A mid-market ecommerce brand in the home and kitchen category

The challenge

The brand sells a single-category lineup of mid-priced kitchen products, competing against both legacy housewares brands and a wave of newer DTC competitors. Their organic channel was healthy, but the team had noticed something specific in the data: revenue from "best [product] for [use case]" queries had dropped significantly through 2025, even though impressions on those queries had increased. The drop matched the rollout of shopping-intent answers in ChatGPT, where users asking "what is the best [product] for [use case]" were getting curated product comparisons inline. The client was not consistently being named. When they were named, it was usually as the third or fourth option in a list. The category leader was getting named first in close to every relevant response, and a newer competitor was getting named in the "best for beginners" and "best value" slots that the client should have owned given their actual market position. The team had tried a generic comparison content push and seen no movement.

What we shipped

We approached this as a structured cluster build rather than a content campaign. The work centered on a 40-page comparison cluster covering the buying decisions a customer actually makes in the category. The pages broke down by use case, by price tier, by feature priority, and by lifestyle context, and each one named real competing products with honest comparison.

Each comparison page followed the same structure: a direct one-sentence recommendation, a structured comparison table with five to seven competing products including the client's, three to five paragraphs of substantive comparison covering specific tradeoffs, and a clearly labeled FAQ block. We used Product, Review, and FAQPage schema across the cluster, with ItemList markup on the comparison tables. Critically, we included real prices, real feature data, and honest weaknesses for the client's own products. The pages that admitted weaknesses in some dimensions got cited more often than the pages that did not.

We shipped llms-full.txt at 1.3MB containing the comparison library, the full product specifications, the materials and sourcing documentation, and the return and warranty terms. The materials documentation in particular drove citations on durability-related prompts that we had not initially targeted.

We did not bother with off-domain press for this engagement. The category is dominated by review sites for off-domain citation, and we made a deliberate choice to focus on becoming a citation-worthy source ourselves rather than trying to get cited by Wirecutter. We did do structured submissions to two product database aggregators that ChatGPT shopping responses lean on.

The numbers

The conversion rate on the comparison pages themselves was the headline funnel change. The pages were doing real qualification work, and customers arriving from a ChatGPT citation had already absorbed the comparison logic before they hit the page. They came in further along the decision, and they were converting at almost twice the rate of pre-engagement organic on those URLs. Average order value lifted modestly, mostly because the comparison content was bundling related products into the recommended use cases and customers were buying the bundles. The return rate on organic-acquired customers dropped about a point and a half, which the operations team attributed to better expectation-setting on product fit. Branded search lifted meaningfully, and the brand started appearing organically in tertiary categories where the client did not actively market, because ChatGPT had absorbed the brand into its broader category understanding.

What we'd do differently

We undershot on competitor honesty in the first ten pages. The marketing team was understandably nervous about naming competitors and citing their strengths, and the early pages hedged in ways that made them less useful as citation sources. Once we shipped the next batch with more direct comparison language, citation rate on the new batch jumped within two weeks while the hedged pages lagged. We eventually rewrote the first ten pages, which cost a week of work we could have skipped if we had pushed harder on tone in the kickoff. We also should have built the materials documentation page in week one. It became a surprise citation driver and it was a small piece of work that did outsized impact.

What's next

The engagement extended into a retainer focused on cluster maintenance and category expansion. The client is launching a second product category in Q3, and we are building the comparison cluster for the new category in advance of launch so it lands with citation coverage on day one. We are also building a more sophisticated citation-tracking system that maps prompt variations to revenue, so the marketing team can see which specific comparison prompts are doing the conversion work and which are just noise. The longer-term direction is to push the cluster model into international markets, where the brand is launching in two regions next year and the comparison logic needs to be rebuilt for local competitive sets. The team is also testing whether the comparison framework can drive Pinterest and TikTok Search citation, both of which are starting to look more like LLM-style answer surfaces than traditional social feeds.