Run a simple test. Take a recommendation question from your industry — "what are the best tools for X?" — and type the exact same sentence into three or four different assistants. Compare the lists.
They almost never match. Some brands show up everywhere. Others only in one. And some names appear where you least expected and vanish where you assumed they were locked in. This experiment, which anyone can run in ten minutes, is one of the most revealing things about how AI visibility actually works.
How to Run It Yourself
You don't need a tool. Pick 3 to 5 questions a customer would ask an AI before buying in your market. Run each one in every assistant you can access. Note which brands appear, in what order, and with what tone. Repeat a few days later, because results shift over time.
The goal isn't to root for your brand. It's to see the pattern. And the pattern, once you collect the notes, tells a story.
Why the Lists Diverge So Much
Four forces sit behind the divergence, and understanding each helps you read the result.
Different training data. Each model was trained on a distinct set of texts, at distinct times. The brands that became "strong" in each one's memory aren't the same.
Live search or not. A model that retrieves in real time pulls what's on the web now; one that answers from memory brings what existed up to its cutoff. That alone produces different lists for the same question.
Caution calibration. Some assistants name freely; others hold back, offer criteria, and cite less. The same knowledge yields answers more or less populated with brands depending on this tuning.
Randomness. Language models have a degree of built-in variation. The same question, asked twice, can return slightly different answers. That's why repetition matters: you want the pattern, not the luck of one run.
What the Pattern Tells You About Your Brand
When you organize the notes, three diagnoses usually emerge.
If your brand appears in every model, across several questions, you have a robust, well-distributed presence — the kind that comes from real, broad authority. Good news: it's hard to knock down.
If it appears in one or two models and vanishes in the others, you have an uneven presence. You're probably strong in a specific source one model values and absent from the ones the others use. It's a fragile presence, dependent on a few footholds.
If it appears in almost none, even on questions where it should be obvious, you have a baseline visibility problem — and no amount of fine-tuning will fix it before you build presence where the AI looks.
The Mistake of Trusting a Single Test
The risk with this home experiment is stopping too early. You run the question once in your favorite assistant, see your brand there, and relax. But that was one run, in one model, on one day. Randomness may have favored you. A competitor may dominate the other three models you didn't check.
The reverse happens too, and it's crueler: you run it once, don't appear, and conclude you're invisible — when, in half the runs, you would have shown up. A sample of one can't tell absence from bad luck.
That's why serious AI-presence measurement is about volume and repetition, not one clever check. You want many questions, many models, several runs over time — and only then does the pattern emerge with confidence.
From Manual Test to Continuous Measurement
The manual experiment is great for building awareness. It pulls AI presence out of the abstract and shows, in black and white, that your brand has different realities in each assistant. But it doesn't scale, and the results change too fast to keep redoing by hand.
That's the bridge Genoma builds: continuously running the questions that matter in your industry, across every model, and turning the divergence you saw by hand into stable tracking — where you appear, where you vanish, who shows up in your place, and how it shifts week to week. Start with the manual test today. It will already show you that you need to look at more than one model.