How Yovara measures AI visibility
In one sentence
Yovara asks AI assistants the questions your customers ask, several times, through the assistants' official services, and reports how often they mention you, who they mention instead, and which sources they rely on. Then it compares those answers with your website to explain why, and what to change.
1. The questions
- We read your website (only public pages, respecting your robots.txt) and understand what you do, for whom and where.
- From that, we draft about 30 questions a potential customer might ask an AI assistant before choosing a provider, grouped by intent and stage: discovering, comparing, deciding.
- Most questions don't name you (about 25). These measure discovery: whether AI brings you up when someone asks about what you sell. A few name you (about 5). These measure brand understanding.
- You review and edit the questions before anything is measured. Then they are frozen: re-measuring uses exactly the same questions, so results stay comparable over time.
2. The AI services
We currently observe:
- OpenAI, through its official API with web search;
- Perplexity, through its official API.
We don't scrape consumer apps like chatgpt.com, and we don't use services whose terms forbid analysing their answers.
What this means: we see the answer an anonymous user, with no history, in your market would get through each service's official channel. Consumer apps can differ because of personal history, memory, model versions and experiments. Yovara measures a consistent, reproducible proxy. It is not a copy of what a particular person sees on their own phone.
3. Repetitions and states
AI answers vary from one run to the next, so each question is asked 3 times on each service (6 answers per question). A question is:
| State | Rule |
|---|---|
| Visible | you are mentioned in half of the answers or more |
| Sometimes | you are mentioned in some answers, but fewer than half |
| Competitors appear instead | you are never mentioned, and a competitor is named in half of the answers or more |
| Invisible | neither you nor a specific competitor is named consistently |
4. The numbers
- AI visibility (the main number): across the questions that don't name you, the share of answers that mention you. Partial visibility counts (appearing in 2 of 6 answers counts as 33 %).
- Brand understanding: the same, across the questions that name you. It's shown separately, so it never inflates your visibility.
- Every number links to the answers behind it. You can open any question and read what each AI actually said, with its sources.
5. Competitors and sources
- Competitors: the ones you tell us about, plus the ones AI itself recommends in its answers. When a name recurs across answers, we add it as a discovered competitor.
- Sources: for each answer we record the pages AI cited, and (when the service reports it) the pages it read but didn't cite. We tell apart your own sites, competitors' sites and third-party sites.
6. Why you are (or aren't) visible
We compare every question where you're not visible with your website and the sources AI used. That tells us whether you have no page that answers it, a page that answers it weakly, a page AI never read, a page AI used without naming you, or whether AI relies on competitors' pages or third-party sites where you're absent. We also run a technical check of your site (for example, whether AI crawlers are blocked or content only appears with JavaScript).
7. Recommendations
Each recommendation is derived from those observations and links to its evidence. It's labelled Observed (seen in the answers), Inferred (a pattern in the data) or Good practice. We use AI to phrase recommendations in plain language, never to decide what to recommend.
8. Re-measuring
Re-measuring asks the same frozen questions again, with the same settings. We compare with your first measurement. We also show how much AI's sources changed on their own, because AI answers move even when nothing changes on your side.
Is a change real? We compare each question with itself (the same question, before and after) and calculate the margin of the overall change. We only call it a clear change when it is larger than that margin; otherwise we say it is within the margin. With few questions, or answers that move a lot, the margin is wider.
How sure is the main number? We show it with its margin (for example, 64 % ± 12 points). Questions about one topic are related to each other, so we treat each question, not each answer, as one unit: with other, similar questions your result could differ by about that much.
9. What we don't do
- We don't promise rankings or guarantee what any AI will say. No one can.
- We don't manipulate AI systems, generate spam or publish anything on your behalf.
- We don't scrape consumer AI apps.
Yovara · Be found. Be understood. Be recommended.