Answer first: on 2026-06-13, Yanse's blended AI visibility rate was 44.2% (53 appearances in 120 runs) across ChatGPT, Gemini, Perplexity and Claude — Gemini 100%, ChatGPT 30%, Claude 27%, Perplexity 20%. Benchmark: most companies start at 5–15%; >40% indicates strong visibility. Two stories dominate this issue: a 100%→0% collapse on Perplexity in 48 hours, and Gemini hallucinating our domain in 38% of its answers.
6 real buyer prompts (verbatim-identical to our 2026-06-12 baseline) × 5 independent runs each × 4 platforms = 120 runs via DataForSEO AI Optimization API, US/EN, web search enabled. Models: Perplexity sonar (temp 0.7), ChatGPT gpt-5.3-chat-latest, Gemini gemini-3.5-flash, Claude claude-fable-5. Metric: visibility rate = appearances/N; median rank reported as secondary only. Raw JSON retained for every run; replication cost ≈ $8.76.
| Buyer prompt (abridged) | Perplexity | ChatGPT | Gemini | Claude |
|---|---|---|---|---|
| Oil blotting paper OEM, low-MOQ private label (head term) | 0/5 | 3/5 · med #1 | 5/5 · med #1 | 1/5 · #1 |
| Oil-control film OEM, low MOQ | 0/5 ⚠ was 5/5 | 1/5 · #1 | 5/5 · med #1 | 1/5 · #1 |
| Bamboo charcoal blotting paper OEM | 4/5 · med #1 ↑ | 3/5 · #1 | 5/5 · med #3 | 1/5 · #7 |
| Whole oil-control & acne line under one roof | 0/5 | 0/5 | 5/5 · med #1 | 1/5 · #1 |
| Private-label blotting paper suppliers, China | 2/5 · #5 ↑ | 1/5 · #1 | 5/5 · med #1 | 0/5 |
| OEM suppliers for US beauty brands | 0/5 | 1/5 · #1 | 5/5 · med #2 | 4/5 · #8 |
| Platform visibility rate | 20% | 30% | 100% | 27% |
On June 11, the oil-control film prompt returned Yanse 5/5 times on Perplexity at rank #1–2. On June 13: 0/5 — the entire citation pool had rotated to competitor and directory pages. This is Profound's "40–60% of cited domains replaced within a month" happening live, in 48 hours. The same volatility worked for us elsewhere: bamboo charcoal rose 60%→80% (now citing our new domain — the 301 + sameAs entity merge propagating), and the China private-label query rose 0%→40%. The lesson we'll repeat every issue: AI visibility is an operating rhythm, not a ranking you win once.
Across 120 answers we counted every Yanse domain mention: 22× the correct yansecos.com, 23× the legacy yansebeauty.com (301 set, AI indexes not yet refreshed — all from ChatGPT and Claude), and 11× pure hallucinations: yansecosmetics.com ×9, yanse-cosmetics.com ×2, yansecosmetic.com ×1 — mostly Gemini (11 of its 29 domain mentions, 38%). Action taken: registering all three hallucinated domains and 301-ing them to the real site — converting phantom referrals into real traffic and closing a squatting risk. If AI talks about your brand, audit which address it hands out.
| Engine | Top cited sources (count) | Tactical read |
|---|---|---|
| Perplexity | alibaba.com (20) · taikicosmetics.com (20) · gspmed.com (15) · made-in-china.com (15) · accio.com (12) · youtube.com (11) | Directory sources decide life and death. Our 8 marketplace listings go live next — the direct weapon to retake the film query. |
| ChatGPT | yansebeauty.com (15) · made-in-china.com (12) · beaza.com (6) · taikicosmetics.com (6) | Reads owned sites deepest — but remembers our old domain. Issue #2 will verify the 301 switch. |
| Gemini | (citation URLs masked behind vertexaisearch redirects) | 100% visibility driven by deep owned-site crawl; the weakness is domain hallucination, countered by entity consistency. |
| Claude | accio.com (104) · alibaba.com (73) · taikicosmetics.com (21) · gspmed.com (15) | Practically lives on B2B aggregators. accio.com coverage audit added to the action list. |
yansecosmetics.com + 2 variants, 301 → yansecos.com. ~$30/yr converts Gemini's phantom referrals into real traffic.
Oil-control film listings on Alibaba International, titled in buyers' exact phrasing — aimed at Perplexity (alibaba: 20 citations) and Claude (73).
LinkedIn ranks #2 among cross-platform AI citation sources in industry indexes. Assets ready, publishing next.
Same 6 prompts, same models, same N. Verifies: 301 propagation into ChatGPT, listing effect on Perplexity/Claude, Gemini hallucination rate after domain registrations.
Microsoft Copilot is not covered (no API access in our harness) — "four of the big five." API calls approximate but do not perfectly equal consumer apps (no personalization, location or chat history). Gemini's citation URLs are masked by redirects, so its source profile is partially blind. N=5 is a cost-precision compromise: a single cell swing of one run ≈ ±20pp, so trends across issues matter more than any single number. Everything here is replicable: identical prompts, stated models, ~$8.76.