How long until
it actually works?

Two different clocks, and conflating them is how expectations get broken. A connected system, lead capture through follow-up, runs in weeks. Getting an answer engine to name the business takes 30 to 60 days before anything moves and about 90 before it compounds. Anyone promising either result inside a month is selling a timeline rather than a system.

The Diagnostic takes two weeks and ends in a written scope. The build after it is measured in weeks, and the driver is the number of systems being connected rather than the complexity of the reasoning. The first thing that lands is usually output arriving somewhere a person reads it, on a schedule, without anyone starting it by hand.

Answer engines are slower and the reason is structural. Roughly half of buying questions never trigger a web search at all: the model answers from what it already holds, which was built from what other sites say about the brand across a training cycle. That does not respond to publishing a page this week. Changes to your own pages can land faster, in days, but they only affect the half where the engine actually looks.

Recall moves last. Citations move first, and the fastest documented on-site change is restructuring content into question-shaped pages, which produced first citations in four days in one published test. Watch mention rate and share of voice monthly, not traffic. Most AI answers produce no click at all, so a buyer the engine recommended arrives as direct traffic and looks like someone you already had.

A brand new domain, a recent rebrand, or no third-party presence at all. All three mean the model has little to remember. That is a mentions problem rather than a content problem, and it is fixed by getting written about rather than by writing more.