The AI Visibility Check measures whether your company is mentioned in the responses from ChatGPT, Perplexity, Gemini, and other AI chatbots. And what you can do to change that.
Your customers have long since stopped searching only on Google. You ask an AI—and instead of getting a list of ten results, you get an answer with two or three names. Anyone not mentioned there is simply not considered when the decision is made.
The problem is: you don’t even notice it. There is no spike in the statistics, no report, no warning. Customers who have never found you don’t show up in any analysis.
The AI Visibility Check shows you what’s happening right now. About forty questions, three systems, measured multiple times over several days. In the end, you’ll receive a report with clear figures and a prioritized list of actions to take.
For twenty years, online visibility was a matter of placement. You were in third place or eleventh place, and anyone who wanted to could keep scrolling. The idea was to show a lot of options, and let the user choose.
Language models work differently. They don’t present a choice; they give an answer. When asked who can help with a specific problem, ChatGPT gives two or three names—and that’s that. There is no second page to turn to.
The questions have gotten longer
Someone typed “Treuhänder Zug” into Google. An AI is asked: “I’m self-employed in Zug, I do everything myself, and I’m having trouble figuring out the value-added tax.” “Who can help me, and how much will it cost, roughly?” These are entirely different questions—and they are answered by entirely different content.
The answer is being rebuilt
A language model doesn’t link to your website; it formulates its own response based on what it knows and finds about you. You have no control over how you are portrayed—but you can have a significant influence on the foundation upon which that portrayal is based.
And you can’t look it up yourself
Google offers tools that allow any business to track its ranking. That doesn’t exist for language models. You can test yourselves—but a single test tells you almost nothing, because the same question will yield a different answer tomorrow. That is exactly why accurate measurement is necessary.
The good news is: This can be measured. And many of these factors can be influenced, often with a manageable amount of effort. The first step is always the same—knowing where you stand.
The Check is a structured assessment of your visibility in language models. I ask the systems about forty questions that your customers would actually ask, and analyze how often, where, and in what context your company is mentioned.
The result isn’t a dashboard with flashing numbers, but a report that even your executive team can understand: what was measured, what the results were, how to interpret them, and what specific actions to take based on them.
Measured multiple times: Each question is asked several times, spread out over several days. A single query is a coincidence, not a measurement.
Tailored to your industry: The questions will come up as we talk. Not a standard catalog that looks the same for every industry.
With recommendations for action: You won’t just get numbers—you’ll get a prioritized list: what to do first, what to do later, and what not to do at all.
MODULE 1: About forty questions in four categories
The list of questions will be determined together with you during the preliminary discussion. It covers four types of questions:
MODULE 2: Three Systems
Measurements are taken in ChatGPT, Perplexity, and Gemini. These three cover the majority of actual usage and are based on different sources. If you measure only one of these, you’ll get a distorted picture—and, ironically, the system where you perform well is rarely the one your customers use.
MODULE 3: Multiple measurements over several days
That’s where in-depth analyses differ from quick tools. Language models don’t always respond the same way. The same question may include your name today but not tomorrow. If you measure just once, you’re measuring by chance.
That is why each question is asked multiple times and spread out over several days. Only through repetition can you establish a reliable mention rate—that is, a figure you can use for comparison in three months.
What This Method Cannot Do
Visibility in language models can be influenced, but not controlled. No one can guarantee that you’ll be mentioned in a specific place after taking certain steps—and anyone who promises that should make you suspicious. What I can provide you with is a reliable baseline assessment, a well-founded evaluation, and measures that are effective based on current knowledge. What I can’t give you are guarantees.
The report is written in such a way that you can share it internally without having to translate it. That’s on purpose: In most cases, the person who requests the check has to convince someone else of it.
Most companies don’t believe there’s a problem until they see it. So here’s a simple offer for you: Send me a question that your customers might ask. I’ll measure it and send you the results. Free of charge, with no obligation, and no sales pitch afterward.
Here’s how it works:
The comprehensive assessment for your company: about forty questions, three systems, measured multiple times over several days. In addition, the report, the raw data, and a thirty-minute discussion.
For each competitor you include in the comparison, the same set of questions will be evaluated. The result is a direct comparison: Who is mentioned, how often, in what context, and based on which sources. My recommendation: two to three competitors. One example on its own is a random comparison; more than five make the report confusing. Competitors are not included in any package and cost the same everywhere.
A comprehensive assessment followed by work on the recommended actions: a half-day workshop with your team, a detailed action plan outlining responsibilities and the order of tasks, and a follow-up meeting after about sixty days. This makes sense when multiple people within a company need to work together on a project—typically marketing, the website team, and a business unit. A report alone rarely brings about change. A workshop where everyone leaves with their own assignment—sure.
The same measurement, repeated quarterly, with a delta report: what has changed, what is attributable to your actions, and what is due to changes in the models. In addition, three specific recommendations for the next quarter in each case. Visibility is not a state, but a process. A quick check will show you where you stand today. Only by repeating it can you tell whether what you're doing is actually working.
The most common reaction to a visibility report is concern, followed by the question: “What now?” That is why the report ends not with numbers, but with a list. The measures vary by industry, but there are patterns that almost always emerge.
Your team can implement much of this on its own. The report is written in such a way that this is possible. If you’d like support, there’s the “Check Plus Implementation” package.
