Three turns and a check. Keep this open next to your chat: each step is the thing to paste next, with what you typed already carried across. Nothing is sent from here, and nothing on this page is a transcript. The conversation stays where you are having it.
The question as it was asked, in the words it was asked in.
The states or countries in scope and why each is on the list.
The operations that create exposure: people, customers, revenue, entities.
What you actually have, since it decides how many jurisdictions get deep work.
Databases, subscriptions, and surveys you have access to.
Watch for, the whole way through
The model has no reliable knowledge of any state's current rule and will produce thresholds, effective dates, and citations that are wrong. This prompt exists to scope the work, not to do it.
A filled cell never gets re-checked. An empty cell marked "not researched" is safer than a plausible guess, and this is the reason the matrix ships blank.
Home-rule cities and counties defeat state-level surveys routinely. A state column that is correct can still be wrong for the city where most of the workforce sits.
Effective dates and grandfathering are where multistate work goes wrong quietly. Confirm which version applies to agreements already signed before you rely on any answer.
Do not paste client-confidential operational data unless your company's AI policy permits it. Headcount by state is often more sensitive than it looks.