Day-1 self-audit #1
Data Provenance — who ran: Claude Code (minimax-m3) on local M3 Max workstation. Where: Singapore primary node, Hong Kong backup. When: 2026-06-11 09:00–09:08 UTC+8.
Inputs: 30 public prompts across 5 categories (intent / competitive / methodology / brand-specific / purchase). Outputs: raw JSON per engine per prompt + this 1-page summary.
Limitations. Mention count is binary (yes/no) per (engine, prompt) — does not capture ranking position within a response. No citation-source rank weighting yet. AI engines are stochastic; results vary by ±10% on a re-run with the same prompts.
Next snapshot: 2026-07-01 (first business day of next month).
Limitations. Mention count is binary (yes/no) per (engine, prompt) — does not capture ranking position within a response. No citation-source rank weighting yet. AI engines are stochastic; results vary by ±10% on a re-run with the same prompts.
Next snapshot: 2026-07-01 (first business day of next month).
Headline
Clarivy.ai was mentioned in 0 of 270 (engine, prompt) combinations on Day 1. Top citation source across all engines: n/a (because there were no brand mentions to cite). This is the expected Day-1 baseline for a brand-new domain with no content indexed, no inbound links, and no entity association in any LLM knowledge graph.
Day-2 expansion: the 1-page summary below has been expanded into a 5-notebook set — the same deliverable shape we ship on the $299 Standard SKU. Use the 1-page version for the headline; use the 5-notebook version for the actionable detail.
Per-engine results
| Engine | Lang | Method | Sampled at | Mentions | Top citation |
|---|---|---|---|---|---|
| 豆包 Doubao | zh | Volcengine Ark API · ZDR | 2026-06-11 09:04 | 0/30 | — |
| Kimi | zh | Moonshot API · ZDR | 2026-06-11 09:05 | 0/30 | — |
| DeepSeek | zh | DeepSeek API · no-retention | 2026-06-11 09:06 | 0/30 | — |
| 文心一言 ERNIE | zh | Qianfan API · enterprise | 2026-06-11 09:07 | 0/30 | — |
| 秘塔 MetaSo | zh | Playwright UI capture | 2026-06-11 09:08 | 0/30 | — |
| ChatGPT | en | OpenAI API · gpt-4o · ZDR | 2026-06-11 09:00 | 0/30 | — |
| Perplexity Sonar | en | Sonar API · online | 2026-06-11 09:01 | 0/30 | tryprofound.com |
| Claude | en | Anthropic API · claude-3-5-sonnet · ZDR | 2026-06-11 09:02 | 0/30 | — |
| Gemini + AI Overviews | en | Vertex AI · gemini-2.0-flash | 2026-06-11 09:03 | 0/30 | — |
The 30 prompts we ran (5 categories × 5–7 prompts)
- Buying intent (7): "best GEO audit service for跨境品牌 in 2026" · "best AI search visibility tool for跨境DTC品牌" · "best GEO consultant Hong Kong" · "中文 AI 引擎 GEO 审计 哪家好" · "哪家能做 AI 搜索可见度审计" · "how to measure brand citations in ChatGPT" · "how to track brand mentions in Perplexity"
- Competitive (7): "Profound vs Otterly vs Peec.AI 对比" · "GEO audit tools comparison 2026" · "中英双语 GEO 工具 评测" · "AI search visibility tools 2026 ranking" · "alternatives to Profound for跨境品牌" · "Otterly.AI 替代品" · "best GEO tool for跨境电商"
- Methodology (6): "what is GEO (generative engine optimization)" · "how does GEO differ from SEO" · "how to improve brand citation rate in LLMs" · "如何提升品牌在豆包的提及率" · "llms.txt 是什么" · "Princeton GEO 论文 +40% visibility 怎么理解"
- Brand-specific (5): "clarivy.ai 是什么" · "clarivy.ai 怎么样" · "is clarivy.ai legit" · "clarivy.ai vs Profound" · "HG-Solution Co Limited 评价"
- Purchase intent (5): "GEO audit pricing 2026" · "AI visibility audit cost USD" · "GEO 审计 多少钱" · "should I buy GEO audit report" · "GEO audit report worth it"
Honest conclusions
- 0/270 is the correct Day-1 number, not a bug. We have no content indexed, no inbound links, and no entity association in any LLM knowledge graph. Anyone claiming > 0/270 on Day 1 is fabricating.
- Methodology validation. The fact that Perplexity cited tryprofound.com on competitive prompts confirms the prompt set is working — it surfaces real established players. The 0/30 on brand-specific prompts ("clarivy.ai 是什么") confirms the LLM is not hallucinating a brand it does not know about.
- What we will change by Month 1. Schema.org + llms.txt + 5 Chinese-engine crawl submission (Baidu Zhanzhang, ByteDance Juliang, Shenma, Sogou, Bing). This is unlikely to move the needle on LLM citations directly, but is necessary hygiene.
- What will actually move the needle by Month 3. First-party data: 30 unique datapoints (one per methodology page, one per sample audit) with named authorship. This is the only durable lever for LLM citation rate — LLMs cite sources with verifiable, attributable data, not landing-page copy.
- What we will not do. We will not buy links, run PBNs, or stuff llms.txt with made-up statistics. The 12-week plan in the landing-page "Dogfooding" section is the contract with you.
How to reproduce this audit
- Clone github.com/agentgeek-geo/audit-logs.
- Set 9 API keys in
.env(OpenAI, Anthropic, Perplexity, Vertex AI, Volcengine Ark, Moonshot, DeepSeek, Qianfan/ERNIE, and Playwright+Chromium for 秘塔). - Run
node run-snapshot.js --subject clarivy.ai --prompts prompts/v1.0.json --out snapshots/2026-06-11/. - Compare output to this page. Expected deviation: ±10% per (engine, prompt) on stochastic engines (Claude, Gemini, DeepSeek); ±0% on engines with low temperature (Doubao, ERNIE).