It’s worth checking if:
The check isn’t worth it if:
If you’re unsure, take advantage of the free one-time measurement. After that, you’ll know more than you would from any consultation.
There are now numerous providers that automatically measure visibility in language models, often for just a few francs a month. Some of them are well done. Nevertheless, they solve a different problem than the check. Three differences are crucial.
My name is Sophie Hundertmark. I conduct research and teach at the Institute for Financial Services in Zug at the Lucerne University of Applied Sciences and Arts, and I also advise companies on artificial intelligence. I completed my doctoral dissertation in 2026 at the University of Freiburg.
AI visibility is one of my main areas of research. Among other things, I have conducted an industry analysis of the Swiss banking market and am working on a framework that outlines how corporate content must be structured so that language models can reproduce it accurately.
The difference from an agency lies in the origin of the method. I don’t measure simply because I bought a tool, but because I’m interested in how these systems generate answers. This is particularly evident in the report’s analysis section—where the numbers are turned into recommendations.
STEP 1: Request
Let me know what this is about: the company, the industry, and what you want to know. I’ll get back to you within one business day.
STEP 2: Preliminary consultation, about thirty minutes
We’ll work together to determine the list of questions. You know your customers, and I know the systems—that combination yields the questions that really get to the heart of the matter. We’ll decide here whether to include your competitors in the comparison.
STEP 3: Measurement
The queries are spread out over several days across all three systems. We don’t need anything from you at this stage.
STEP 4: Analysis and Report
I analyze the data, organize it, and write the report.
STEP 5: Meeting, 30 minutes
We’ll go over the results together and figure out what should happen first.
DELIVERY TIME: Two to three weeks
It usually takes two to three weeks from the initial consultation to the final report, depending on our workload. The bulk of the process is the measurement itself, which cannot be meaningfully expedited—it is precisely the fact that it is spread out over several days that makes the result reliable.
If you’re in a hurry, let me know. Sometimes it goes faster, and I’ll be honest with you when you make the request about what’s realistic.
No. Visibility in language models can be influenced, but not controlled—the systems change too quickly and in ways that are too difficult to understand. Anyone who gives you a guarantee is selling you something they can’t deliver. What I promise: a thorough baseline assessment, a well-reasoned evaluation, and measures that are effective based on current knowledge.
Usually weeks to months, not days. Language models draw partly on continuously updated sources and partly on knowledge acquired during training. Changes to your website can therefore take effect relatively quickly, while a company’s overall image shifts more slowly. That is why it makes sense to repeat the measurement after one quarter.
No, even if there are some overlaps. Search engine optimization aims to achieve a high ranking in search results. The question here is whether your company is even mentioned in a written response—and how. Different questions, different metrics, and in some cases, different measures. Good search engine optimization never hurts, but it’s not enough.
Then you’ll have a reliable baseline measurement, a strong case to present to your management, and you’ll know which sources work for you. That’s a useful result, not a wasted effort. I’ll tell you that clearly, instead of coming up with reasons to take action.
ChatGPT is the best known, but the systems draw on different sources and provide different answers. It often happens that a company is clearly visible in one system but not at all in another. Measurements are therefore taken in ChatGPT, Perplexity, and Gemini—together, they account for the majority of actual usage. Additional systems can be added upon request.
The two of us, during the preliminary discussion. I bring the structure and experience from other industries; you know how your customers speak. The final catalog is included in the report to ensure that the measurement remains reproducible.
Absolutely. Questions that are actually asked of you—whether in sales, customer service, or during consultations—are particularly valuable. They’re usually much better than anything you come up with at your desk.
Two to three. One example on its own is a random comparison; more than five make the report confusing. Experience shows that the most insightful competitors aren’t necessarily the biggest ones, but rather those whose content is most similar to yours.
No. The measurement takes place exclusively within the AI systems. No one is contacted, and no trace is left at the companies surveyed.
Just me. Results will not be disclosed or used in presentations or publications unless you expressly consent to it in writing. I’d be happy to sign a confidentiality agreement.
That depends largely on what the report finds. A significant portion of the measures relates to content and can be handled internally—in which case, the main cost is time. If you would like support, the “Check plus Implementation” package is available for CHF 6,000. I provide separate quotes for larger projects—always in advance and always in writing.
Yes. The questionnaire can be tailored to a specific brand, service, location, or business unit. This is especially helpful if you’ve been reassigned to a new area.
It depends on whether your customers do their research before making a purchase. If so, it’s worth it regardless of size—small providers actually benefit more than average because they can catch up more easily in language models than in traditional search results. If you rely solely on referrals, I’ll be honest with you.
Every quarter, if you’re actively working on your visibility. Every six months, if you just want to observe. It rarely makes sense to do this more frequently than quarterly, because changes take time.
Two to three weeks after the initial consultation, depending on our workload. If it’s urgent, ask me—sometimes it can be done faster.
Then tell me. I don’t deal in fine print: If the report doesn’t do the job, we’ll find a solution. I’d rather have that than a dissatisfied customer who doesn’t say anything and never comes back.
You can continue to speculate on how your company is mentioned in AI responses. Or you can just measure it accurately once and you’ll know.
If you’re still unsure, start with the free individual assessment. Just one question is all it takes, and within a few days you’ll see how people are talking about you. If you want to get started right away, send me a message—I’ll get back to you within one business day, and we’ll spend ten minutes figuring out whether and to what extent the check makes sense for you.
