From brand facts to qualified enquiries: a practical GEO audit

We start a GEO audit with four questions: which questions mention the brand, how the answer describes it, whether the user can find the brand, and whether a qualified enquiry follows. Record these stages separately to decide whether to improve content, evidence or the contact journey. This guide covers the workflow from fact checking to lead review.

Appearing in “What is industrial inspection?” has a different business meaning from being shortlisted for “We have two production lines and need to detect scratches on reflective surfaces; how should we select a system?” The first may support learning; the second may support purchasing. Neither guarantees a sale. Observe them separately instead of combining them into an impressive exposure figure.

The first deliverable is a question list. Before discussing how many more articles to write, ask sales to collect recurring questions from recent enquiries and check whether public pages answer them. Coverage without evidence of customer demand does not establish business value.

Bring marketing, product and sales together to review the source material. Check whether the brand name, operating entity and product names are being confused. Document service regions, eligible customers and excluded services. Distinguish similarly named companies using verifiable locations, entity relationships or product information.

Replace “industry-leading” with a capability that can be checked. When evidence is missing, mark the claim as awaiting confirmation rather than filling the gap with a larger number. Fact cards reduce conflicting statements; they do not create a supposed platform trust score. Consistency helps people verify information, but does not establish a universal AI scoring mechanism.

Extract questions from recent enquiries, quotation emails and support records, removing personal information. Preserve budgets, regions, objects and constraints instead of reducing everything to “industry + recommendation.” Ask sales to classify questions as pre-purchase, initial research or after-sales. Support questions may deserve content, but should not be mixed into acquisition metrics.

Start by selecting 20 questions manually. This is a manageable working sample, not a platform requirement. Score observed frequency, service fit and decision impact from 0 to 2; record evidence gaps and production costs separately. The scores rank work, not search demand. Prioritize frequent customer questions that your evidence can answer but existing pages cannot.

Every claimed advantage should explain the object, conditions, action and evidence. “High accuracy” alone gives no basis for judgement. A stronger page specifies the test object, defect definition, sample size, false-negative and false-positive counting methods, and applicable version. If commercial data cannot be disclosed, describe the testing and verification process instead of inventing a percentage.

Assign one primary answer page to each question. Begin with conditions and a direct answer, then provide evidence, limits and a next step. A case study should explain the original problem, actual implementation and measurement method; customer logos are insufficient. Keep cases without permission in internal research.

Classify external material by its relationship to the claim: company statement, partner confirmation, independent test or user feedback. Ten copies of one press release still represent one original source. They may increase discovery opportunities but are not ten independent pieces of evidence. Paid content must not be described as independent evaluation.

Check identity and capabilities with branded queries first. Then use buyer questions without the brand name to observe natural discovery. Report the two groups separately. A mention produced after inserting the brand into a prompt does not demonstrate unprompted discovery.

For each target platform retain the exact prompt, interface, web-search state, time, session method, original answer and citation URLs. Use new sessions for single-turn questions. Preserve context within a multi-turn decision chain and report it separately. Repeat under comparable conditions and retain absences and errors, not only successful screenshots.

Create a ticket for each error: the incorrect sentence, corresponding fact card, conflicting public sources, page to change and next review date. If an answer has no citations, record “no visible citation.” Similar wording alone does not prove that a particular article supplied the answer.

Keep two dashboards: one for controlled-query answer performance and one for actual visitors and sales leads. Sample mention rates are not user exposure, and crawler visits are not potential customers. These records can inform each other, but cannot be multiplied to estimate sales.

Add an optional source field to the enquiry form or CRM. Let users name the AI platform and original question, or say they do not remember. Keep observable referral information without filling gaps with assumptions. Direct visits and branded search may also follow advertising, events or referrals. Deduplicate qualified enquiries and exclude bots, repeated requests and recruitment messages.

The schedule delivers an audit, revisions and a first review. It does not promise rankings or orders within two weeks. One person may hold several roles, but fact approval and writing responsibilities must be explicit.

Start with sample size and missing data. Then report identity errors, non-branded mentions, cited pages and actual leads by question type. Branded search and site traffic provide clues, but record simultaneous advertising and events. Growth during the same period does not establish GEO causation.

Continue when a human review confirms the repairs and fixed questions repeatedly produce more accurate descriptions. Adjust when query visibility rises but enquiry fit falls, revisiting the audience and service boundaries. Pause expansion when facts cannot be verified, the site fails or there is no evidence of customer demand.

The deliverable is a reviewable set of questions, facts, pages and lead records. It helps decide where to spend money and time. It cannot guarantee citations or establish that an AI platform has certified the brand.