This lesson takes a single cluster and builds the whole ad group from it, so you can see how the parts interlock. The cluster: a regional accounting firm, 60 staff, triggered by a compliance deadline, constrained by a data-residency requirement, wanting document workflow automation live before year end.
Start with the scope line, because it is the hardest constraint and the one most competitors will fail. Data residency is binary and verifiable: either your material states where data is stored or it does not. Put it in the hint, put it in the feed, and put it on the landing page in the same words.
Next write the disqualifier from the situations you would lose. You are not for solo practitioners, and you are not for firms that need on-premise deployment. Saying so removes you from prompt families where a cheaper or heavier competitor would beat you anyway, and concentrates your presence where you win.
Then the core promise. A promise is one sentence the engine can quote verbatim, stating the outcome and the condition of achieving it: "Document workflow live in four weeks with EU-only data residency, without replacing your practice management system." Conditions make promises credible; unconditional superlatives make them ignorable.
Now the landing path. Send answer traffic to a page that restates the same constraints in the same order. A mismatch between the quoted promise and the page is one of the fastest ways to lose both conversion and future citations, because engines do sample destination content.
Finally, record the expected prompts and add them to the monitoring set. An ad group is not finished when it is written; it is finished when you know which prompts should surface it and you are watching them.
The interlock is the point. Cluster defines the situation, hints target it, the promise gets quoted, the feed proves it, the page confirms it, and the monitoring set tells you whether it worked. Break any link and the group underperforms in a way that is hard to diagnose from the reporting alone.