Sycophancy

Sycophancy means an AI model gives you the answer you «want to hear» rather than the most accurate one. In a GEO context this is critical: a user asking a leading question («which CRM is best for growth companies?») may receive a different answer than a user asking a neutral question («compare the five most widely used CRMs»). This means your mention rate in Brand Analyzer can vary significantly depending on how the prompt is framed – not just who actually has the best product. In 3LA, the judge model currently scores sentiment (−1.0 to +1.0) and whether a competitor appeared stronger, but does not explicitly detect sycophantic patterns. Full sycophancy detection requires multi-variant prompt batches (neutral, leading and counter-leading framing) combined with sentiment-delta analysis across variants.