CClarivyBilingual AI Search Visibility

2026-06-11 · 8 min read · Chaoli (founder, Clarivy) · All posts

Day 1 of building Clarivy: here is our own GEO audit

This morning, around 09:00 UTC+8, I registered clarivy.ai on Cloudflare and pointed the nameservers. The domain propagated within 30 minutes. By 09:08 I had run the same 9-engine GEO audit that we will sell to customers.

Result: 0 of 270 (engine, prompt) combinations mention clarivy.ai. Zero. Not "low." Not "needs work." Just zero. This is the Day-1 dogfooding baseline, and we are publishing it in full because the alternative is asking you to trust a methodology that we have not used on ourselves.

What we ran

9 AI engines:

30 prompts across 5 categories:

  1. Buying intent (7) — "best GEO audit service for [vertical] in [year]"
  2. Competitive (7) — "Vendor A vs Vendor B vs Vendor C"
  3. Methodology (6) — "what is GEO," "how to improve brand citation rate"
  4. Brand-specific (5) — "clarivy.ai 是什么," "is clarivy.ai legit"
  5. Purchase intent (5) — "GEO audit pricing 2026"

9 × 30 = 270 datapoints, each with the 4-column methodology matrix (engine, method, sampled_at, reproduce link). Raw JSON is at /audit/self-audit-01.html and on the audit-logs GitHub repo.

What the engines said (the most useful cell)

Perplexity cited tryprofound.com on at least one competitive prompt. ChatGPT listed Profound, Otterly, Peec, and LLMrefs in its top-of-mind vendor list. Claude wrote a generic explainer about GEO methodology. None mentioned clarivy.ai — because none could have.

This is the part of the audit that is actually useful to you, dear future customer: the prompt set works. The fact that established competitors surface means the queries are not too narrow, not too easy, and not biased toward my brand. If the engines had returned no vendor names at all, I would have worried the methodology was broken.

What 0/270 actually means

It does not mean "Clarivy has bad content." It means we have no content at all, and the LLM knowledge graphs have not associated "clarivy.ai" with any entity. There is no entity_id, no inbound links from trusted sources, no Wikipedia entry, no Crunchbase page, no Crunchbase-style entity record. Of course we are 0/270 — we are not even 1/270 yet.

Anyone claiming "we got mentioned by AI search in 30 days" on a brand-new domain is either (a) running on a typo'd query that happens to contain a substring of their brand, (b) measuring differently from the way we measure, or (c) lying. There is no Day 1 escape from this.

The 12-week plan (this is the contract with you)

We will not bullshit you with "30 days to top of LLM answers." Here is the actual plan, in priority order, with the expected citation-rate impact called out at each step:

  1. Week 1–2 — Hygiene. Add schema.org Organization + Product + Offer markup to every page. Add a llms.txt at /llms.txt describing what clarivy.ai is. Submit the sitemap to Baidu Zhanzhang (百度站长平台), ByteDance Juliang (字节巨量引擎), Shenma (神马搜索), Sogou (搜狗站长平台), and Bing Webmaster Tools. Expected impact: 0–1 mention. This is not optional, but it is also not sufficient.
  2. Week 3–4 — First-party data. Publish 30 unique datapoints, one per content page. Examples: "9 AI engines surveyed in our methodology," "$99 Snapshot deliverable includes 5 specific fixes," "30 prompts × 9 engines = 270 datapoints per audit." Each is a small, attributable fact that LLMs can cite. Expected impact: 2–4 mentions on methodology queries.
  3. Week 5–6 — Named customers. After the first 3 deliveries, publish 3 case studies with named customers, their written consent, and verbatim quotes. (We have not collected any quotes yet; we will not write fake ones.) Expected impact: 4–8 mentions on competitive queries.
  4. Week 7–8 — Bilingual press. IndieHackers post, Show HN submission, an 即刻 thread, a 知乎回答. Cross-linked so LLMs see clarivy.ai referenced from multiple trusted sources. Expected impact: 6–12 mentions across buying-intent queries.
  5. Week 9–10 — ProductHunt launch. One-day launch, all-hands-on-deck for the comment thread. Expected impact: 10–18 mentions, especially in English buying-intent queries.
  6. Week 11–12 — Re-audit & publish deltas. Re-run the same 30-prompt × 9-engine matrix. Publish a side-by-side diff in a blog post titled "Day-90: 0 → 18 mentions, here's what worked and what didn't." Expected impact: this is the credibility move that pays back over years.

What we will not do

What this means for you, the buyer

If you are a跨境DTC brand evaluating GEO services, here is what you should ask of any vendor — including us:

  1. Show me your own brand's most recent GEO audit. (We just did. 0/270.)
  2. Show me the raw JSON of an audit you ran on a paying customer. (We will, on first delivery, with the customer's written consent.)
  3. Tell me which 9 engines you commit to, in writing, with versioning. (We do — see /audit/methodology.html.)
  4. Tell me the methodology matrix you attach to every datapoint. (We do — engine, method, sampled_at, reproduce link.)
  5. Tell me your refund policy. (30 days, no questions.)

If a vendor can't answer all five, walk away. The GEO market has more heat than light right now (June 2026), and the 12-week window for serious players is open, but it is not open forever.

How to follow along

If you want to be one of the first 3 customers whose audit we publish with their consent, book a 30-min briefing: calendly.com/clarivy/30min. We respond in English or 中文 within 1 business day.

See the three SKUs →   Read the full Day-1 audit