One in Seven Tracked Prompts Can Never Show You a Win

The Dead Slot

When the answer names vendors, the tracked brand shows up 32.5% of the time. When it names none, that falls to 4.3%. The gap is 7.6×.

The cost is not only the missed measurement. A tracked prompt is a recurring expense: it runs on every engine on every cycle, and on most plans it occupies one of a fixed number of slots. A question that returns "what is generative engine optimization" produces a perfectly good answer and a permanently flat line.

Being Honest About That Number

That 7.6× is partly definitional and it would be dishonest to present it as a discovery. If an answer names no company, then in almost every case it also fails to name yours. The two events are close to the same event.

Two things survive that caveat. The left bar is not zero: at 4.3% a brand is occasionally the only name in an answer, usually when the question is about the brand itself. And more importantly, the size of the dead group is a real, checkable fact about how people build prompt sets. 14.8% is not a rounding error, and nobody chose it on purpose.

The useful question is therefore not "did this answer name vendors", which you learn too late. It is which prompts tend to produce answers that do, which you can decide in advance.

Category Is the Lever

The pale bar is the one to read first. Discovery prompts return an answer naming some vendor 92.5% of the time; expert prompts manage 69%. That is a 23.5-point spread decided entirely by how you phrase the question.

The outcome follows it. Discovery prompts name the tracked brand 36.5% of the time against problem / solution at 22.4%, a 14.1-point gap. A question shaped like "best tools for X, top providers" is asking for a shortlist. A question shaped like "how do I fix or handle X" is asking for advice, and advice often arrives without a single product in it.

Notice that expert prompts are the weakest on vendor-naming yet not the weakest on mentions. When a deep technical question does name products, it names fewer of them, so a brand that makes the cut takes a bigger share of it. That is a reason to keep a few rather than to drop them.

What a Dead Prompt Looks Like

Dead prompts are rarely bad questions. They are usually good questions of the wrong shape, which is why they survive review so easily. Three patterns account for most of them.

• 01The definition. Anything shaped like "what is X" returns an explanation. It is often the highest-volume phrase in a category, which is exactly why people track it, and the answer has no room for a vendor.
• 02The yes or no. "Does X still matter in 2026" gets an opinion and a couple of caveats. Nobody is named because nobody needs to be.
• 03The pure how-to. "How do I improve X" returns steps. Steps sometimes mention a tool and mostly do not, which is why this category sits in the middle of the chart above rather than at either end.

The fix is usually a rewrite rather than a deletion. "What is answer engine optimization" is dead; "best answer engine optimization tools for a small team" is the same underlying interest expressed as a question that has to name someone. The topic was never the problem.

If you already have a set running, the audit is quick. Sort your prompts by how many distinct companies have appeared in their answers across all runs so far. The ones sitting at zero are not slow starters, and after a few cycles they are safe to replace. Prompts that consistently return a list, even a list you are absent from, are the ones worth keeping: being absent from a shortlist is a measurable, fixable position, which is more than a definition will ever give you.

How Many Prompts to Track

More than feels necessary, for a reason that has nothing to do with coverage. A visibility score is an average over your prompt set, and a small set produces a number that swings on its own. We measured that separately: a single category slice moves 27 points between runs with nothing changed, against 15 for the full set, precisely because fewer prompts sit inside it.

So a ten-prompt account is not just narrower than a fifty-prompt account. It is noisier, and it will hand you apparent wins and losses that are neither. Track fewer things badly and you get a number you cannot act on.

For a starting set we use 3 discovery, 3 problem / solution, 2 use case and 2 expert, 10 in total. Discovery gets a full share because it leads on both measures. The second three-slot category is a coverage decision rather than a vendor-naming one: problem-solution questions are how a lot of buyers actually open a search, even though use-case prompts score higher on naming anyone.

The Branded Prompt Trap

The opposite failure is more tempting and does more damage. A prompt containing your own name is guaranteed to mention you, so it lifts the score immediately and permanently. It is the fastest way to make a dashboard look good and the fastest way to make it useless.

The reason is that visibility is supposed to answer one question: does AI bring you up when someone describes their problem without knowing you exist. Ask "is [your brand] any good" and you have measured whether the model can read your name back to you. That is worth knowing occasionally, for accuracy and sentiment, but it is not visibility and it should never sit in the same average.

It is also the number that quietly breaks comparisons. Two accounts with different branded shares are not measuring the same thing, so neither their scores nor their trends can be read against each other. If you inherit an account whose score looks unusually strong, the branded share of its prompt set is the first thing to check, and the most common explanation.

We generate starter sets unbranded by default and discard outright any prompt that comes back carrying the brand name, rather than editing it out. The check is worth running on any set you did not build yourself: count how many of your tracked prompts contain your company name, and treat anything above zero as a decision someone should be able to justify.

How to Choose Yours

• 01Ask the answer-shape question first. Before adding a prompt, ask whether a good answer to it could name companies. If not, it is a dead slot no matter how relevant the topic.
• 02Weight toward discovery. It is the strongest category on both measures, and it maps to the moment a buyer is actually choosing.
• 03Keep them unbranded. A prompt containing your own name will mention you and teaches you nothing. It also inflates the score, which is worse than useless when you report it.
• 04Audit the dead ones after a few runs. Any prompt that has returned no vendor at all across several runs is not going to start. Replace it; the slot is worth more elsewhere.
• 05Then aim your content at the format. Discovery prompts mostly return ranked lists, so the work that follows is getting onto those lists, not rewriting your homepage.

One caveat on the numbers above. Categories are our own labels, applied when a prompt is created, and the boundaries between them are fuzzy at the edges. The ordering has held across our corpus, but treat the gap between two adjacent categories as directional rather than precise.