Challenge
When consumers ask ChatGPT which Quooker to buy and where, ChatGPT answers with a shortlist of products, each linked to one or more shops. At the baseline on February 9, 2026, Spoelbakhuis appeared in 9 of 20 tracked prompts, but usually low in the list: on average at product #2.9, and around #5 for the PRO3 Flex. For the energy distributor, ChatGPT never linked to Spoelbakhuis at all.
Solution
Spoelbakhuis's product pages were given a complete JSON-LD structured data profile, so AI models can read the products, variants, prices, availability and the shop itself without guesswork. Twenty realistic buyer prompts across four product categories served as the measuring point: per prompt, which products ChatGPT showed, which shops it linked and in what order.
Spoelbakhuis moved from the back of ChatGPT's product lists to the front, just as ChatGPT started showing fewer products than ever.
Optimizing without measurement is guessing. Spoelbakhuis's case shows why position matters more than volume once AI answers get shorter.
Present, but at the back of the list
Quooker buyers increasingly start with ChatGPT: which Quooker suits a small kitchen, PRO3 or COMBI+, where to buy a loose Flex tap. ChatGPT answers with a few products, each with links to shops. For a webshop that sells the same brand as dozens of competitors, the real question is which shop gets linked, and how high.
At the baseline, 15.6% of all products ChatGPT showed carried a Spoelbakhuis link. When Spoelbakhuis did appear, it sat on average at product #2.9 and was usually the second link, not the first. For the energy distributor, Spoelbakhuis was absent from all 15 products shown.
Structured data, measured prompt by prompt
The work focused on one lever: making Spoelbakhuis's product information machine-readable. Each product page got structured data describing the product, its offer and the shop behind it, so an AI model no longer has to infer them from the page layout.
The measurement covered 20 buyer prompts across four categories: the Quooker PRO3 Flex, the Flex COMBI+, the energy distributor and the Flex loose tap. For every answer, three things were recorded: whether Spoelbakhuis was linked, at which product position, and at which link position within that product.
Results after six weeks
Between the two measurements, ChatGPT moved from GPT-5.2 to GPT-5.3. The new model shows around 60% fewer products per answer: 52 instead of 128 across the same 20 prompts. Fewer slots means every remaining slot counts more.
Spoelbakhuis held its presence at 9 of 20 prompts and grew its share of shown products from 15.6% to 17.3%. The bigger shift is in position. The average product position improved from #2.9 to #1.8. For the PRO3 Flex it went from #5.2 to #1: wherever Spoelbakhuis appears for that product, it is the first product shown and the first link. For the Flex loose tap, the average position improved from #5.3 to #2.
The link order improved too. Within a product, the Spoelbakhuis link moved from position 1.8 to 1.6 on average, and for the COMBI+ from 2.7 (almost always the third link) to 1.7. And for the energy distributor, where Spoelbakhuis was invisible at baseline, it now appears in 2 of 5 prompts, both times as the first product.
Not every number went up. For the Flex loose tap, Spoelbakhuis's share of shown products fell from 22.5% to 13.3%, even as its position improved. That is exactly what continuous measurement is for: it shows where the next round of optimization should go.