GuidesNovember 3, 2026 · 4 min read
What is an AI visibility score? How we calculate ours
A plain-English guide to the AI visibility score: what it measures, how rank in the answer is weighted, why failed answers don't count, and what's good.
Key takeaways
- The score is the share of AI answers that name you, weighted by how high you're named, from 0 to 100.
- Named first counts in full; second counts 0.85, third 0.7, fourth 0.6, and fifth or lower 0.5.
- Answers that fail are left out entirely, so an engine outage never looks like a drop.
- Most small brands start below 20. The trend over weeks matters more than any single day.
"AI visibility" has become a common phrase, and every tool measures it a little differently. This post explains exactly what Citemark's AI visibility score measures and how it's calculated, so you can decide how much to trust it and what to do with it.
What the score measures
The score answers one question: when your buyers ask AI assistants about your category, how often are you recommended, and how prominently?
To measure that, Citemark keeps a list of buyer questions for your brand, like "best invoicing tool for freelancers" or "Harvest alternatives". Each question is put to each AI engine you track. Every question × engine pair is one "cell". For each cell we read the full answer and note whether your brand is named and, if it is, in which position among the products named.
The score is the weighted share of cells that name you, shown on a scale from 0 to 100.
How rank in the answer is weighted
Being named first in an answer is worth more than being named fifth. Buyers read top-down and often go with one of the first suggestions. So each cell is weighted by your position:
| Your position in the answer | Weight |
|---|---|
| Named first | 1.0 |
| Named second | 0.85 |
| Named third | 0.7 |
| Named fourth | 0.6 |
| Named fifth or lower | 0.5 |
| Not named | 0 |
The score is the average weight across all cells, times 100. If you're named first in every answer, the score is 100. If you're named in half the answers, always in third place, the score is 35.
A worked example
Say you track 5 questions on 2 engines, so 10 cells. In one day's answers you're named first twice, second once, fourth once, and not named in the other 6.
- Add up the weights: 1.0 + 1.0 + 0.85 + 0.6 = 3.45.
- Divide by the number of cells: 3.45 / 10 = 0.345.
- Multiply by 100: a score of 34.5.
Why failed answers don't count
Sometimes an engine doesn't answer: an API error, a timeout, a rate limit. Those cells are marked "no data" and are left out of the calculation completely. They're not counted as "not mentioned".
This matters more than it sounds. If an outage counted as a miss, your score would drop every time an engine had a bad hour, and you'd chase problems that aren't real. With failed cells excluded, a drop in the score always means the answers that did come back named you less.
Why we use each question's latest answer
On paid plans your most important questions are checked every day and the rest every week. A day's score uses each question's most recent answer, not only the answers from that day. Otherwise the score would jump up and down depending on which questions happened to run that day.
Share of voice: the other number to watch
The score only looks at you. Share of voice looks at everyone: of all the brand mentions across the answers, what share is yours, and what share goes to each competitor you track? It answers "who is winning my category in AI answers?" and it's the best way to spot a rival that's gaining ground before it shows up in your sales.
The dashboard also shows visibility per brand (the share of answers that name each brand), your average rank when you're named, and the sentiment of the sentences that mention you, so you can see whether you're recommended or mentioned with a warning.
What is a good AI visibility score?
It depends heavily on your category and how specific your questions are. From what we've seen:
- Category leaders with near-synonym brands (think Wikipedia for encyclopedias) score 80 to 95.
- Strong, well-known products in competitive categories often land between 50 and 80.
- Smaller and newer brands usually start below 20, often near zero for broad questions.
In our 20-brand test the scores ranged from 95 to 7, with a made-up website at 0. A low starting score isn't a failure; it's a baseline. What matters is the direction over weeks after you make changes.
Why the score moves around
AI answers aren't deterministic. The same question can get a different answer an hour later because of model updates, changes in what web search returns, or plain randomness in how the model writes. Why AI answers change goes into this in detail. The practical upshot: read the trend line, not a single day's number. A 7-day or 30-day view smooths the noise.
Per-engine scores
The headline score combines the engines you track. The dashboard also breaks it down per engine, because the engines often disagree. It's common to be named regularly by one and rarely by another: each engine searches the web differently, leans on different sources and was trained on different data. When one engine lags, look at the pages that engine cites for your questions. They're often different from the pages the others cite, and they tell you where to work for that engine specifically.
You can also filter the dashboard to a single engine to see your visibility, average rank and sentiment on it alone, and compare that with your competitors on the same engine.
What to do with your score
The score tells you where you stand; it doesn't tell you what to fix. For that, look at the cells where you're missing and read the answers. Who was named instead? Which pages did the engine cite? Those cited pages are the most concrete lead you'll get, because they're the sources the engine used to build the answer. The playbook for getting recommended turns that into a to-do list.
To see your own score today, run the free AI visibility check. It takes about a minute and needs no signup.