Direct answer
Your customers don't all ask the same question. Some type your category name ("best note-taking app"), others describe a much broader need ("I'm overwhelmed, how do I get organized?"). The broader the question gets, the harder it is for AI to think of you: it first cites general solutions, adjacent players, sometimes nothing from your field at all. So there's a breadth threshold beyond which your brand disappears from the answer — and that threshold isn't the same for every brand. A well-established brand holds up on fairly broad questions; a more low-key brand drops off quickly. Knowing where you drop off on the "specific → broad" scale, and who AI cites before reaching you (or in your place), means measuring your presence on the most upstream terrain — the one you didn't think to watch.
The problem
When you test your AI visibility, you generally ask the "clean" question: the one that names the category. And often the result is reassuring — on that question, the brand shows up.
But that's only one door among several. Many customers arrive through a much fuzzier question, where neither your brand nor even your category is named. And there, AI's behavior changes: it often stays general, suggests methods rather than products, cites players from nearby. If you only measure the specific question, you never see that moment where you disappear — even though it corresponds to a real share of your customers' conversations.
The idea to grasp
The breadth of the question is a slider, not a switch. Sliding from specific to broad, you observe a drop-off:
- Specific question ("which all-in-one notes app") → AI is on marked-out ground, your category is explicit, your brand has every chance of showing up.
- Medium question ("how do I bring together my notes and my projects") → the category is still there, but the competition widens.
- Broad question ("how can I be better organized at work") → AI heads toward the general: methods, habits, tools of all kinds. Your brand, or even your category, may no longer appear at all.
Two things get measured in this slide:
- The drop-off threshold. At what level of breadth your brand stops being cited. According to our tests, a very well-known productivity app held up to a medium question and then dropped sharply on the broadest question — going from near-systematic presence to almost nothing. The drop-off isn't gradual: there's a step.
- The substitution landscape. Who AI cites in your place once you've dropped off. On broad questions, these are often generalist players, or adjacent brands you didn't consider competitors. According to our tests again, on a very broad organization question, AI cited communication and calendar tools — not the notes app that nonetheless reigned on the specific question. The vacant ground is rarely empty: someone occupies it.
And a third factor runs through it all: memory versus web search (see the fiche "What AI knows from memory, and what it goes and searches for"). On broad questions, AI relies much more on what it searches for live — and therefore on the type of pages it finds. According to our tests, a brand could be entirely absent from memory on a broad need, and appear only a little when AI ran a web search. So the drop-off threshold isn't the same depending on whether AI answers from memory or goes and searches.
What you hear everywhere
"On my question, I come out first, so I'm well positioned." On that question, yes. But a broader wording can make you disappear entirely. Testing a single question means measuring a single door out of several.
"If I drop off on vague questions, that's normal, they don't concern me." Maybe. But those are often the ones asked by customers who don't yet know you — precisely the ones you're trying to reach. The drop-off isn't trivial: it tells you how far AI spontaneously associates you with a need.
"Whoever replaces me, it doesn't matter — what counts is me." On the contrary: knowing who AI cites in your place on broad questions reveals adjacent competitors you weren't watching, and the type of answer that takes your spot. It's information about the terrain, not just about you.
Our stance: facts only. We don't decide whether a broad question "should" concern you. We measure, question by question, whether AI cites you — and from what point it stops. You read the drop-off yourself.
Our approach: measure several breadths, and read the drop-off
From here on, the register changes: we describe the instrument.
On the "indirect search" axis of mAIr Signal, the principle is simple: you provide your questions, from specific to broad, and mAIr measures each one separately, without merging them. For each question, and for each AI:
- Your presence (cited or not, at what rank), distinguishing memory from web search.
- The players cited in your place when you drop off — declared and discovered.
- The sources AI consults during web search: what types of pages feed it on these broad questions. This is the most concrete spot in the landscape.
mAIr doesn't choose your questions or calibrate the gradient for you: you're the one who lays out the range (a help block guides you from specific to broad). mAIr returns the facts, question by question; the drop-off threshold, you read it on your own results.