Last updated: 13 April 2026
What my AI Visibility Monitor looks like
Before I discuss the results, here is the setup. I use a spreadsheet called the AI Visibility Monitor. It contains 30 prompts divided into five clusters: direct brand recognition, category discovery, comparison, problem-solving and thought leadership. I run each prompt three times per platform—ChatGPT, Gemini and Perplexity—to filter out chance variation.
This is not an automated measurement. I enter the prompts manually and save the results. It is a periodic check, not something I do every day. The full monitor covers 270 planned data points (30 prompts × 3 runs × 3 platforms), which I am collecting over time. In this first round, 165 runs were completed. ChatGPT is fully measured (all 26 measured prompts, 78 runs), Gemini is on schedule (22 of 26 prompts, 66 runs), and Perplexity is still limited (7 of 26 prompts, 21 runs, mostly direct brand prompts).
That means the Perplexity results are not directly comparable with ChatGPT and Gemini at this stage. I discuss them separately.
Expectation versus reality
I expected low scores overall, with perhaps a few outliers on prompts that explicitly asked about me by name. That would make sense given that my site had very little content when I measured it. What I found was more nuanced. Across all 78 measured ChatGPT runs, my site or name appeared 22 times—a mention rate of 28.2%. Gemini scored 4.5% (3 of 66 runs). What interested me more was the pattern behind those averages.
The surprise is in cluster A
The first cluster—direct brand recognition—tested prompts such as “Do you know the website handsongeo.com?”, “What does Hands on GEO write about AI visibility?” and “Is Hands on GEO a reliable source about GEO?”, plus their English equivalents.
Across the 18 measured ChatGPT runs, my site scored 15 out of 18—an 83% recognition rate. All three misses came from the same prompt: “What is Hands on GEO?”, which ChatGPT interpreted as hands-on work with geographic data. I will return to that separately.
Gemini recognised my site in 3 of 18 runs in this cluster (17%). That sounds low, but the detail is interesting: the only prompt where Gemini consistently mentioned me—all three runs—was “Do you know the website handsongeo.com?” For every other direct question about my content or reliability, nothing. Gemini knows the domain, but apparently has little context about what is on it.
For Perplexity, only the Dutch branded prompts have been measured so far (4 prompts, 12 runs). Recognition is notably strong: 10 of 12 runs were positive—83%, comparable to ChatGPT. The English Perplexity prompts have not yet been run.
The signal that stood out most
Besides direct brand recognition, some prompts mention my name without being asked. Those mentions came only from ChatGPT. Gemini and Perplexity scored zero on the measured questions outside the branded prompts.
When asked, “My website does not appear in AI answers. What can I do?”—a typical problem-focused question from a B2B marketer struggling with this issue—ChatGPT casually brought up handsongeo.com as a practical example in all three runs. It was not answering a direct question about my website, but advising someone with a concrete problem. It is worth mentioning that each chat was a fresh conversation, with no context, and I was not logged in.
My name also appeared occasionally in comparison prompts. For “Compare the best GEO sources for marketers” (Dutch), ChatGPT named my site in 1 of 3 runs as a “specialist GEO source, Advanced level, still niche/limited reach”. For the English version, it scored 2 out of 3, once describing it as “Best niche resource for NL/B2B”. My URL also appeared in 1 of 3 runs for “Who are GEO experts in the Netherlands?”, although the wording there was somewhat ambiguous.
I consider that a more relevant signal than brand-recognition scores. If an AI tool spontaneously names me as an example in response to a question with clear intent, my site must be associated with that specific problem in its training data. Apparently, I am relevant in that context.
A surprising observation about the Dutch brand prompt
One result I did not expect: the Dutch prompt “What is Hands on GEO?” scores zero on ChatGPT and Gemini. Not because they do not know my site, but because they interpret the phrase incorrectly. ChatGPT explains “Hands on GEO” as “practical work with geographic data”—GEO as in geodata, not Generative Engine Optimisation.
The English version—“What is Hands on GEO?”—scores three out of three on ChatGPT. Perplexity does recognise the Dutch version, though not consistently (2 out of 3 runs).
This is an odd asymmetry I want to investigate further. I do not yet know whether the brand needs to be clearer in Dutch or whether GEO terminology in Dutch is still too ambiguous.
What I take away from this
This is measurement round 1, and it is still partly incomplete. One measurement has limited value. What matters is the pattern over time: does my visibility in the discovery clusters grow as I publish more content? Does Gemini recognise me on more than one prompt? Will Perplexity show the same picture once the other prompts have been measured?
Those are the questions I want to answer in the next round.
What I can already say is that the measurement is worthwhile. Not because all the scores are high—they are not—but because the pattern provides information. Something is already there, and it is more specific than I expected. The distinction between brand recognition, question-driven discovery and generic category visibility is exactly what a GEO strategy should address.
Frequently asked questions
How does this AI Visibility Monitor work?
I use a spreadsheet with 30 prompts across five clusters (brand name, category, comparison, problem-solving and thought leadership). I enter each prompt manually and run it three times on ChatGPT, Gemini and Perplexity. Each prompt gets a score from 0 to 3 per platform: how many of the three runs mentioned my site or name.
Do you do this every day?
No. This is a periodic manual check, not automated daily tracking (I would like that, but I have not found the right script yet). I run the prompts over time to build a baseline. Round 1 ran from 7 to 12 April 2026. At the moment, 165 of the 270 planned data points are filled in.
Why run each prompt three times?
AI systems regularly give different answers to identical questions. I consider one run unreliable. Three runs give an initial indication of consistency without making the process unworkable.
What is the difference between the five clusters?
Cluster A tests direct recognition of my brand. Cluster B tests whether I appear in generic category questions about GEO. Cluster C looks at comparison prompts. Cluster D simulates problem-focused searches. Cluster E focuses on authority and thought leadership.
Why does Gemini score so much lower than ChatGPT?
Gemini recognises my domain when asked directly (3 out of 3 for “Do you know handsongeo.com?”), but almost nowhere else. Based on one measurement round, I cannot say whether this relates to training data, how Gemini classifies entities or something else. The four English discovery prompts that have not yet been measured may qualify the picture.
What does it mean when an AI tool “knows” your site?
It means information about your site is represented in the training model—enough for the model to say something sensible about it. It says nothing about how recent or accurate that information is, or whether the site is considered authoritative. It is a starting point, not an endpoint.
Will you keep measuring?
Yes. The aim is to compare each round with the previous one—not to prove that my approach works, but to see whether anything changes. If it does not, I want to know that too.
Sources:
– Hands on GEO AI Visibility Monitor—measurement round 1, 7–12 April 2026 (own data, 165 of 270 planned data points recorded)
– Measurements on ChatGPT (GPT-4o), Gemini and Perplexity—manual, periodic runs
Also read: Everything a B2B marketer should know about GEO in 2026 · What is AI Search? · What is Entity Optimization?
This translation was made by AI.