sentence a human could quote in a group chat.

How you buy it and what happens next. Price or clear path to price, shipping regions, lead times, returns in one short block. Answer engines punish mystery here because a wrong shipping claim is a bad citation.

One checkable differentiator. Stocked by named retailers, used in a documented project, tested to a named standard, reordered by a segment you can describe without inventing percentages. One real differentiator beats five adjectives.

I keep the brand story. I stop letting the brand story occupy the only paragraph above the fold.

Collection pages are citation surfaces too

Product brands pour effort into individual SKUs and leave the collection page as a grid with a one-line intro. Collection pages are often the URL an Overview can use when the query is categorical: linen bedding Australia, ceramic dinner sets for restaurants, kids’ waterproof shells.

A useful collection intro does four jobs in under 150 words:

• Names the range and who it is for
• States materials or make method at range level
• Notes shipping or wholesale availability if that is a real advantage
• Points to the two or three hero products with a factual reason each

Then the grid can be beautiful. Beauty without those lines is a catalogue with the labels peeled off.

Claims that survive a sceptical read

I use the same standard on client product copy that I use on McKee Creative’s own service pages. Weak claims sound warm and die under scrutiny. Stronger claims sound almost dull and travel further.

Weak: “Our bestsellers are loved worldwide.”

Stronger: “This jacket has been in continuous wholesale reorder with Australian stockists since 2021. Online, the same cut is sold in sizes 6–18 with free exchanges inside Australia.”

The second version can be checked against invoices, size charts, and a returns policy. The first version cannot. Answer engines lean toward the checkable line when they have a choice. I use the same pattern on service proof for the Brand System Diagnostic inclusions page, where fee, scope, and boundaries sit in plain language so a buyer or a model can quote them without guessing.

If you would not put the sentence in a wholesale line sheet, it probably should not be the only sentence on the product page either.

Where AI-written product copy goes wrong

A lot of product descriptions I audit now were drafted by a model and lightly polished. The tell is consistent: balanced paragraphs, symmetrical benefits, and almost no numbers a warehouse or a pattern-maker would recognise.

That pattern is expensive twice. Humans bounce when they cannot find fit, fabric, or shipping. Models skip the page when they cannot extract a defendable attribute. I have written about marketing written by a machine and approved by nobody. Product pages are where that failure shows up as lost discovery, not only as soft brand feel.

The fix is unglamorous. Pull the tech pack, the supplier email, the care label, and the shipping rules into the description. Let the model draft from those sources if you want speed. Do not let it invent a personality paragraph in place of the facts.

A one-week pass for a live catalogue

A full site rebuild can wait. Start with the pages that already make money.

Pick your top five revenue products and your two busiest collection pages. For each product, add or correct: materials, origin, dimensions or size system, care, shipping region, and one proof line you will stand behind. For each collection, write a short factual intro and name the hero products with a reason. Align the same facts on the product schema fields and the visible copy so the page does not contradict itself.

Then run three real queries a customer would type into ChatGPT, Perplexity, or Google with an Overview: category plus use case plus country. Read which brands get named and which sentences get reused. Compare those sentences to yours. The difference is usually specificity, not budget.

Broader generative engine optimisation still matters: topical articles, entity consistency across the web, crawlable structure, proof pages at brand level. None of that rescues a money SKU whose only public text is a mood board caption.

When the catalogue is not the real problem

Sometimes the product pages are already clear and the brand still never appears in answers. That is when I stop patching SKUs and read the whole marketing setup: whether proof exists anywhere, whether the brand entity agrees with itself across Google Business, About, and wholesale materials, whether content exists for the questions buyers ask before they reach a product URL, and whether paid, email, and organic are sending mixed stories about what the brand is.

That full read is the Marketing Diagnostic. It is a $2,500 one-time audit of ads, email, SEO, and strategy together, including whether you are set up to be cited or only set up to look good in a browser tab. Documented client results sit on the case studies page when you want proof before you book.

I still open Chrome the old way. Type the query. Read what the Overview quotes. If the quote could have come from your tech pack and did not, the work is on the product page first. Fix that before another campaign, another blog calendar, or another round of adjectives on a page that already had too many.