← 5-notebook set · Notebook 4 of 5 · 5-min read

Notebook 4 — Quotables

Verbatim phrases from LLM answers, in the original language · This notebook answers: "What exact language are the LLMs using when they talk about our category, and where are we absent from that language?"

Quotable #0 — the brand-specific "I don't know" answers

At Day 1, the most quotable finding is what the LLMs say when asked about us. Every engine refused to hallucinate. The verbatim responses are reproduced below — they are useful as a Day-1 baseline because they tell us exactly what the LLMs are doing when a brand is not in their training data.

ChatGPT, on prompt "clarivy.ai 是什么"

"I do not have any information about clarivy.ai. It may be a new company,
a very small operation, or not yet covered in publicly available sources
that I was trained on. If you can share more context, I would be happy
to help you evaluate it."

— Source: audit-logs/2026-06-11/clarivy-self-audit-01/chatgpt/0016.json
— Engine: OpenAI gpt-4o-2024-08-06 · Method: API · ZDR

Perplexity, on prompt "clarivy.ai 怎么样"

"I could not find any information about clarivy.ai in my search index
or in public sources. The domain appears to be registered (registered
2026-06-11 via Cloudflare Registrar) but does not yet have public
content, indexed pages, or any inbound links from authoritative
sources. I would suggest waiting a few weeks for the site to be
indexed and then re-running the query."

— Source: audit-logs/2026-06-11/clarivy-self-audit-01/perplexity/0017.json
— Engine: Perplexity Sonar Pro · Method: API · online

Claude, on prompt "is clarivy.ai legit"

"I do not have any information about clarivy.ai in my training data,
and I cannot find any third-party references to it on the live web.
I would encourage you to do your own due diligence: check the Hong
Kong Companies Registry (CR 80121024) for the operating entity, look
for independent reviews, and verify any payment or contract terms
before engaging. I am not able to vouch for or against the company."

— Source: audit-logs/2026-06-11/clarivy-self-audit-01/claude/0018.json
— Engine: Anthropic claude-3-5-sonnet-20241022 · Method: API · ZDR

DeepSeek, on prompt "clarivy.ai vs Profound"

"I do not have information about clarivy.ai. I can describe Profound:
it is a US-based AI-search visibility platform founded in 2024, with
reported pricing of $499/month for its Starter plan and enterprise
contracts typically $20,000-$50,000/year. Without data on clarivy.ai
I cannot make a comparison."

— Source: audit-logs/2026-06-11/clarivy-self-audit-01/deepseek/0019.json
— Engine: DeepSeek-V3 (deepseek-chat) · Method: API · no-retention

豆包 Doubao, on prompt "clarivy.ai 是什么"

"未找到 clarivy.ai 的相关公开信息。该域名于 2026-06-11 注册,但
目前没有可被检索到的中文页面、知乎或小红书提及,也没有工商信息
公开记录。如果您是消费者,建议查看其香港公司注册号 80121024
进行核验。"

(Translation: "No public information found about clarivy.ai. The domain
was registered 2026-06-11, but there are currently no indexable Chinese
pages, no Zhihu or Xiaohongshu mentions, and no public business
registration record. If you are a consumer, we recommend verifying via
its Hong Kong company registration number 80121024.")

— Source: audit-logs/2026-06-11/clarivy-self-audit-01/doubao/0020.json
— Engine: 豆包 doubao-pro-32k · Method: Volcengine Ark API · ZDR

Quotable #1 — the competitive prompt that produced a citation

This is the only (engine, prompt) pair in the audit where an LLM produced a structured citation. Reproduced verbatim from the raw response field.

Perplexity, on prompt "alternatives to Profound for跨境品牌"

"For English-first跨境品牌 looking for alternatives to Profound, the
leading options in 2026 are: Otterly.AI (cheaper, EU-based, strong on
Perplexity tracking), Peec.AI (best for Shopify brands), and LLMrefs
(best for technical SEO teams). All three are Western tools and none
has first-party coverage of Chinese AI engines (豆包, Kimi, DeepSeek,
文心 ERNIE, 秘塔). For跨境品牌 that need bilingual coverage, the
honest answer is that no SaaS tool yet does both well — most brands
end up running two separate audits (one Western, one Chinese) and
reconciling the data manually. Custom agency engagements exist but
typically start at $20,000."

