Pilot case study · Moissanite watches
How Kloira took monthly AI citations from 2 to 37 in sixteen weeks.
No new product. No repriced SKU. No ad spend shift. Kloira ran the Surfient AI-indexing audit, shipped the six interventions it flagged, and by week sixteen ChatGPT and Perplexity were citing them in the exact comparison queries they'd been losing for two years. This is the receipts page — week-by-week data, engine breakdown, and the parts that didn't work.
0.0x
Citation growth
2 → 37 / month
0 wks
Time to result
Oct '25 → Feb '26
+0 pts
GEO Score lift
42 → 89 / 100
0
Engines now citing
of 9 tracked
Background
A moissanite watch brand betting on answer engines
Kloira sells moissanite-set timepieces on www.kloira.com. The category is unusual — moissanite is a diamond-alternative mineral most shoppers haven't heard of, every product has a dozen technical specs (stone carat, dial size, movement, water resistance, warranty), and the comparison queries are dense: "moissanite watch vs diamond watch", "best moissanite watch under $500", "is moissanite real".
These are exactly the queries that answer engines — ChatGPT, Perplexity, Google AI Overviews — answer by citing a handful of sources. Before Surfient, Kloira was almost never in that handful. They ranked reasonably on classic Google for the short-tail terms but were invisible in the generative surface.
2 / month
Across 9 engines · week 1 of pilot
16 weeks
Oct 20, 2025 → Feb 9, 2026
84 SKUs
Single collection · Shopify Plus
80+
Run weekly · diff-based alerts
Week 1 · starting point
GEO Score: 42/100. The catalog was invisible to AI.
Surfient's first run found 311 individual issues across 84 product URLs. None of them were novel — they were the same pattern we'd seen across the 500+ Shopify merchants Onviqa shipped engineering work for over the previous decade. The twist was how many of them existed on one store at once.
- No llms.txt — LLMs had no canonical map of the catalog.
- No products.ndjson feed — crawlers were parsing HTML banners.
- Incomplete Product schema on every SKU (missing GTIN, material, priceValidUntil).
- No FAQ schema anywhere — the questions buyers ask ChatGPT were unanswered on-site.
- Brand entity unresolved — Kloira wasn't in Wikidata, no sameAs graph, no Organization schema.
- Product copy written for SERP clicks — keyword-stuffed, hero-first, unquotable.
Audit summary · Oct 20 2025
42/100
GEO Score
Surfient's AI-indexing readiness index. Anything below 60 is effectively uncitable; 85+ is where compounding starts.
Weeks 1–16 · what we shipped
Six interventions, in this order, compounded into the result.
The sequence matters. You can't rewrite copy for AI quoting before the schema exists for answer engines to pick up the specs — and you can't monitor citation drift until the catalog is citable in the first place. Every Surfient install runs this same playbook; Kloira is the receipts.
- 01Week 1
Audit the whole catalog against 80+ AI-indexing checkpoints
Kloira started at 42/100 on the Surfient GEO rubric. Product pages had hero copy and images but no machine-readable spec tables, no FAQ schema, no brand entity, and no canonical llms.txt. The audit surfaced 311 individual fixes across 84 product URLs.
42 → 89 GEO score
- 02Weeks 2-3
Publish AI-ready files: llms.txt, products.ndjson, ai-sitemap.xml
The first files Surfient publishes on every install. llms.txt hands LLMs a plain-English catalog map, products.ndjson gives crawlers the structured feed they prefer to HTML, and ai-sitemap.xml tells answer engines which URLs are citable vs. transactional.
84 products, 1 feed
- 03Weeks 3-5
Schema rewrite: Product + Spec + FAQ + Review graph
Every watch page got a full Product graph with GTIN, material, movement, dial diameter, water resistance, warranty, and priceValidUntil. FAQ schema lifted from the most-asked questions in customer-support transcripts. Review schema backfilled from the 4,200+ real reviews already in Shopify.
84/84 pages schema-valid
- 04Weeks 5-7
Brand-voice copy rewrite for AI quoting, not SERP clicks
The biggest shift. Copy that ranked on Google (keyword-stuffed, above-fold hero) doesn't get quoted by ChatGPT; copy that gets quoted reads like a short, confident answer to a buyer question. We rewrote every product lead paragraph to the answer-engine template: one claim, one spec, one social proof sentence.
