GEO figures by access group, with ranges and samples

AI Visibility now separates what it measures from how sure it is.

• Two groups, never pooled. ChatGPT and Gemini (the apps) and Qwen, Doubao and DeepSeek (the API panel) each get their own mention rate, source rate and share of voice. The Growth Score reads the apps when they have answers, else the API panel.
• Ranges and a floor. Rates average prompt and engine pairs and come with a 95% range. Under 30 pairs the page says not enough answers yet and does not call a change; with enough, a change against the previous 14 days reads higher, lower or no change detected.
• Several answers per prompt. Run can ask for up to 10 answers per engine. Each is a separate paid run and a stored sample, and a same-day rerun no longer overwrites the last answer.
• Prompt versions. A new text on a prompt with answers becomes the next version; the old one is retired with its answers. Prompts also take a kind (tracked or control) and a target page, and a prompt that names your brand is shown but not counted.
• Source categories. Baidu Baike, Zhihu, WeChat, Toutiao, Baijiahao, Dianping, complaint sites and video join the existing ones, and the page lists the other sources behind answers that name you.
• Fact check. Open a stored answer and compare it with your confirmed brand facts. It is a paid AI check (see the pricing table), and every finding quotes the answer itself.