Depra AI turns stored AI answers into four scores: visibility, share of voice, position and sentiment. Each one is calculated for your brand and for every tracked competitor, on the same questions. Read this page when you need to know what a number means before you act on it.
How is each score calculated?
| Score | What it tells you | Calculation |
|---|---|---|
| Visibility | How often AI names you | Answers that name the brand, divided by all answers in the period |
| Share of voice | Your share of all brand mentions in the answers | Mentions of the brand, divided by mentions of all tracked brands |
| Position | Where you come when AI lists options | The average order in which the brand is first named, where 1 is first. Lower is better |
| Sentiment | Whether AI describes you positively, neutrally or negatively | A score from 0 to 100: positive is 100, neutral is 50, negative is 0, averaged across mentions |
Visibility counts answers. Position and sentiment count mentions, so they rest only on the answers that name the brand.
Share of voice has its own guide: AI share of voice.
How does Depra AI read an answer?
An AI model reads each stored answer. For every tracked brand the answer names, it records four things.
- Position: the order in which the brand is first named, where 1 means named first.
- Sentiment: whether the answer describes the brand positively, neutrally or negatively.
- Quote: the exact sentence that names the brand, so you can check every mention.
- Confidence: how sure Depra AI is that the text refers to the brand.
These rules decide whether a mention counts.
| Case | What happens |
|---|---|
| The mention has no supporting quote from the answer | It is dropped |
| Confidence is below 50% | It never counts towards any score |
| The answer uses another spelling of the brand, or its domain | It counts as the same brand |
| The brand name sits inside a longer word | It does not count |
What does the 95% band mean?
The 95% band is the range the true figure most likely sits in. AI engines can name different brands when the same question is asked twice. So a score is only as precise as the number of answers behind it.
| Answers behind the score | 95% band at 50% visibility |
|---|---|
| 20 | About 30% to 70% |
| 400 | About 45% to 55% |
Your report shows this band for visibility, as a 95% confidence range, next to the number of answers. Reports read through the Depra AI MCP server, the read-only connection for AI apps, show a band for visibility and for share of voice.
Changes are judged the same way. The daily brief, a page inside the app, lists a visibility change only when the two days' bands do not overlap. Each day must also have at least 5 answers.
What if a number rests on too few answers?
Do not treat it as a finding. Reports read through the Depra AI MCP server flag a row that is too small to read.
The same goes for a comparison marked as having no clear gap or no clear winner. Do not act on it as if it were a result.
Note: An empty list of recent changes is a real result. It means no score moved by more than the normal variation between runs.
What is left out of the scores?
- Questions that name your brand: left out by default, so the score is not inflated.
- Searches with no AI Overview: stored separately and kept out of every score. A missing Overview never counts against you.
- Weak mentions: a mention with no supporting quote, or below 50% confidence, does not count.
- Answers from before you tracked a competitor: a competitor is scored only on answers collected after you started tracking it.
Why branded questions are left out is explained in prompts and topics.
What does the calculation look like with numbers?
Note: This example uses made-up numbers. It is an illustration, not customer data.
Say a brand has 200 stored answers for the period, and 50 of them name the brand. All tracked brands together are mentioned 400 times in those answers.
| Score | Numbers used | Result |
|---|---|---|
| Visibility | 50 answers name the brand, out of 200 answers | 25% |
| Share of voice | 50 mentions of the brand, out of 400 mentions of all tracked brands | 12.5% |
| Position | Named first in 10 answers, second in 20 and third in 20 | 2.2 |
| Sentiment | 30 positive, 15 neutral and 5 negative mentions | 75 |
Position is 10 x 1 plus 20 x 2 plus 20 x 3, divided by 50 mentions, which is 2.2. Sentiment is 30 x 100 plus 15 x 50 plus 5 x 0, divided by 50, which is 75.
How do I check a number?
Open the answers behind it. Every mention carries its quoted sentence, and every answer is stored in full with the websites it cited. If a score looks wrong, read the answers.
The full rules are on how Depra AI measures AI visibility. A method you can run by hand is in how to measure AI visibility.