The short answer
A visibility score is a statistic computed over a sample of questions. The sample has to be sized so that the effect of one month's work is larger than the measurement's own noise. Foliora Personal measures twenty-five buyer questions and ships three approved changes a month, and at that ratio a single gained citation moves the score by about four points — large enough to read as a result. Widen the panel to five hundred and the same citation moves it by two tenths of a point, which is indistinguishable from the run-to-run variance of the score itself.
So the number is not a ration of how much you are allowed to look at. Foliora reads your entire catalog up to your plan's managed page count and will propose work anywhere in it. The panel is the instrument, not the scope.
The questions are generated from your catalog, not typed in by you
Most tools in this category hand you an empty box and ask which prompts you would like to track. That sounds like control and behaves like a tax. It puts the hardest research problem in the workflow — what do buyers in this category actually ask, and which of those questions is this business qualified to answer — onto the customer, on day one, before they know anything.
Foliora derives the panel from the catalog itself: your products, collections, services, pages, and the buying decisions they imply. The output is a set of questions your own inventory gives you standing to answer, which is a stricter filter than either your intuition or a keyword tool. Nobody picks from a menu of generic prompts, because a generic prompt panel measures a generic category rather than your business.
You can see the panel, and you should read it. If a question in it is wrong for your business, that is a finding about how your catalog describes itself, and it is usually worth more than the question was.
Two different things reduce noise, and only one of them is prompt count
Assistant answers are non-deterministic. Ask the same question twice and the cited sources can differ. There are two separate ways to deal with that and they are routinely conflated.
Asking the same question repeatedly reduces the uncertainty on that question. Asking more different questions reduces the uncertainty on the aggregate while telling you nothing more about any single one. A product advertising several hundred prompts sampled once each has bought breadth with the budget that repetition needed, and the per-question numbers underneath its confident average are single draws.
Both cost the same thing, which is sampling runs. Choosing a smaller panel is what makes it affordable to ask the same questions consistently and keep every full response, rather than storing a score and discarding the evidence that produced it.
- Repetition per question buys reliability on that question
- More questions buys precision on the average and nothing else
- Both are paid for out of the same budget, so a panel of hundreds is a choice against repetition
Size the panel to the size of the intervention
This is the part that gets skipped. A measurement is only useful if it can detect the thing you are doing. Foliora Personal publishes three approved changes a month; those changes will plausibly touch a handful of questions, not hundreds. The panel should therefore be small enough that a handful of questions is a visible fraction of it.
At twenty-five questions, one citation gained is four points. At fifty it is two. At one hundred it is one. At five hundred it is two tenths of a point, at which stage you have built an instrument that cannot see your own work — and, worse, one whose month-to-month wobble is larger than any change you could have caused. That is not a more rigorous measurement. It is a smoother chart.
The corollary is honest and worth stating: if you were shipping two hundred changes a month, twenty-five questions would be the wrong panel and a much larger one would be correct. Panel size should track the rate of change, in both directions.
The panel is fixed, because a moving panel cannot produce a trend
Changing which questions are measured between two months makes the difference between those months uninterpretable. You have not observed a change in your visibility; you have changed the instrument and then read it. This is the most common way an AI visibility number becomes decorative, and it usually happens innocently, one added prompt at a time.
So the panel holds. When it does need to change — a new product line, a category that did not exist last quarter — the change is recorded and dated, and the trend restarts from there rather than pretending continuity across the seam. Every stored observation carries the engine, the question, the date, the full response, and the citation URLs, so any number in any month can be recomputed from the evidence rather than taken on trust.
Then why can you buy more questions?
Because scope and allowance are different things, and it would be dishonest to pretend the add-on does not exist. A business selling one category of product needs one panel. A business with four genuinely distinct buying decisions across four ranges needs more questions, not because it earned extra usage but because there is more than one category being measured.
The test for whether you need more is not whether you want a bigger number. It is whether there is a buying decision your current panel does not cover at all. If the answer is no, adding questions dilutes the resolution of the score you already have, and we would rather say so than sell the pack.
What this means when you compare us to a monitoring tool
Several platforms in this category track five to twenty times as many prompts as Foliora measures, and some of them do it daily where we sample monthly on the entry plan. Those are real advantages for the job of knowing where you stand, and our comparison pages say so plainly.
What a prompt count cannot tell you is whether anything changed as a result. That is why the number we put on the pricing page and the plan card is approved changes published and verified, and why the panel sits below it in the comparison table rather than at the top. A monitoring tool can match a question count at a lower price. It cannot match a published, verified change, and the published change is the thing that moves the answer.
Common questions
Is 25 buyer questions a usage limit?
No. It is the size of the measured panel. Foliora reads every page in your catalog up to your plan's managed page count — 250 on Personal, 1,500 on Startup, 5,000 on Agency — and proposes work anywhere in it, whether or not that page maps to a tracked question. The panel exists to detect movement, not to cap what gets examined.
Can I choose the questions myself?
You can read the panel and challenge anything in it, and a question that is wrong for your business is a signal worth acting on. What Foliora does not do is start from an empty box, because deciding what buyers ask and which of those questions your catalog can credibly answer is the research the service is supposed to perform, not a form for you to fill in before you have any data.
Wouldn't tracking more prompts give a more accurate score?
A more precise average, and a less sensitive one. Precision on the aggregate and sensitivity to your own work pull in opposite directions once the panel is large relative to the number of changes you ship. With three changes a month, a five-hundred-prompt panel produces a stable number that cannot register what you did. Separately, more distinct prompts is not the same lever as asking the same prompt repeatedly, which is what actually reduces the uncertainty on any individual question.
How much does one citation move the score?
About four points on a twenty-five-question panel, two on fifty, one on a hundred. That is the arithmetic of the panel size and nothing more sophisticated is going on. It is stated here so you can check any movement we report against how many questions changed hands, rather than accepting a score with an undisclosed denominator.
What happens if the panel needs to change?
It gets changed, dated, and recorded, and the trend restarts from that point rather than being compared across the seam as though nothing happened. Silently amending the panel and continuing the same line is the most common way these numbers stop meaning anything.
How can I verify a score you report?
Every observation is stored with the engine, the question, the market, the date, the full response text, and the citation URLs. A score is recomputable from those records months later, and the individual responses are readable. If a vendor cannot show you the prompt, the date, the raw answer, and the citations behind a number, the number is decorative — that applies to us on exactly the same terms.