Common GEO Mistakes to Avoid

The common GEO mistakes fall into three groups. The ones sold to you: a single blended visibility score, which averages across assistants that share roughly a tenth of their cited domains and so describes none of them; llms.txt treated as a confirmed ranking factor; the 40% figure from the original GEO paper quoted as a promise rather than as a benchmark result; and optimising a tool's own score, whose method can change without notice. The ones caused by competence: publishing more on the same question when selection discards material before merit, writing for the phrase a buyer typed rather than the category the system actually searches, treating schema markup as the work rather than as a record of consistency, blocking every agent carrying a vendor's name and losing live citations with the training crawlers, and auditing everything before fixing anything. And the ones you only see later: measuring with no noise floor, rewording the question set between quarters, letting one person hold the method, and claiming a win from work shipped two weeks before the measurement.