Case Study: Product Pages Rewritten for AI Answers

Most product pages are written for a human who has already decided to shop. But a growing share of purchase research now happens inside AI assistants, and those assistants need a different kind of page. A direct-to-consumer home fitness brand rebuilt its catalog for that reality.

Background

A direct-to-consumer brand sold premium home fitness equipment through its own Shopify store. The catalog ran to about ninety products, from compact rowing machines to adjustable dumbbell sets. The marketing team was competent at paid social and traditional SEO, and product pages ranked for branded and category terms.

What worried leadership was a shift in how customers described their journey. In post-purchase surveys, more buyers said they had asked an AI assistant "what is the best rowing machine under five hundred dollars" or "compare adjustable dumbbells for small apartments." When the team tested those queries, their own products rarely appeared, while competitors with thinner catalogs but clearer pages won the recommendation.

The product pages themselves were the weak link. They led with lifestyle imagery and a price, then manufacturer specs, then scattered reviews, with no plain-language explanation of what each product was for, who it suited, or how it compared to alternatives.

The audit that set the scope

Before rewriting anything, the team pulled the actual questions: the phrases shoppers typed into the site's own search box, and the questions buyers wrote in the post-purchase survey. The overlap became the rewrite backlog, a far more useful brief than improve all ninety pages.

It surfaced recurring question types the catalog had never answered plainly: how much floor space does this need, is it loud in an apartment, can two people use it at once, what is the learning curve, and what breaks first.

Challenge

The brand faced a structural mismatch between how its pages were built and how AI engines extract product information. The pages prioritized persuasive imagery over extractable text: a model scanning one found a hero photo, a price, and a spec table, but little natural-language description it could quote.

Comparisons were missing. Buyers asking AI for the best option for a use case need pages that state fit: who the product is for, who it is not for, and how it differs from others.

Structured data was inconsistent. Some products had Product schema, others did not, and none included rich attributes like weight, dimensions, or use-case that help engines match a product.

Finally, the pages were siloed. Each product stood alone with minimal links to related items or guides, so the site conveyed no coherent catalog intelligence to a model.

The catalog problem nobody had named

A fifth problem surfaced only once the team read pages side by side. Nearly every product described itself with the same generic vocabulary: compact, premium, space-saving, professional-grade. The pages were not wrong, but they were interchangeable, and an engine answering which of these fits a small apartment had nothing to distinguish them on. That reframed the project: for answer engines the catalog is evaluated as a whole, so the pages must form a system where each product holds a distinct slot.

What They Did

The team treated each product page as a candidate answer to a buyer's question, not just a sales asset. The five moves below ran in sequence, chosen so each made the next easier.

1. Wrote a plain-language answer block for every product. Near the top of each page, they added a two-to-three sentence description a model could lift directly: what the product is, the problem it solves, and who it fits. No marketing throat-clearing, just a clean definition.

The team wrote these blocks in a specific voice: short sentences, no adjective that could apply to any competitor's product, and one concrete use case each. A compact rower that folds to a two-foot depth is quotable. A space-saving solution for modern homes is not, because it could describe anything.

2. Added explicit comparison and fit guidance. Each page now included a short "Who it's for / Who it's not for" section and a paragraph on how the product compared to the brand's other options, giving models the contrast language they need for versus questions.

3. Standardized and enriched schema. They rolled out complete Product schema across all ninety pages, including weight, dimensions, intended use, and skill level. Uniform markup let a model parse the range consistently.

4. Built buying-guide hubs. The team created a handful of guides, for example one on choosing a rowing machine for a small home, linking out to relevant products. These hubs became the authoritative pages models referenced, with products supporting them.

5. Added FAQ sections to high-intent pages. Selected product pages received a short FAQ answering the practical questions buyers asked: assembly time, space needed, compatibility with existing gear. This captured long-tail, intent-rich queries. The FAQ blocks came last, because the product team had to approve the numbers.

How the work was executed

Ninety pages is a lot to rewrite well, and the first instinct, writing every block by hand, stalled immediately. The hard part was not typing sentences, it was deciding what each product's honest position actually was. Several products genuinely competed, and the team had to resolve which of two overlapping ones to recommend for a given buyer.

So the work split into two passes. First, a short internal brief per product, stating the one-line definition, the primary buyer, the disqualifying condition, and the closest in-catalog alternative. The second pass applied the template. That concentrated the hard decisions into a few hours per product while the bulk of the writing became fast and consistent. Schema went in one pass, the hubs came last.

Results

Within two months, the brand's products began appearing in AI-generated buying recommendations for several target queries. On "best rowing machine for small apartments," a guide hub plus two product pages surfaced in Perplexity's answer, the brand's first citation there.

The FAQ sections proved especially effective. Product pages with FAQs saw a noticeable lift in assisted conversions, the team's term for orders that began with an AI-assisted research session and closed through a direct visit. The brand could not attribute every sale to a citation, but the signal was clear: rewritten pages converted better than the old spec-led versions.

Most valuable, the enriched schema improved the brand's traditional rich results too, tightening the link between GEO work and existing search performance.

How the team tracked it

The tracking combined three self-reported signals, each imperfect, and the team was careful not to over-read any of them. The first was a weekly query audit: a fixed list of buyer questions, drawn from the same source as the original audit, run through three engines, recording whether the brand was cited, which page was cited, and whether a competitor appeared instead. The list never changed over the quarter, because changing questions makes a trend unreadable.

The second was a self-reported checkout field asking how the shopper first heard about the brand. It was easy to game but confirmed direction.

The third was an assisted-conversion rate on rewritten pages, compared against a control group not yet touched. Holding back a slice of the catalog as a control is what let the team claim the lift.

Lessons

Write for the answer, not just the sale. A page that explains clearly will sell and get cited; one that only persuades will do neither well.

Structure your catalog as a system. Buying-guide hubs gave models a stable reference point, with product pages as supporting evidence. Siloed pages leave engines guessing.

Schema is the silent partner of GEO. Consistent, attribute-rich markup does quiet work that pays off across AI answers and classic search.

Fit guidance is a citation magnet. Stating who a product is for and who it is not for gives models the contrastive language they prefer to quote.

A reusable catalog rewrite template

• One plain-language definition, two or three sentences, above the imagery block.
• One named primary buyer, in the words that buyer would use.
• One disqualifying condition, so the page can lose a comparison and stay useful.
• One named in-catalog alternative, with the reason to choose either.
• At least three FAQ entries drawn from real support questions.
• Product schema generated from the same fields the page renders.
• A link from a buying-guide hub, and a link out to the closest alternative.

Frequently Asked Questions

Do product pages need different copy for AI than for humans?

Not entirely different, but the order and clarity matter. The same page can serve both audiences if it leads with a plain-language definition and fit guidance before the persuasive and visual elements. Humans skim the answer and then browse; models lift the answer. One page, structured for both.

Will adding FAQ sections to product pages hurt conversions?

In this case study they helped. Short, genuinely useful FAQs answer the practical objections buyers carry into research, and they give AI engines quotable content for long-tail questions. Keep them tight and specific to the product rather than generic filler. If a question has no honest answer, leave it out rather than padding the section.

How much schema is enough for product GEO?

Enough to describe the product the way a shopper would ask about it: category, key attributes, intended use, and differentiators. You do not need every possible field, but consistency across the catalog matters more than completeness on one page. Uniform markup lets engines parse your range as a whole, which is what lets them recommend one product for one buyer.