Make AI mention you, correctly
GEO (generative engine optimisation) means getting a brand accurately mentioned and recommended inside AI answers. Three tiers: free audit, Geoclaw🦞 from CNY 999, project work.
What GEO is
Generative engine optimisation is the practice of getting a brand accurately mentioned and recommended inside AI-generated answers. People no longer open results one by one; they read the paragraph the model produces. The contested position has moved from "which rank" to "is your name in that paragraph, and is what it says about you true".
Where it differs from SEO
| SEO | GEO | |
|---|---|---|
| What is optimised | The order of results on a search page | What the model says when it answers directly |
| User action | Clicks a link, lands on a site | Reads the answer, often clicks nothing |
| Success metric | Rank, clicks, traffic | Mention rate, first-recommendation rate, description accuracy |
| Unit of work | A single page | How consistently the brand is described across the whole web |
| Feedback loop | Days to weeks, rank is checkable | Re-test the same question set on a cycle |
In one line: SEO optimises which link a person clicks, GEO optimises how a model describes you. They are not mutually exclusive, but the work differs — porting keyword stuffing across achieves nothing, because what the model reuses is checkable fact, not term frequency.
How we work
- Measure the current state. Fix a set of questions real buyers ask, run them across several AI platforms, and archive the raw answers and screenshots. This produces the mention rate, first-recommendation rate and description accuracy baseline. No simulations, no synthetic data.
- Find the gap. Compare where competitors sit on the same questions and identify which layer the problem is in: the brand is never mentioned, it is mentioned incorrectly, or it is mentioned behind competitors. The three have completely different fixes.
- Fix the source facts. What a model reuses is extractable fact: who you are, what you do, for whom, with what verifiable proof. The work is making those facts consistent and extractable across the web, not stuffing pages with terms.
- Re-test. Same questions, same platforms, on a cycle, compared against baseline. GEO without a re-test is superstition.
We do not promise rankings. Any claim of a "guaranteed top-three position in AI recommendations" has no technical basis in a generative system.
Project-recap figures
Kechuang No.1
Shanghai Institute of Industrial Transformation
Yichi Garden
Evidence note: 100% / 80% is a 15 June 2026 Yuanbao snapshot for Kechuang No.1 (15/15 mentioned; 12/15 first-recommended). 90% / 70% is a 14 June 2026 Yuanbao snapshot for the Shanghai Institute (9/10; 7/10). The fixed question sets, raw JSON answers and screenshots are archived internally. These are dated single-platform snapshots, not a current state or a promise; later two-platform snapshots were 80% / 60% and 80% / 70%, respectively. They are not an industry benchmark or a commitment for a new engagement.
From "knowing the current state" to "having someone run it"
Each tier answers a different question. Read the boundary line before you choose.
Free GEO audit
CNY 0Answers: what your brand looks like inside AI today
Does not answer: how to fix it
- A real question set, run live on multiple AI platforms
- Mention rate, first-recommendation rate and description accuracy baseline
- Where competitors sit on the same questions
- A problem list grouped by root cause
Geoclaw🦞
From CNY 999Answers: how to do the work yourself, step by step
Does not answer: execution on your behalf
- The full workflow: diagnosis, positioning, content production, monitoring
- Standalone tools including Chinese de-AI-ification and landing page cloning
- Tests executed on real browsers through Playwright automation
- Installs into Claude Code or the Claude desktop app
Advisory, bootcamp, in-house
Project-basedAnswers: what to do, who does it, and when you re-test — within your constraints
Does not answer: guaranteed rankings
- GEO advisory: define the questions, the basis and the actions
- AI marketing bootcamp: nine cohorts delivered, teams learn by running it
- In-house training: hand the method to your own team
- Re-test on a cycle and compare against baseline
Geoclaw🦞 pricing is the ClawdMart listing price and the product page governs. Advisory and bootcamp engagements are quoted per project.
Questions we get asked
How is GEO different from SEO?
SEO optimises the ordering of a results page and is judged on rank and clicks. GEO optimises the content a model produces when it answers directly, and is judged on mention rate, first-recommendation rate and description accuracy. A user who reads the answer may never click anything, so the two require different work — keyword density carries almost no weight in GEO, because models quote verifiable facts.
Which platforms does this cover?
For the Chinese market: Doubao, Tencent Yuanbao, Kimi, DeepSeek and Baidu. For exporters: ChatGPT, Perplexity and Google’s AI answers. Each retrieves and cites differently, so each is measured separately rather than averaged into one score.
How far does the free audit go?
It produces a current-state report: how your brand is mentioned on each platform, where competitors sit on the same questions, and a problem list grouped by root cause. It answers "what does this look like now", not "how do we fix it". Fix plans, content work and re-test schedules are outside the free scope.
How quickly does GEO show results?
It depends which layer the problem sits in. "The model has never heard of you" needs extractable source material built first and takes longest. "It mentions you but gets it wrong" usually traces to inconsistent descriptions across the web and moves faster once corrected. We do not give a blanket timeline; we re-test the same question set on a cycle and compare against baseline.
Can you guarantee a top position in AI recommendations?
No, and we do not offer that. A generative system does not expose a controllable ranking table, so any guarantee of a top-three position has no technical basis. What we do guarantee: measurements use real platform answers and raw screenshots, never simulations, and every round uses the same basis so the data can be audited.
Which tier should we take?
If you do not know where the problem is, start with the free audit. If you have an internal team willing to do the work, buy Geoclaw🦞 (from CNY 999) for the full workflow. If you need priorities set within real business constraints and someone running the cycle with you, take the advisory engagement or the bootcamp.
This page in Chinese: GEO 生成引擎优化