A common GEO mistake is to ask an AI assistant a few times whether it recommends a brand and treat an occasional mention as success. Generative answers vary by model version, market, language, context and phrasing. A reusable baseline fixes a question set, records test conditions and separates each answer into comparable measures. That gives content, technical and PR work a defensible order of priority.
Define the audit scope first
Choose questions from business objectives rather than mechanically converting a keyword list. A practical set covers brand awareness, service discovery, use cases, comparisons and purchase decisions. Record language, market, platform, date and login state for every test. A company serving Chinese and international buyers should create distinct Chinese and English questions because translation does not always preserve intent.
Metric 1: brand presence
Brand presence is the percentage of fixed questions in which the brand is named. Separate primary recommendations, shortlist inclusion, secondary mentions and absence. Presence alone is not positive. If the brand is placed in the wrong category or connected to irrelevant services, visibility can amplify misinformation. Always interpret presence together with accuracy.
Metric 2: answer accuracy
Create a fact-check sheet for company name, location, service scope, audience, capabilities, case descriptions and contact information. Classify each statement as accurate, partly accurate, outdated, unverifiable or wrong. Flag high-risk errors such as invented clients, incorrect pricing, wrong markets or services the company does not offer. Accuracy is the floor of GEO.
Metric 3: citation and source quality
Record whether the answer cites sources, which domains appear, whether links work and whether the cited page supports the claim. Distinguish owned websites, independent media, encyclopaedic sources, communities, directories and aggregators. Citation volume is not the goal. Independence, relevance and credibility matter more.
Metric 4: competitive share of answer
Across the same question set, record which competitors appear, where they appear and why they are recommended. Identify whether their visibility is supported by technical content, cases, media coverage, community discussion or general brand recognition. The purpose is not to copy competitors, but to find evidence the market already recognises and questions WEPR has not answered.
Metric 5: position and strength of description
A brand may be the first recommendation, one name in a list, an additional resource or part of a warning. Apply consistent labels and capture description length, relevant service terms and whether the answer gives a reason to choose the brand. Raw answers plus transparent labels are more useful than a mysterious composite score.
Metric 6: technical accessibility
Verify that public pages return successful status codes, robots rules permit access, the sitemap includes published articles, canonical links are correct, mobile content is complete and critical copy is available to crawlers. Test access for relevant search and AI search crawlers. Administration pages, drafts and customer data must remain protected and excluded from indexing.
Metric 7: citability of content
Review priority service pages and insights. Does the title address a real question? Does the opening give a direct answer? Does the body explain steps, conditions, evidence and limitations? Are facts connected to a source and date? Pages made of slogans, repeated keywords or interchangeable paragraphs may be crawlable without being useful building blocks for an answer.
Metric 8: movement from visibility to action
GEO should not end with screenshots. Separate AI referrals, organic search and direct visits in analytics, then observe whether visitors move to cases, services or contact pages. Add source tracking to enquiries and compare it with sales feedback. Early samples will be small, so avoid inflated conversion claims, but establish consistent measurement from the beginning.
Turn the audit into an action plan
A useful report contains the method, question set, raw answers, all eight measures, critical errors, major competitors, source distribution, technical issues and 30/60/90-day actions. Prioritise factual risk, crawl barriers, unanswered high-value questions, weak external evidence and user experience. Retest the same core set monthly while reserving a few exploratory questions.
Minimum viable checklist
Confirm eight things: core brand facts are correct; the brand appears for priority service questions; cited sources are credible; competitor recommendations have identifiable reasons; public content can be crawled; each article answers a question independently; visits can move toward cases and contact; and every test records platform, date, language and market. Consistent measurement turns GEO into a manageable growth discipline.
Limitation: AI platforms do not disclose their complete ranking and generation systems. Audit findings are observations tied to a specific time and test environment, not guarantees of deterministic placement.