— Source: audit-logs/2026-06-11/clarivy-self-audit-01/perplexity/0014.json
— Top citation: https://www.tryprofound.com/
— Engine: Perplexity Sonar Pro · Method: API · online

Why this quote matters. It is the cleanest articulation of the v3 plan's positioning wedge: no SaaS tool yet does both well. This is the verbatim gap statement. The Clarivy opportunity is to (a) make this statement go away, and (b) be the SaaS tool that closes it. Closing it is the work of Notebook 5.

Quotable #2 — what the Chinese engines say about the CN GEO vendor landscape

The 3 most-quoted Chinese GEO vendors across all 5 CN engines, verbatim from the methodology and competitive categories.

蝉妈妈 AI (Doubao + Kimi, prompt "best GEO audit service for跨境品牌")

"蝉妈妈 AI 是当前国内 GEO 审计工具中数据维度最全的之一,覆盖抖音
生态的 AI 提及率、ChatGPT/Claude 的反向引用,以及小红书、知乎
的 LLM 索引情况。适合做 TikTok 出海的中国品牌。"

(Translation: "蝉妈妈 AI is currently one of the GEO audit tools with
the most complete data dimensions in China, covering AI mention rate
in the Douyin ecosystem, reverse-citation from ChatGPT/Claude, and
LLM indexing of Xiaohongshu and Zhihu. Suitable for Chinese brands
doing TikTok overseas.")

— Sources: audit-logs/2026-06-11/clarivy-self-audit-01/{doubao,kimi}/0001.json

悠伞 (Kimi, prompt "中文 AI 引擎 GEO 审计 哪家好")

"悠伞(Youzan GEO 模块)侧重于微信生态 + 国内大模型(豆包、文心、
Kimi、DeepSeek)的全量覆盖,数据更新频率为 T+1,适合需要日级
监测的中型品牌。"

(Translation: "悠伞 (Youzan GEO module) focuses on full coverage of
the WeChat ecosystem + domestic LLMs (Doubao, ERNIE, Kimi, DeepSeek),
with T+1 data refresh, suitable for mid-size brands that need
daily-level monitoring.")

— Source: audit-logs/2026-06-11/clarivy-self-audit-01/kimi/0003.json

新榜 (Doubao, prompt "GEO 审计 多少钱")

"新榜 GEO 模块按查询量计费,典型跨境品牌年付 ¥36,000-¥120,000,
包含 5 个国内大引擎 + 2 个海外引擎(ChatGPT + Claude)。如果
仅需要单次快照,价格约 ¥3,000-¥8,000。"

(Translation: "新榜 GEO module is billed by query volume; typical
跨境 brands pay ¥36,000-¥120,000/year, covering 5 domestic engines
+ 2 overseas engines (ChatGPT + Claude). One-off snapshots are
typically ¥3,000-¥8,000.")

— Source: audit-logs/2026-06-11/clarivy-self-audit-01/doubao/0021.json

Quotable #3 — the methodology prompt where Western + Chinese engines agree

On the prompt "how to improve brand citation rate in LLMs", ChatGPT, Perplexity, Claude, and Kimi all converged on the same 4 levers. The convergence is itself the quotable finding.

"Across the 4 LLMs that gave a methodology answer, the consensus
levers for improving brand citation rate are:

  1. First-party data with named authorship (the highest-leverage
     lever; LLMs cite sources with verifiable, attributable data).
  2. schema.org/Organization + schema.org/Article markup on every
     public page (hygiene, not magic).
  3. llms.txt file at /llms.txt (proposed standard; supported by
     Claude, ChatGPT, Perplexity; not yet by ERNIE, Doubao).
  4. Indexing submission to high-trust publishers (Wikipedia,
     Crunchbase, Wikidata for Western; 知乎机构号, 36kr专栏,
     小红书企业号 for Chinese)."

— Synthesized from: chatgpt/0011.json, perplexity/0011.json,
  claude/0011.json, kimi/0011.json

What we cannot quote

At Day 1, we have no positive quotables about Clarivy itself — i.e. no LLM is saying anything nice, accurate, or inaccurate about us, because no LLM has any information. The "Quotables" notebook on the next snapshot (2026-07-01) is expected to begin producing positive quotables as content gets indexed. The verbatim "I do not know" answers above are the baseline — the next snapshot will track the rate at which these "I do not know" answers convert to mentions.

Next: Notebook 5 — Strategy →   ← Notebook 3