Method

    How to measure AI visibility without fooling yourself

    AI answers change between sessions, so one screenshot proves little. A simple, repeatable method for tracking whether ChatGPT, Perplexity and Gemini name you.

    By NoorRank · Updated

    In short

    Measure AI visibility with a fixed set of questions, written in advance and asked the same way on each platform, in clean logged-out sessions, on a regular schedule. Log every answer with its date. Track three things: whether you are named, whether you are cited with a link, and whether what is said about you is accurate.

    Why one screenshot isn't evidence

    Ask ChatGPT the same question twice and you can get two different answers. Location, account history, the date and the exact wording all change the result. So a single screenshot of your brand in an answer, or missing from one, tells you very little. What tells you something is the same test, repeated.

    Our own experiment showed why. A new entity was recognised by Perplexity on day 17, dropped on day 19 and recovered on day 20, with nothing on the site changing. A weekly check would have recorded a clean success and missed the drop entirely. Read the Miracle Berry case study for the full log.

    The method, step by step

    1. Write the question set before you start. 20 to 40 questions a buyer would ask, in their words. Most should not include your brand name, because those are the questions where you have to earn a mention.
    2. Split it by intent. Category questions (‘best fit-out companies in Dubai’), problem questions (‘how do I reduce my clinic's no-shows’), comparison questions, and a few brand questions (‘what is [your company]?’).
    3. Cover each language you sell in. Arabic and English answers can differ, so ask both versions.
    4. Use clean sessions. Logged out or in a fresh private window, with a consistent location where possible.
    5. Ask the same way on each platform. ChatGPT, Perplexity, Gemini and Google (for AI Overviews), at minimum.
    6. Log everything. Date, platform, question, whether you were named, your position in any list, whether you were linked, which sources were cited, and anything the answer got wrong about you.
    7. Repeat on a schedule. Monthly is a reasonable minimum. During active changes, weekly or even daily checks catch movement a monthly check would miss.

    What to track

    MeasureWhat it tells you
    Mention rateShare of non-brand questions where you are named at all
    Citation rateShare of answers that link to your site as a source
    PositionWhere you appear when the answer is a list
    AccuracyWhether the facts stated about you are correct
    Competitors namedWho appears instead of you, and on which questions
    Sources citedWhich sites the AI relies on in your category, and therefore where you need to be mentioned

    Data you already have

    • Analytics referrals. Visits from chatgpt.com, perplexity.ai and gemini.google.com show up as referral traffic. ChatGPT also adds utm_source=chatgpt.com to many links it shows.
    • Search Console. Clicks and impressions from AI Overviews and AI Mode are included in the Performance report, though not shown separately.
    • Bing Webmaster Tools. Shows how your site appears in Microsoft's search features, which also feed Copilot.

    Mistakes that make the numbers lie

    • Changing the questions between rounds. Then you are measuring the questions, not your visibility.
    • Only testing brand queries. Being named when someone asks for you by name is the easy part.
    • Testing while logged in. Personal history can make an assistant favour brands you have searched for.
    • Counting a mention as a win when the facts are wrong. An inaccurate description can cost you more than no mention.
    • Reporting only the good rounds. If a platform drops you, log it. That is the information you need to act on.

    Questions people ask

    Are there tools that track AI visibility automatically?

    Yes, several SEO platforms now run question sets across AI assistants on a schedule. They save time, but the method is the same: fixed questions, consistent conditions, logged results. Check how a tool runs its queries before trusting its numbers.

    How many questions do I need?

    Enough to cover your main categories, services and languages. 20 to 40 is a practical range for most businesses. Fewer than ten makes the results too noisy to read.

    Can I see AI Overview traffic separately in Search Console?

    Not currently. AI Overview and AI Mode clicks and impressions are included in the overall Performance report without their own filter.

    See what AI says about your business

    The free Ghost Audit shows how ChatGPT, Perplexity, Gemini and Google's AI Overviews describe you today, and who they recommend instead.