84 pages rewritten
- 05Weeks 6-10
Entity + site graph — who is Kloira, what do they make
Every engine needs to know that Kloira is a moissanite watch brand before it'll cite a Kloira product page. We shipped an Organization graph, a Brand graph, a sameAs list linking Instagram / Trustpilot / YouTube, and an About page built to be the canonical answer to 'who is Kloira.'
Entity resolved in 5 engines
- 06Weeks 10-16
Weekly monitor loop — diff the citation panel, fix what regressed
The compounding part. Surfient runs the 80+ checkpoint audit every week and diffs the citation counts engine-by-engine. When Gemini stopped citing a spec table in week 13, the monitor flagged it; we found the schema break, shipped the fix, and Gemini came back in week 14. This is what the $39/mo Premium plan buys.
Week-on-week delta diffing
Result · week-by-week
The curve that convinced us we had a product
Weeks 1–4 were the audit + plumbing — feeds, schema, llms.txt. The bend only starts in week 5 when the schema graph was populated enough for ChatGPT and Perplexity to trust the spec data. From week 8 it compounds — every new engine that cites Kloira feeds the share-of-voice loop for the next engine.
Per-engine breakdown
Not every engine moves at the same speed.
ChatGPT and Perplexity picked up the changes fastest — both prioritise machine- readable catalogs and honour llms.txt within days. Claude and Google AI Overviews followed once the FAQ + spec schema was in place. Gemini took longer; DeepSeek and You.com are still early surfaces for DTC jewelry.
We publish the full methodology in our weekly monitor logs — every citation is timestamped and engine-attributed. No aggregate gaming.
Fastest mover — picked up llms.txt within 9 days
Best citation quality — full product cards with price
Picked up the FAQ + spec schema rewrite
Google AI Overviews began quoting the spec tables
First citation landed in week 11
Week-1 baseline vs. week-16 pilot data. Citations are engine-attributed mentions tracked by the Surfient monitor across the nine engines on the left.
Business impact
Citations are the input. Sessions and orders are the output.
Surfient tracks "AI-attributed" sessions via a combination of referrer data (ChatGPT, Perplexity, Bing Copilot, Google AI Overviews and llms-enabled answer surfaces all leave identifiable referrers or UTM chains) and a Shopify pixel that tags orders originating from those sessions. Weeks 1–16 of the Kloira pilot:
AI-attributed sessions
up from 58 / mo
AI-attributed orders
up from 1 / mo
AOV (AI-attributed)
up from $348
Conversion rate
up from 1.72%
Numbers lifted from the Kloira Shopify admin + Surfient monitor, rolled up week 16 vs. week 1. AOV includes bundle orders. Conversion rate measured on AI-attributed sessions only; site-wide conversion was comparatively flat (non-AI traffic was the same, which is the point — no cannibalisation).
Operator quote
We didn't launch a new product, rewrite the homepage, or change the price of a single watch. We shipped the six things Surfient audited for — feeds, schema, entity, copy, llms.txt, weekly diff monitor — and by week sixteen ChatGPT and Perplexity were citing us in the exact comparison queries we'd been losing to competitors for two years.
What didn't work / what's still open
The honest counter-column
- Grok + You.com didn't move. We shipped the same audit to both, and neither cited Kloira in any of the 100+ comparison queries we tracked. Both surfaces index DTC catalogs less densely than the leaders — we're not sure yet whether this is a feed issue or a trust-graph issue, and we're running a follow-up experiment in Q2 2026.
- Gemini regressed once in week 13. A schema change on the FAQ module broke the spec-table quote path for six days. The monitor caught it, we shipped the fix, Gemini came back in week 14. This is the kind of regression only weekly diff-monitoring catches — and the reason the Premium plan pays back.
- No change in classic-SERP rank. Kloira's Google position for the short-tail terms was stable throughout the pilot. This work is additive to classic SEO, not a replacement. Stores running both pipelines see the compounding on both surfaces.
- Category matters. Kloira is a tight, spec-heavy category (watches) with a limited SKU count (84). A 10,000-SKU fashion retailer would see the same curve shape but slower — the audit surface area scales with catalog size.
Run the same audit on your own Shopify store.
The 80+ checkpoint rubric Kloira ran is the same one every Surfient install gets in the first five minutes. Free tier keeps the audit current. Premium at $39/mo ships the fixes, the feeds, and the weekly diff monitor that caught the Gemini regression in week 13.