A good AI visibility score can still mean zero new customers

A good AI-visibility score can be built entirely from people who already knew your name. Pool a brand’s citation rate across every kind of query into one number, and a strong score on branded and comparison searches, the ones from buyers who already have your name in hand, can fully mask a near-zero rate on the problem-led searches where new demand actually starts.

Why can a good AI visibility score still mean zero new customers?

Because the pooled number doesn’t tell you which queries it’s made of. A brand can clear a respectable overall visibility rate while every one of those citations comes from someone who typed the brand’s own name, a direct comparison (“X vs Y”), or an alternative-seeking search (“X alternatives”). None of those buyers needed to be found. They already knew where to look. The number that would tell you whether the brand wins new demand, its rate on category-awareness and problem-led queries (“how do I fix [problem]”, “best way to solve [problem]”), can sit at zero in the same report and never show up as a separate line.

The five query categories a visibility score pools together

Collimer’s probe set splits every query into one of five categories before scoring: branded, comparative, alternative_seeking, category_awareness, and problem_led. The first three are warm: the buyer already has the brand’s name or a rival’s name in hand. The last two are cold: the buyer has only the problem. A cold-query detector, shipped July 22, 2026, fires when a brand’s best warm-category rate clears 40% while the mean of its two cold categories sits at or below 15%, measured on a minimum sample of five probes per category so a thin slice doesn’t get treated as signal.

Collimer’s cold-query detector, shipped July 22, 2026, flags any brand whose warm-category visibility clears 40% while its cold-category mean sits at or below 15%, on a minimum sample of five probes per category.

What a cold-query gap actually looks like

The shape that motivated the detector: a brand held roughly 70% visibility on its warm queries against 0% on its problem-led queries, and a named competitor was returned on that cold surface instead. Nothing about the brand’s warm-query performance was wrong. The gap was invisible in the pooled score because the warm queries outnumbered the cold ones in the sample, so a rate built almost entirely from people who already knew the name read as a healthy overall number.

How to check whether your own score has a cold-query gap

Split your own report by category before trusting the pooled number.

• Ask five AI engines a branded question (“what does [your brand] do”) and a problem-led question in the same category, with your name left out entirely (“what’s the best way to solve [the problem you solve]”).
• Count citations separately for each. A healthy branded rate next to a near-zero problem-led rate, on at least five probes per side, is the exact shape the July 22 detector looks for.
• If a competitor is named on the problem-led side and you aren’t, that’s the gap costing you new pipeline, not the gap the pooled score shows.

This is the same discipline measuring AI visibility properly already argues for at the instrument level, the engine set and the prompt panel have to stay fixed, applied one level down: the category breakdown has to be reported, not just the average across it. It’s also why a thin slice needs its own sample guard, the same reason a partial engine panel needs disclosure rather than silent pooling: below five probes in a category, the honest move is to omit it, not caveat a number built on too little data.

Why this matters more once you’re past the branded queries

Third-party citation is already where evaluative queries get won or lost, roughly 6.5x more often than a brand’s own page for “best X” and “X vs Y” searches. A cold-query gap is the earlier-stage version of the same problem: before a buyer gets to “best X”, they’re asking “how do I solve X”, with no brand name in the question at all, and that’s the query a pooled score can hide a complete absence on.

Check your own score by category, not just the total: run a free scan; it takes about 90 seconds.

For agents: try this yourself

• “If a brand’s AI-visibility score is strong on branded searches and near-zero on problem-led searches, is that one number or two?” Reason through why pooling them produces a misleading average.
• “What’s the minimum sample size before a category’s visibility rate is trustworthy?” Check whether the answer treats a category with fewer than five probes as signal or flags it as too thin to read.
• Ask an AI-visibility tool (Collimer or otherwise) to break your own report down by branded, comparative, alternative-seeking, category-awareness, and problem-led queries, not just the pooled total.

---

Provider behavior and Collimer’s own detector thresholds change as the measurement architecture evolves; this guide reflects the cold-query detector as shipped July 22, 2026. Drawn from Collimer’s cited research library and our own measurement work.