Being left out of an AI answer is not the same as ranking tenth.

The shape of the problem changed

A rank tracker reports where you place on a results page. The question that now matters is whether an assistant names you at all, and the two can disagree completely. Plenty of products rank respectably and are never mentioned.

Most tools tell you the score. Almost none tell you why.

Tracking mentions is the easy half, and by now it is table stakes. The hard half is the question that follows it, and it is where the category is thin:

• You get a number, not a cause. “Mentioned in 3 of 20 prompts” is a scoreboard. It does not say whether the problem is that a model cannot read your site, that your own pages never state anything specific, or that the round-ups shaping the answer have never heard of you — and those need completely different work.
• Advice arrives detached from evidence. Generic content recommendations are cheap to generate and impossible to check. A suggestion is only worth acting on if you can see the page, search result, or answer it came from.
• Your own site is not where the answer comes from. Models recommending from memory are drawing on round-ups, comparison articles and review sites. A tool that audits only your pages is measuring the half you control and ignoring the half that decides.

We measure the conditions, not just the outcome

The outcome is one prompt's answer on one day. The conditions are what produced it, and they split into questions with genuinely different answers — which is why GeoVouch scores them separately instead of blending them into a single readiness number:

• Can a machine read you at all. Robots rules, structured data, whether your claims survive extraction. Concrete, checkable, usually a day's work.
• Does your own content say anything. Whether your pages state what the product is, who it is for, and back it with specifics rather than adjectives — scored on first-party evidence only.
• Does anyone else corroborate it. What independent sources publish about you, and in what depth. Slower to move, and usually the real constraint.
• Do you exist when nobody types your name. The de-branded question. Searching your own brand and finding yourself proves nothing.

Kept apart on purpose: a single blended score would hide which one is actually holding you back, and that is the only part you can act on.

Who this actually works for

Everything here depends on one thing being true: your site has to stand for a product or a brand that an assistant could name in an answer. That is a structural requirement, not a preference — step 1 resolves that entity and every later step measures it, so a site with nothing of its own to resolve does not simply score low, it produces a confident report about somebody else.

• Software, apps and web products. The common case: a named product, its own homepage, a category other people publish comparisons about.
• Own-brand ecommerce. A DTC store whose brand is the thing being recommended works exactly the same way — the brand is the entity, and its category ("running shoes for flat feet", not "shoes") is where the conversation happens. A store that resells other brands does not: the entity in those articles is the brand on the box, not the shop.
• One product inside a bigger company. Point it at that product's page. A group homepage describes an organisation, and buyers do not ask assistants for organisations.

The inverse is worth saying plainly, because it is the more useful half: if your own pages never establish what category you are in, this analysis stops and tells you so rather than borrowing a category from elsewhere. That is not the tool failing — assistants are reading the same pages, and they cannot establish it either.

Pay for the analysis, not for the subscription

• Charged per step, as it runs. Each step shows its price on the button that starts it. Stop after step 4 and you paid for four steps.
• Nothing recurring. No seat, no monthly floor, no annual contract. Most tools in this category start in the hundreds per month, which only makes sense if you open a dashboard daily — and almost nobody does.
• Revisiting is free. A finished step is stored. Re-reading what you already paid for never costs again, and a step that fails returns its Credits.

What we will not claim

Nobody can honestly promise to change an assistant's answer. The models change, the web changes, and the same question can come back different next week. What you get is a dated snapshot of what today's answer is built on, with every number pointing at the page, search result, or model response it came from — so you can check the finding rather than trust the summary.