Client name withheld. Figures are from our measurements on the stated dates.
Business: an online retailer of professional kitchen equipment for restaurants, bakeries, and food service operators. Built-to-order catalog, roughly 5,400 active products, no warehouse stock.
Starting point
On September 13, 2026, we measured the client's presence in AI answers with 15 commercial search queries run through Perplexity, covering the main equipment categories it sells. The client's site was cited in 1 of the 15 queries — a narrow product category — ranked third of seven sources, with three citations inside the answer text. The other 14 queries did not mention the client.
One competitor was cited in 9 of the 15 queries, well ahead of the field. Four others were each cited in 3 to 4 queries.
What we found
The obvious explanation — the cited page is thin, add more content to the rest — did not hold up. We compared the one cited category page against a major category that was never cited:
| Measure | Cited category | Uncited category |
|---|---|---|
| Body text | 17,946 characters | 51,036 characters |
| FAQ block | none | 5 questions |
| Linked articles (per sitemap) | 32 | 68 |
| Product listings | 132 | 296 |
On every measure the uncited category was the stronger page. Content length, an FAQ block, and cluster size did not explain the gap. The likely remaining factor, not separately verified: the uncited category has established brand-name competitors already occupying AI answers; the cited one does not.
A separate pull of three months of Search Console data (September 15) showed articles driving 42% of the site's total clicks, while product pages — 74% of indexed URLs — drove 11% of clicks. Local-language product pages had the best position (9.8) and CTR (2.2%) on the site, which narrowed the problem: those pages were reaching the index, they were not getting shown.
A GA4 pull the same month showed the AI-assistant referral channel produced zero recorded conversions over three months (July through September) while engagement on that channel rose over the same period — session duration from 16 to 34 seconds, events per session from 5.17 to 7.63. Traffic and engagement from AI referrals were growing; conversions were not.
What we did
- Published an
llms.txtfile listing the 20 top-level categories with product counts and 19 brands. - Corrected a firewall rule that was blocking Meta's link-preview crawler and confirmed the major AI crawlers (GPTBot, OAI-SearchBot, PerplexityBot) and Googlebot were reaching the site unblocked.
- Fixed product structured data: 4,078 of 5,422 products had an empty description field in their schema markup because the fallback source field was never wired in; added it, and matched spec attributes to the visible specification tables.
- Rebuilt the Google product feed as a dynamic endpoint after finding the static feed file was frozen behind a 10-year edge cache; it now serves 4,577 products per language on each request.
- Fixed category pagination pages that were marked
noindex, nofollow, which had made listing pages past page 1 invisible to crawlers. - Removed a duplicate category branch that was splitting the same products across two listings, and fixed internal search so different dash characters in size specs (e.g. "1500x600" vs "1500×600") return the same results.
- Fixed duplicate purchase-event tracking in analytics so conversion data, including from AI referrals, can be measured going forward.
Status
Re-measure scheduled; results will be published here.