What questions should you run?
Use the questions buyers ask when money is on the table, not the ones you wish they asked. Three types cover most businesses: recommendation questions ("best dental clinic in Jumeirah"), comparison questions ("X clinic or Y clinic for veneers"), and problem questions ("my back hurts after gym, who should I see in Dubai"). Write 15 to 20, keep them exactly as a customer would phrase them, and freeze the list. The same questions get reused every time you measure, or the results stop being comparable.
How do you run the test, step by step?
- Open a fresh chat for every question. History contaminates answers.
- Ask the question exactly as written. Do not mention your brand name.
- Screenshot the full answer before doing anything else.
- Record in a spreadsheet: the question, the platform, every business named, the order they appear in, and what is said about each.
- Repeat the question in Arabic in a new chat, and record it as a separate row.
- Repeat across all four platforms: ChatGPT, Google AI Overviews, Gemini, Perplexity.
At the top of that range, twenty questions, two languages, four platforms works out to 160 answers. Budget three to four hours, and expect the spreadsheet itself to be the hardest part of the exercise.
How do you score what you recorded?
Count the answers that name your brand and divide by total answers: that is your Brand Mention Rate. Then read the mentions themselves. Mark each one positive, neutral or wrong, because an answer calling you "premium but slow to respond" is steering buyers away while technically mentioning you. Note which competitors appear most, and in which language they beat you.
A workable sheet needs six columns: question, language, platform, businesses named in order, what was said about each, and a tone flag. Add a date column so this month's run can sit next to the last one. Build the sheet once, in about half an hour, and every run after that is just filling in rows.
What do you do with the results?
Three findings are common. If you are absent, the fix starts with sources: directories, reviews and structured facts that engines can read, starting with your AI visibility fundamentals. If you are described wrongly, correct the facts everywhere they appear before anything else. If a competitor owns one language, that side of the web is where your next content effort goes.
When is the manual method not enough?
The method is honest work and it is free, which is why we published it in full. Its cost is repetition: 160 answers per month, scored consistently, in two languages, without shortcuts. If that cadence will not survive contact with your calendar, an AI Visibility Audit runs the full test professionally and hands you the scored baseline with a fix list.