Once your AI tool is connected, the easiest way to get going is to just ask. Here are a few favorites our own team finds useful, ready for you to copy and use, just swap in your own project or category name, and send it.

‍

💡 Tip: keep your AI tool's confirmation prompts on. You'll approve each change before it runs — it costs a click, and it's the difference between an assistant and a liability.

‍

‍

Check if it already exists

Before I write a new error message for [a failed download], check whether we already have copy that says something close to this. Show me what we have, what it says, and what status each one is in. Don't create anything yet.

A good answer searches by meaning, not just keyword — and tells you which matches are Final and which are still Draft. The status is what tells you whether it's safe to reuse.

‍

‍

Find what's not Final yet

We ship [the sign-in flow] on Friday. Go through it frame by frame and tell me what copy is not Final yet — count by status first so I can see the size of it, then list what's outstanding.

Ask for the count before the list. A number tells you whether to worry; the list tells you what to fix.

‍

‍

Spot duplicate copy

Are there components in [the Commons category] that say almost the same thing? Show me the overlaps and which one looks like the canonical version.

Worth running once a quarter — every duplicate gets translated separately, reviewed separately, and drifts separately.

‍

‍

Check translation coverage

How many components in [the Calmo Mobile App category] have no [German] translation, and which ones are they? Show me coverage by language first, then the list.

‍

‍

Put a flow up for review

Put the copy on [the sign-in flow] up for review — move what's ready to Review, tag the batch so the reviewer can find it, and leave a comment on each one saying what changed.

Your agent prepares and files the handoff — it can't notify anyone, so you'll still want to share the link and let the reviewer know it's waiting.

‍

‍

Hand off approved strings

I'm implementing [the password reset screen]. Give me the approved copy for it with the keys, and flag anything that isn't Final yet so I know what might still change.

‍

‍

Getting better answers

• Ask for a dry run first. Most of these tools can tell you what they *would* do without doing it — ask for that when a prompt touches more than a couple of components.
• Name the scope. "The Billing project" gets a better answer than "our copy." The narrower the scope, the less your agent has to guess.
• Ask for counts before lists. A number tells you whether there's a problem; a list tells you what it is.

‍

Want more on getting the most out of this? Read getting good results, or head back to connect your AI tools if you haven't yet.