You cannot improve what you cannot measure. But how do you measure visibility in a chat box? Traditional rank trackers (SEMrush, Ahrefs) track positions on a SERP. They do not track mentions in a generated paragraph.

The New Tool Stack

We are building tools to probe LLMs with thousands of permutations of a query to calculate Generated Share of Voice (GSV).

The Methodology

  1. Define a Query Set: “Best CRM,” “CRM software,” “Sales tools.”
  2. Permutation: Use an LLM to generate 100 variations of these questions (“What CRM should I use if I am a startup?”).
  3. Probe: Run these 100 queries across GPT-4, Claude 3.5, and Gemini via API.
  4. Extraction: Parse the text output. Extract Named Entities (NER).
  5. Frequency Analysis: Calculate the frequency of your brand’s appearance vs. competitors.

The “Share of Sentiment”

It is not just about frequency. It is about sentiment.

  • “HubSpot is great.” (+1)
  • “Salesforce is expensive.” (-1)

If you have 50% GSV but 80% negative sentiment, you are losing. Advanced visibility tools now provide a Sentiment-Weighted GSV score.

Actionable Insight

If your GSV is low:

  • You need more co-occurrence in training data.
  • You need to check your robots.txt (are you blocking the crawler?).
  • You need to check your structured data (is the model confused about what you do?).

The “Share of Intention” Metric

Beyond Share of Voice, we are looking at Share of Intention. Does the model recommend you for the Right Reason?

  • Query: “Cheap CRM.” -> You rank #1. (Bad if you are Premium).
  • Query: “Enterprise CRM.” -> You rank #0.

Vector Alignment Audit: You must audit why you are ranking. If you are ranking for “Cheap” because of a hallucination, you will get bad leads. You need to use “Negative Constraints” in your content (“We are not a cheap solution…”) to push your vector away from the “Cheap” cluster.

Glossary of Terms

  • Agentic Web: The specialized layer of the internet optimized for autonomous agents rather than human browsers.
  • RAG (Retrieval-Augmented Generation): The process where an LLM retrieves external data to ground its response.
  • Vector Database: A database that stores data as high-dimensional vectors, enabling semantic search.
  • Grounding: The act of connecting an AI’s generation to a verifiable source of truth to prevent hallucination.
  • Zero-Shot: The ability of a model to perform a task without seeing any examples.
  • Token: The basic unit of text for an LLM (roughly 0.75 words).
  • Inference Cost: The computational expense required to generate a response.