Loading...
Loading...
13Module
Pitfalls and misconceptions that hurt your AI optimization efforts. Learn from others' failures to accelerate your success.
Available3 lessons18 min
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.
A list of mistakes is the easiest module to pad and the least useful to read, so these are grouped by why each one happens rather than by what it is.
Some arrive as products, and are attractive because they turn a slow per-platform problem into one number. Some come from doing a real skill harder in a place where more of it stops helping. And some cost a quarter or a year because nothing about them looks wrong at the time.