A mention does not mean your website was cited
An AI may remember your brand while directing trust and traffic to media sites, aggregators or even a competitor's website.
Four-platform GEO visibility audit
Does AI mention—or cite—
your brand?
Observe Doubao, DeepSeek, ChatGPT and Claude using the same real customer questions. Keep evidence for every conclusion, find the one page most worth improving, then retest with the same questions.
No promise of inclusion, ranking or a fixed uplift. Conclusions are supported by evidence, and unknowns are labelled clearly.
BASELINE / OBSERVATION PLAN
PLANNEDOne question set, four evidence tracks
01DoubaoAnswer · sources · screenshot
02DeepSeekAnswer · sources · screenshot
03ChatGPTAnswer · sources · screenshot
04ClaudeAnswer · sources · screenshot
64planned observations
8core questions
2×baseline / retest
One question set, separate reportsDoubaoDeepSeekChatGPTClaude
WHY CITATIONGUARD
Do not judge a brand's GEO from a single answer.
We do not manufacture an “overall score of 92”. Separating platforms, modes and evidence is how we locate the real issue.
An answer is not automatically auditable
Without the full answer, source URLs, time, region and mode, one screenshot cannot support a sound business decision.
API results are not consumer-product results
App, web and API results—and web search versus model memory—must be recorded separately rather than blended into one attractive score.
A page change is not proof of improvement
Change is worth discussing only after retesting fixed questions in a comparable environment and allowing a reasonable observation window.
THE METHOD
Turn GEO from an impression into a chain of evidence.
Freeze the question set, separate the platforms, record the environment, improve one page and retest. Every step leads back to the original answer and its sources, not just a summary chart.
E0E1E2E3E4Evidence becomes more complete at each level
- 01
Define real buying questions
Confirm the official website, brand aliases, three competitors and target markets. Turn sales conversations into 20–30 customer questions and retain control questions.
- 02
Establish a four-platform baseline
Use the same question set on each platform. Save the complete answers, every visible source, screenshots, time and test environment.
- 03
Inspect the website and its evidence
Check the selected pages for HTTPS, crawl responses, robots directives, noindex, canonical URLs, sitemaps, page copy and structured-data signals.
- 04
Improve one page, then retest
Use the evidence to select one priority page and propose a revision. Retest under comparable conditions after implementation without presenting correlation as a causal guarantee.
DELIVERABLES
Not a vague recommendation, but a handover-ready record.
Four-platform baseline
Evidence checkedBrand visibility audit
Doubao
separate basis
DeepSeek
separate basis
ChatGPT
separate basis
Claude
separate basis
This only illustrates the report structure. It is not a customer result.
- 01Customer question set and collection plan
- 02Complete answer and source evidence index
- 03Separate visibility metrics for each platform
- 04Website technical-eligibility review
- 05Revision plan for one priority page
- 06Baseline and like-for-like retest report
Human responsibilities stay visible
Authorised people handle consumer-product sign-in, complete source collection, product-fact approval and final publication. The tools plan, validate, calculate and report; they do not bypass platform restrictions.
CROSS-AGENT
One core, packaged for the coding agents you already use.
Codex, Claude Code and WorkBuddy share the same Python CLI, measurement definitions and evidence rules. The execution environment can change without changing the audit standard.
01CodexLocal Skill / Plugin
02Claude CodeLocal Skill / Plugin
03WorkBuddyLocal Skill / Plugin
BETA PRICING
Use one real delivery cycle to test the value.
These are founder-led Beta prices. Implementation is quoted separately based on your CMS or codebase.
China two-platform audit
Doubao + DeepSeek
¥499Beta / site
A first check of AI visibility in China for local acquisition and brand teams.
- Question set and collection plan
- Technical eligibility and evidence audit
- Priority-page recommendations
- Like-for-like retest report
Full four-platform audit
Doubao + DeepSeek + ChatGPT + Claude
¥999Beta / site
One question set across Chinese and international platforms, with every surface and mode still reported separately.
- Question set and collection plan
- Technical eligibility and evidence audit
- Priority-page recommendations
- Like-for-like retest report
Global two-platform audit
ChatGPT + Claude
¥499Beta / site
For international teams with business in a target region and compliant testing access.
- Question set and collection plan
- Technical eligibility and evidence audit
- Priority-page recommendations
- Like-for-like retest report
ChatGPT and Claude testing is available only to customers or testers with authorised accounts in supported regions. We do not bypass restrictions to complete a test.
A GOOD FIT
Teams likely to see value sooner
- B2B SaaS, AI tools and industrial-product teams with their own website and valuable leads
- SEO / GEO consultants and marketing agencies that need a repeatable, white-label delivery process
- Teams that can provide real customer questions, public product evidence and access to update website pages
NOT A FIT
Requests we decline
- Guarantees of a number-one rank, rapid inclusion or a fixed uplift
- Product claims with no publicly verifiable evidence
- Requests to bypass account, regional or platform restrictions
- High-risk medical, legal or financial claims in the first Beta cohort
FAQ
Clear boundaries make the work easier.
Add anything still uncertain to the enquiry form.
Can you guarantee inclusion, citations or a number-one position?
No. Platform crawling and answers are outside our control. We provide an auditable diagnosis, actionable page recommendations and a reproducible retest process.
Does the Skill collect data from all four platforms automatically?
It does not bypass sign-in, regional or platform restrictions. An authorised tester collects consumer-product evidence manually. Permitted API experiments are run as separate batches and are never presented as consumer-product results.
Why is there no single overall GEO score?
Platforms, models, accounts, regions and search modes are not directly comparable. We report mentions, official-site citations, target-page citations, factual consistency and stability separately for each platform.
What do I need to provide?
Your website, brand aliases, three competitors, 20–30 real customer questions, public product evidence, and people who can approve website facts and changes. Do not send account passwords, cookies or API keys.
Which coding agents are supported?
The current local packages support Codex, Claude Code and WorkBuddy. All three use the same audit logic, measurement definitions and evidence rules.
START WITH FIT
Before you buy, check whether your website is ready.
Send the website, industry, target market and platforms you want to observe. We will first assess whether you need two platforms, all four, or stronger website foundations.
No sensitive credentials requiredDo not submit passwords, cookies, API keys or private sensitive data.