FR
EN
ES
DE
STS (Semantic Text Similarity) detection measures the semantic closeness between AI-generated responses and a brand's own content. A high STS score indicates the AI heavily relies on the brand's content to formulate its answers, signaling strong GEO influence.
What is STS Detection?
STS Detection (Semantic Text Similarity) is an analysis technique that measures semantic closeness between two texts. In the GEO context, it compares responses generated by AI models with content published by a brand on its own website.
A high similarity score means the AI "borrows" the brand's vocabulary, formulations and information — a sign that content directly influences generated responses.
How STS Detection Works
- Vectorization: texts are converted into numerical vectors via embedding models
- Comparison: cosine similarity between vectors is calculated
- Score: a similarity percentage is assigned (0% = no relation, 100% = identical)
Beyond Exact Matching
STS goes far beyond simple word matching. It captures meaning similarity even when wording differs.
Applications in AI Labs Audit
- Measure brand content influence on AI responses
- Identify which pages are most "borrowed" by LLMs
- Track influence evolution over time
- Compare influence against competitors
Link with Mention Rate
STS complements the mention rate: while mention rate counts if the brand is cited, STS measures how much the brand's own content influences the response formulation.
A brand may not be explicitly cited but see its content widely reused — STS detects this "invisible" influence.
More details in our article AI Bot Tracking and LLM Mentions.