We have no customer case study yet, and we will not write one from a screenshot. What we have is a European furniture store we operate, measured with the same tool, the same repeats and the same test. It is labelled as ours everywhere it appears.
Numbers are from the published datasets and one product run; the store is not named because it is a separate company.
Share of AI answers citing the store on category questions, two competitors measured alongside.
The bands do not overlap, so the gap can be stated as a fact rather than a hope.
Category questions versus the store's own Search Console queries. The question set moved the number by nineteen points; nothing on the site changed.
The same brand in two languages. The intervals overlap, so the report declines to call it a gap.
On generic "best mid-century armchair" questions the store was cited often. On the queries real visitors typed into Google — product names, materials, delivery questions — far less. The first number flatters; the second is the business.
A 100-prompt run killed our own first hypothesis: the difference between languages was small-sample noise. The difference between platforms was not — Google's AI surfaces cited the store far less than Perplexity did.
Of the store's pages that appeared as sources, almost all were product and category pages. The content plan that followed was not "write more articles".
AI assistants were the third-largest referral source to the group's stores, and in one country store more than one visit in six came from them — before any optimisation.
Search Console queries, brand terms removed, five assistants, three repeats, two languages. Intervals printed; minimum detectable change printed with them.
The analysis script for the before-and-after was published with its checksum on 11 September 2026, before the changes went live. It cannot be edited to fit the result.
Direct answers at the top of product and category pages; comparison tables; the delivery and material facts the cited pages had and ours did not.
Same questions, same repeats. The result will be published as "changed", "within noise" or "new baseline" — whichever the test returns — in the fourth study.
The intervention result is not yet published, and this page will not report it early. What is published: the provenance study, the audit and the measurement study, with datasets.
A category score is a different instrument. Connect Search Console and measure what people actually type.
Per-platform rates, per-platform fixes. One number across five assistants hides the one that matters for your market.
That is where the citations point. Articles come later, from the questions the product pages cannot answer.
Otherwise the before-and-after will always find what it was paid to find.