TL;DR
- You cannot improve what you never look at. AI visibility needs its own measurement, separate from Google rank tracking.
- Build a prompt set from the real questions customers ask, then test it across ChatGPT, Google AI Mode, Gemini, Perplexity and Copilot.
- For each answer, record four things: are you mentioned, are you cited, is it accurate, and how do you compare to competitors.
- Run it on a fixed cadence and log results in a simple spreadsheet. Automated tools exist once you outgrow the manual version.
- Answers vary by user, region and run, so track trends over weeks, not single snapshots.
More of your buyers now start with a question typed into ChatGPT, Perplexity, Copilot, or Google’s own AI Mode instead of a plain search box. The assistant reads across the web and hands back a short, confident answer, often naming a handful of providers. If your brand is in that answer, you are on the shortlist. If it is not, you may never learn that the conversation happened at all.
That is the uncomfortable part. With Google you can at least see your ranking and your traffic. With AI answers there is no dashboard waiting for you, so most companies have no idea what these tools say about them. The good news is that you can measure it yourself with a simple, repeatable routine. This guide walks through how to set that up.
Why AI visibility needs its own measurement
The old rule still holds: you cannot manage what you do not measure. But AI visibility does not behave like a search ranking, so you cannot just bolt it onto your existing SEO reports. And it is no longer a fringe channel. Google now leads Search with AI-generated answers by default, through AI Overviews and its conversational AI Mode, so even buyers who never open a separate chatbot are reading an assistant’s summary before they reach a list of blue links.
A Google result is relatively stable. Search the same term twice and you see roughly the same page, in roughly the same order, with your position clearly numbered. An AI answer is different in three ways that matter for measurement:
- There is no ranking. You are either named in the answer or you are not, and being named once is not the same as being recommended.
- The output is generated fresh each time, so two people asking the same question can get different brands, different wording and different sources.
- The answer blends many sources into new text, which means it can describe you inaccurately even when nothing on your own site is wrong.
So the thing you are tracking is not a number on a results page. It is presence, accuracy and how you stack up against competitors, measured across several tools and over time.
Step one: build a prompt set
Your prompt set is the list of questions you will ask the AI tools, over and over, each time you measure. Treat it as a fixed test so results stay comparable month to month. Aim for fifteen to thirty prompts to start.
Base them on the real questions customers ask in your category, not the keywords you wish they used. Pull these from your sales calls, your support inbox, and the way prospects actually describe their problem. Cover a few types:
- Category questions, where your brand is not mentioned at all, such as ‘What is the best payroll software for a Dutch SME?’ These show whether you surface unprompted.
- Comparison questions that name you against rivals, such as ‘How does [your brand] compare to [competitor]?’
- Brand questions about you directly, such as ‘What does [your brand] do and who is it for?’ These test accuracy.
- Problem-first questions with no brand or product named, phrased the way a buyer would describe their situation.
Write down your own brand name and your main competitors alongside the list. You will look for all of them in every answer, so decide upfront who counts as a competitor.
Step two: test across the models
Run each prompt through the main assistants your buyers use. In most Dutch markets that means ChatGPT, Perplexity, Copilot, and Google’s AI answers, whether the AI Overview at the top of a normal search or the newer AI Mode. Gemini is worth adding if it is popular with your audience. These tools pull from different sources and behave differently, so results genuinely vary between them, and you want to know where you are weak. Google in particular matters because it reaches people who would never think of themselves as using AI at all.
For every prompt and every model, record four things:
- Mentioned. Does your brand appear in the answer at all? A simple yes or no.
- Cited. Is there a link or a named source pointing to you? Being quoted with a citation is stronger than a passing mention.
- Accurate. Is what it says about you correct and current? Note any wrong claims about your products, pricing or focus.
- Share of voice. Which competitors show up, and how do you compare? Counting how often each brand appears across the whole prompt set gives you a rough share-of-voice figure.
The goal is not a perfect score on any single answer. It is a clear picture of where you show up, where you are missing, and what the tools get wrong about you.
Step three: log it on a cadence
A one-off check tells you almost nothing, because any single answer could be a fluke. The value comes from repeating the same test and watching the trend.
Monthly works well for most businesses. Keep it in a simple spreadsheet with one row per prompt-and-model combination and columns for the four measures above, plus the date and a note on what the answer actually said. Over a few cycles you will see whether your presence is climbing, holding or slipping, and whether a piece of work you published moved the needle.
Doing this by hand is fine at fifteen prompts across four tools. If you scale up, automated monitoring tools exist that run your prompt set on a schedule and chart mentions, citations and share of voice for you. Start manual so you understand what you are measuring, then automate once the routine proves its worth.
Step four: act on what you find
Measurement only pays off if it changes what you do. Three patterns in your log point to three clear actions:
- Inaccuracies: correct them at the source. If a tool describes you wrongly, find where that idea comes from, usually an outdated page, a stale directory listing or a third-party profile, and fix the underlying information so future answers have better material to draw on.
- Content gaps: fill them. If you are absent from a category question you should own, it usually means there is no clear, well-structured content on the web that answers it and points to you. Publish it, plainly and factually.
- Weak third-party support: strengthen it. AI answers lean heavily on sources they trust, such as review sites, industry directories, reputable articles and comparison pages. Getting accurate, up-to-date entries in those places often does more than another page on your own site.
A few honest caveats
Keep your expectations realistic so you read the data correctly. AI answers are non-deterministic: the same prompt can return different brands on different days. Results also vary by the person asking, by their location and language, and by which version of a model is live that week.
This is why you track trends, not single results. One month where you slipped on one prompt is noise. A steady three-month decline across several prompts and tools is a signal worth acting on. Measure consistently, give it time, and let the pattern, not any one answer, guide your decisions.
Where to start this week
You do not need a tool or a budget to begin. Open a spreadsheet, write down ten questions your customers actually ask, add your brand and two competitors, and run them through ChatGPT and Google’s AI Mode. Note who gets mentioned and whether anything said about you is wrong. That first hour usually surfaces at least one inaccuracy and one gap you can fix straight away, and you will have the beginnings of a baseline to measure against next month.
Frequently asked questions
How often should I check how my brand appears in AI answers?
Monthly is a sensible default for most businesses. It is frequent enough to catch real movement and to see whether new content had an effect, without turning into a full-time job. Check more often only around a launch or a rebrand, when things are actually changing.
Do I need a paid tool to monitor AI visibility?
No. You can start with nothing more than a spreadsheet and the free versions of the assistants. Run a fixed set of prompts by hand and log the results. Paid monitoring tools become worthwhile once your prompt set grows and running it manually across several models each month takes too long.
An AI tool says something wrong about my company. How do I fix it?
You cannot edit the answer directly, but you can improve what the model draws on. Trace the claim to its likely source, often an outdated page of your own, a stale directory listing or a third-party profile, and correct the information there. Over time, as the tools re-read the web, accurate sources tend to push out the wrong ones.