Understanding Your AI Visibility Score

The Bingly visibility score is a single number between 0 and 100 that answers one question: when a user asks an AI model about a topic related to your keyword, how likely is it that your website gets cited?

Traditional SEO rank tracking tells you where your page appears in a list of blue links. AI visibility tracking is different — there is no ranked list. A model either includes your site in its answer or it does not, and if it does, it may feature you prominently or bury you in a footnote. The score captures both dimensions.

The aggregate score is a weighted average across all models you selected, combining two components:

Each model contributes equally to the aggregate unless you have applied custom model weights in your account settings. The final score is normalised to a 0–100 scale.

Use these bands as a guide for prioritising your optimisation work. They reflect thresholds observed across thousands of Bingly checks — not arbitrary cutoffs.

Not all citations are equal. When an AI model answers a question, sources mentioned first — or named as the definitive answer — carry far more weight than sources buried in a long list. Bingly tracks the position and context of every citation and converts it to a prominence multiplier.

Improving prominence — moving from a secondary mention to a primary citation — has a larger impact on your score than merely increasing the number of models that mention you. One strong primary citation beats five incidental mentions.

The aggregate score tells you your overall position, but the per-model breakdown is where the actionable insights live. Each model has its own panel showing:

• Citation status — Cited / not cited for this run, plus a run-average citation rate.
• Prominence level — The highest prominence achieved across sampled runs for this model.
• Competitors cited instead — Which other domains the model mentioned when yours was absent — useful for competitive intelligence.
• "How the model sees your page" — The model's own characterisation of your content: what it thinks the page is about, what topics it associates with your domain, and what gaps it identified.

Look for patterns across models. If ChatGPT and Gemini both cite you but Claude does not, that suggests a training-data or content-structure issue specific to how Anthropic's models index your domain — not a universal problem. Address the universal issues first (those flagged by all models), then work on model-specific gaps.

Large language models are non-deterministic: even with the same prompt, they do not produce identical responses every time. This is by design — the "temperature" parameter that controls response variety is what makes LLMs useful for open-ended tasks.

For AI visibility tracking, this means a single check is a data point, not a verdict. Bingly mitigates this by:

• Sampling each model multiple times per check and averaging the citation rate.
• Flagging when a result is based on a small sample so you can interpret it accordingly.
• Storing all historical checks so trends over time are more reliable than individual snapshots.

Different models also differ in training data, knowledge cutoffs, and citation behaviour. A model trained primarily on academic text will cite differently from one optimised for conversational search. This is not noise — it is real signal about which AI surfaces your audience actually uses.