1. Where the data comes from
We collect the answers that AI engines give to real buyer questions.
- Engines we track: ChatGPT, Gemini, Google AI Overviews, Google AI Mode, Perplexity, Microsoft Copilot, Grok and Claude.
- You choose the engines. Each brand starts with ChatGPT, Gemini and Google AI Overviews. You can switch to any of the other engines at any time.
- The same answers people see. For most engines, we collect the answers from the public interfaces that people use, not from developer APIs, because API answers can differ from what a buyer sees. For Claude, we collect the answers through its official API.
- Live and memory answers. With web search on, AI builds its answer from the pages it finds today: we call this live visibility. Without search, AI answers from what it learned in training: we call this memory visibility. To measure memory visibility, we run conversations through the engines’ APIs, so we see what each model already knows about your brand.
2. Which prompts we run
- Built from your brand. Vosapio writes the prompts from your brand profile and your topics: 8 active prompts per topic, plus suggestions you can switch on.
- Your own prompts too. You can add, edit or remove prompts at any time.
- In your buyers' language. Prompts can be written in the language your buyers use, so you see the answers they see.
3. How we run prompts
- Every day. Vosapio runs each tracked prompt once per day on each engine. Some accounts choose a weekly schedule instead.
- Per country. AI engines answer differently in different countries. Each prompt runs with a country setting, where the engine supports it.
- Trends, not single answers. One answer from one day proves little, because AI answers change from run to run. We look at the same prompt over many days and report the trend.
- Conversations. Buyers often ask follow-up questions. We also run conversations, with several questions in a row, to see how the answers change from the first question to the last.
4. How we read each answer
MentionWe find the brand name in the answer, including common variants. For example, “Acme Pro” counts as a mention of “Acme”.
PositionThe order in which the brands first appear in the answer. The first brand named is position 1.
RecommendationCounted only when the answer clearly recommends the brand or calls it the best choice. A passing mention does not count. When in doubt, we do not count it.
SentimentAn AI model rates how the answer describes the brand, on a five-step scale from very positive to very negative. It also writes a one-line reason that you can read.
CompetitorsOnly the competitors you accept count in your numbers. Before we suggest a competitor, we check that it is a direct competitor.
SourcesThe links the answer cites. Your own site and your competitors' sites are marked. Other sites are sorted into types: corporate, editorial, user-generated, reference, institutional or other.
We keep the full answer text, so you can open each answer and see where every number comes from.
5. How we calculate the numbers
| Metric | How it is calculated |
|---|
| Visibility | Answers that mention the brand, divided by all collected answers, × 100. With no answers yet, we show “no data”, not 0%. |
| Share of voice | Mentions of the brand, divided by the mentions of all brands in the same answers. |
| Average position | The brand’s position, averaged over the answers that mention it. Lower is better. |
| Sentiment score | From 0 to 100: very positive = 95, positive = 80, neutral = 60, negative = 35, very negative = 10, averaged over the answers that mention the brand. |
6. When an answer cannot be collected
- Automatic retries. When an engine does not answer, Vosapio tries again. When an engine is down, the prompts wait and run later. You do not need to do anything. We make sure your prompt is not lost.
- No false data. A login page, a bot check or an empty answer is not counted as an answer, so it cannot change your numbers.
- Shown openly. When no answer could be collected, you see “No answer was collected for this run” instead of a gap.
- Not charged. A run without an answer does not use credits.
7. Where your data is hosted, and how we treat it
- Made in the EU. Vosapio is built and run by a company based in the European Union.
- Hosted in the EU. The Vosapio app and its database run in France (France Central region). This website runs in the West Europe region.
- Only what is needed leaves Vosapio. To get the answers, your prompts are processed by the AI engines and by the infrastructure we use to run them. Some of these services operate outside the EU. With each prompt we send only the prompt text and its country setting, never your name, email or account details.
- We do not sell your data. We do not sell, trade or transfer your personal information to third parties for marketing.
- Your rights. You can ask to see, correct or delete your personal data. Our Privacy Policy explains how.
8. How we use data in articles
- When an article gives a number, it states the period and the number of prompts it is based on.
- Examples on our website and in our articles use made-up brand names. We never publish a customer's data without their permission.
9. How we keep articles current
- Every article shows the date it was last updated.
- AI engines change often. When a change makes an article wrong or outdated, we update the article and its date.
10. Found an error?
Write to office@vosapio.com. We will check it and correct the article if it is wrong.
See these numbers for your own brand
Track how AI engines answer the questions your buyers ask, every day.