// Straight answer
Short answer
Write a fixed list of the questions your buyers ask, run each one in a fresh or temporary chat with memory off, repeat it a few times, and record whether you are mentioned and whether your own site is cited. Then repeat in Perplexity, Gemini and Copilot. One answer is a sample, not a score, so compare patterns across runs and dates.
Answered by Tarek Abou Rjeily, AI consultant & builder in Beirut, for businesses in Lebanon and the Gulf.
Typing your business name into ChatGPT and seeing a good (or bad) answer tells you very little. Answers vary with the exact wording, your location, the model or mode in use, your chat history and the day. If the assistant already knows you from earlier chats, it can also flatter the result.
So treat every answer as a sample and build a small method around it. The steps below take about an hour for a first baseline.
Start with the buyer's side, not yours. A name search ('Tell me about Acme Clinic') only tests whether the assistant has heard of you. The questions that matter are the ones a stranger asks before they know you exist.
A mention means the answer names your business. A citation means it shows a link to a source. You can be named with no source behind it, or a directory profile can be cited while your own site is absent. The fixes differ: a mention gap is usually an evidence and entity problem, while a missing citation is more often a crawl, indexing or content problem.
Also note who is cited instead of you. If the same three directories or a competitor's blog appear every time, that list is your to-do list.
Run the same prompts in ChatGPT, Perplexity, Gemini and Microsoft Copilot, because each uses different sources and behaves differently. A business named by one and absent from another is useful information.
As of November 2026, where available, Bing Webmaster Tools has an AI Performance report showing citation activity across Microsoft Copilot and Bing AI summaries, by page and by the queries used to ground answers. Bing says plainly that citations are not rankings or proof of traffic, so use it for trends. Google announced generative AI performance reports in Search Console in June 2026, initially for a subset of sites. If your property does not have either, do not substitute a third-party estimate and call it platform data.
Keep a simple log (a spreadsheet is enough) and re-run the same set monthly. Tarek Abou Rjeily's guide on this site, 'How to measure AI visibility without inventing an AI rank', sets out the full method, including how to log valid runs and attribute leads. You can run this baseline yourself; it does not need a consultant.
How to measure AI visibility (guide)
Answers are generated, not looked up, and they shift with wording, mode, location and which sources are retrieved that day. That is why a method uses several runs of identical prompts and looks at patterns, not at any single response.
Include a few name searches to see how you are described and whether the facts are right, but do not rely on them. The harder and more valuable test is whether you appear for category questions asked by someone who has never heard of you.
Several tools sell one, but there is no universal AI rank. A score is only meaningful if it names the prompt set, assistants, countries, languages and dates behind it and keeps the raw mentions and citations. Be cautious of a single number with no log behind it.
Yes, if any of your buyers search in Arabic. Run the same questions in Arabic and in English, and in Arabizi if that is how your customers write, because the sources retrieved, and so the businesses named, can differ between languages.
Third-party facts can change. These primary sources were checked on the page's update date.
Send the question on WhatsApp and I'll tell you honestly whether it needs building, buying, or fixing by hand.
Ask on WhatsApp →