How to Measure LLM Citations and LLM SEO Performance: The LLM Visibility Index Explained

Executive stakeholders consistently request a single composite metric for generative recommendation performance. The LLM Visibility Index is AskLLM's 0–100 score for how often ChatGPT and Gemini name and endorse your brand on realistic buyer queries. We weight mention frequency, position in the answer, sentiment, and cross-model consistency.

A score of 78 should indicate materially stronger recommendation signal than 45, not a vanity mention in paragraph four. The following sections explain how we construct the index and what it cannot tell you.

We score query responses, not crawls

The starting point is a query, not a site audit.

We generate exactly eight buyer questions from your category, use case, persona, and competitive context: problem-oriented prompts a real buyer would type, not "[your brand] review."

We run them across ChatGPT and Gemini, multiple times with phrasing variation so we are not scoring a single fortunate snapshot.

Per query and model we record: did you appear, where in the answer, how you were described, with what confidence. The index compresses that grid into one number executives can track.

Four components

Mention frequency measures what percentage of responses include you. Straightforward, but incomplete alone.

Position weighting recognizes that first recommendation with rationale outperforms buried "also consider." Frequency is adjusted for where you appeared.

Sentiment distinguishes "go-to for teams that need Y because of Z" from neutral list filler and qualified "has limitations with Y." We capture whether mentions convert to consideration.

Cross-model consistency identifies fragmentation. Strong in ChatGPT but absent in Gemini indicates fragmented presence, distinct from durable visibility across both. High consistency usually means broad training-source absorption, not one-model luck.

What the ranges mean (directionally)

Below 30 indicates weak or fragmented recall. Typical for new brands without corpus depth, or brands with inconsistent third-party description.

30–60 is where most established brands land. You are in the consideration set, not the default pick.

Above 60 indicates consistent appearances with endorsement language, often first or second recommendation.

100 represents category-defining saturation: unprompted detail, consistent characterization. Rare, and typically earned over years of independent coverage.

Use bands as direction, not precision instruments.

What the score does not tell you

It is a snapshot, not a trend line. 65 today does not prove improvement; run audits over time for delta.

It does not diagnose why without reading recommendations. 40 could mean coverage gap, inconsistent description, or a genuinely immature category vocabulary.

The index is a starting point for diagnosis, not a final verdict.

The honest limitation

Outputs are stochastic. Training data is not fully transparent. Two audits a week apart might shift a few points.

We smooth noise with repetition and phrasing variation. Relative ordering (70 vs 35) is reliable even when absolute points drift.

Why we built it

Almost no organization runs manual two-model query batteries consistently. Most assume they are visible because they rank on Google.

A number forces action. 42 → 61 three months later demonstrates that the work paid off in a way informal observation never will.

Startups: Put the index in context with our AI brand visibility audit for startups.

Disclaimer: Educational research from AskLLM (askllm.io). Production visibility audits run buyer-intent prompts on ChatGPT and Gemini only; see askllm.io/methodology for the protocol.

Run your free audit at askllm.io. See your score and what's driving it.