A visibility score you can actually check.Four measured numbers, one published formula.
The four components
Presence rate
40%The share of sampled answers that name your brand at all. This is the floor — every other metric is conditional on it. A brand named in 41% of answers has a very different problem from one named in 4%.
Citation rate
30%The share of answers where your own domain is cited as a source. Rarer than presence and worth nearly as much: a mention is reputation, a citation is traffic, and it is the position hardest for a rival to take from you.
Share of voice
20%Your mentions divided by every brand mention across the sample. This is what separates dominance from being the ninth vendor in a list of nine — presence alone cannot tell those apart.
Average position
10%Where you land in the order named, when you are named. It matters, and it matters least: being named third instead of first is a far smaller problem than not being named. Decays to zero by sixth.
Weights sum to 1 and are unit-tested to stay that way. Changing them would change every past customer's score, so we treat them like a price, not an implementation detail.
What the grade means
| Score | Grade | What it means |
|---|---|---|
| 80–100 | A · Dominant | Named consistently and cited as a source. The job is defending it. |
| 60–79 | B · Visible | You show up reliably, but rivals are taking citations you could own. |
| 40–59 | C · Inconsistent | Present for some buyer questions, absent on others. The gaps are winnable. |
| 20–39 | D · Barely present | Mentioned rarely, almost never cited. Buyers researching your category are not seeing you. |
| 0–19 | F · Invisible | Effectively absent from AI answers in your own category. |
What gets measured, and what gets disclosed
Four engines — ChatGPT, Claude, Gemini and Perplexity — each through its own API with web search or grounding enabled. About 40 buyer-intent prompts, generated from your site and then frozen. Three passes per prompt. 480 answers.
Every report states the sample size, the engines, the dates, the prompt mix, the score formula, and how many runs failed and were excluded from the denominators. It also states the limitation: API answers are close to, but not byte-identical with, what a person sees in the consumer ChatGPT app or in Google's AI Overviews.
We publish that because a score with no provenance is worth nothing. Read the methodology in full.
Questions people ask before buying
- What is AI visibility?
- How often, and in what way, AI answer engines surface your brand when someone asks about your category. It has four measurable components: whether you are named at all, whether your own site is cited as the source, how much of the total brand mention volume is yours, and where in the answer you appear. Anything reported as a single number without those four underneath it cannot be checked.
- How is the visibility score calculated?
- Presence rate 40%, citation rate 30%, share of voice 20%, average position 10%, combined into 0–100. Position decays linearly from first mention to zero at sixth. Citation is weighted almost as heavily as presence despite being much rarer, because being the source is what sends traffic and is the hardest thing for a competitor to take from you.
- Is a low score bad?
- It is common. Most challenger brands in B2B software land in the 20s and 30s: named in some answers, almost never cited. That is the normal starting point, not a catastrophe — and it is winnable, because the answers are usually built from a small set of third-party pages you can go get onto.
- How many answers is the score based on?
- The paid audit samples 480: about 40 buyer-intent prompts across four engines, three passes each. The free scan samples 20 — five prompts, four engines, one pass. Twenty answers is enough to see whether you have a problem and not enough to size it precisely, which is exactly what the free tier is for.
- Can I re-run it later to see if the work paid off?
- Yes, and that is what the frozen prompt set is for. The prompts generated for your first audit are stored and replayed verbatim on a re-scan, so the two runs are comparable prompt for prompt. Regenerating them would turn a before/after into two unrelated samples.
- What do you do with my data?
- We fetch your homepage to infer your category and competitors, then store the prompts, the answers and the scores against your scan. We do not sell data or share it with third parties beyond the AI providers we query and the infrastructure that runs the product. The report is reachable only through a long random token in its URL — there is no login, so treat that link as private.
Related
- GEO audit
The generative-engine-optimization framing, and the difference between a readiness check and a measurement.
- AEO audit
Why the answer-engine checklists you will find elsewhere are not audits.
- The tool landscape
Page checkers, answer-sampling platforms, agency engagements.
- Sample report
The full paid deliverable, rendered from demo data.
Find out what the engines actually say about you.
The free scan gives you the same score computed the same way, on a smaller sample — 20 answers instead of 480. It is the honest way to find out whether the paid audit is worth $349 to you.