The Answer Engine's whole premise is that we measure how you're showing up, not just promise to. So the method is published, not a black box. Read it before you have to trust a single number we report.
Method live. LocalCraft's own baseline: pending. Client measurement: Measurement goes live with our founding clients; we're onboarding founding clients now. Status as of .
Every answer below is the real, current method, and the same text is published as this page's FAQPage markup.
A frozen panel of roughly 30 buyer-intent prompts per client: phrasings like "best double glazing perth" or "upvc windows cost perth", plus suburb and variant forms a real buyer would type. The panel is versioned: its exact contents ship with the measurement harness, not enumerated on this page, so a client can't be gamed by knowing the exact wording in advance.
A panel change is a new version. Trend lines are never silently re-based. If the prompts change, the history says so, rather than quietly presenting a new baseline as a continuation of the old one.
ChatGPT, Perplexity and Gemini will be measured through their APIs, each prompt asked as a Perth buyer would ask it, close to what a buyer sees, not identical to it. Google AI Overviews will be captured as the real Perth search result, with a genuine Perth location set on the query, which is why it's the anchor: the one true consumer surface, not a proxy for it. Every reading will be labelled with which kind it is, so the two are never blurred together.
Three samples per prompt per week. We'll report a 4-week rolling share-of-voice against 2 to 3 named competitors, and only flag movement outside a stated noise band. A single answer on a single day is close to pure noise, not a result. For each sample we'll record whether the business was mentioned, its position, whether the mention was linked or unlinked, and which source domain the engine cited. An AI Overview not appearing for a given query will be recorded as a datapoint, not treated as a failure.
The noise band, and one reading outside it.
No. Nobody can honestly guarantee what an AI recommends, and anyone who promises that is selling you something else. We improve the signals that earn recommendations, we measure how you're showing up, and we show you exactly how we measure it. The work and the number. Never magic.
Measurement goes live with our founding clients; we're onboarding founding clients now. This page publishes the method before the first trend line exists, on purpose: you should be able to check how we measure before you have to trust a single number we report.
Our own category, our own panel. The number publishes here once it's measured.
The method above isn't just something we sell: it's built for LocalCraft's own category too: a frozen, versioned panel of buyer-intent prompts about what we do, held to everything this page describes, including positive controls. Those controls are prompts where a known business should be named in the answer; they exist so an empty reading can be told apart from a broken run. A run whose controls detect nothing produces no publishable result. That outcome is unmeasured, not zero.
Baseline: not yet measured. When the first measurement runs, it will be published here, including if the answer is unflattering. Too early to call is a result we will publish.
Questions about the method: info@localcrft.com.au.