Visibility Score
The headline number in Omnipresence: what the Visibility Score counts, how its confidence band is calculated, and how to report it honestly.
Updated October 20, 2026
The Visibility Score is the share of grounded probes in which a model recommended you. It is the number at the top of the overview, the one the trend chart plots per platform, and the one worth putting in a report.
What it counts
For every probe, the answer is read and the brands it recommends are extracted. If your brand, or any of its registered name variants, is in that recommendation set, the probe counts.
Two consequences:
- It is a rate, not a count. Twelve mentions means nothing without the number of probes behind it.
- It depends on Name matching. A spelling the model used that you have not registered reads as a miss, which is why a surprisingly low score is a mapping problem more often than a market problem.
Why it is the grounded rate
The score uses probes that ran with web search available, so it reflects what a model can find about you today. That is the part your work changes within a quarter.
The same figure computed on training-mode probes answers a slower and different question, and the product keeps them apart rather than averaging them. See grounded and training.
The confidence band
Every rate here comes with a band, calculated as a Wilson interval on the number of probes behind it. Read the band as the range the true rate is likely to sit in.
| What you see | What it means |
|---|---|
| A narrow band | Enough probes. Treat the number as real. |
| A wide band | Not enough probes yet. The number is not telling you much. |
| A move inside the band | Noise. It will come back. |
| A move outside the band | Something changed. Worth explaining. |
This is why a measurement fires a batch of probes rather than one: a single probe on a question like this is close to a coin flip.
Reading it per platform
Visibility Score by Platform splits the same rate across ChatGPT, Claude, Gemini and Perplexity. Expect the spread between them to be larger than most week-on-week movement, because they retrieve from different places and weight sources differently.
A platform that is far below the others is the most useful finding on the screen: compare its cited sources against a platform where you do well, and the gap is usually one or two sources it trusts that do not carry you.
Reporting it
Three things make the number defensible to someone who was not there.
- The band, or at least the probe count. "38 percent, n=120" is a measurement; "38 percent" is a claim.
- The prompt set. "Across 24 buyer prompts" tells the reader what was asked.
- The mode. Say it is the grounded score, so nobody compares it against a figure computed another way.
Put the control prompt on the same chart. A rising score next to a flat control is the strongest evidence that the work did it, and it costs one extra prompt.
Questions about visibility score
- Is a mention the same as a recommendation here?
- Yes. The score counts probes where the model recommended you, so what the product calls a mention is already a recommendation rather than a passing reference.
- Why does it say N/A instead of 0 percent?
- N/A means no grounded probes in the range, so there is nothing to average. Zero percent means probes ran and never recommended you, which is a real finding.
- Why does it move when I have not changed anything?
- Sampling. Each measurement is a batch of probes against a model that answers differently every time. If the move is inside the confidence band, nothing happened.