Five questions to answer before AI touches client work
Permissions, data and review: the rules to settle first, so nothing you build has to be undone.
The Newhour team · · 5 min read
The fastest way to lose a client's trust with AI is to use it on their material without asking. The fastest way to waste a project is to build something you later have to switch off. Both are avoidable with five questions.
1. Which clients allow it?
Check your agreements and ask where they are unclear. Record the answer for each client, and treat "unknown" as "no" until you hear otherwise.
2. What information can go in?
Decide which information can be used with which tool, based on the tool's data terms and your own obligations. Personal details, client names and unreleased work usually need the most care.
3. Where does the output live?
Keep drafts and results where your client files already live, so the access rules you have set up still apply.
4. Who checks it, and against what?
Anything AI produces that reaches a client needs a named person to review it, and a short list of what they check: facts, claims, tone, and anything the client has restricted.
5. How will you know it is working?
Agree up front what good looks like: less time on a step, fewer revision rounds, fewer details chased. Try it on real work before you roll it out.
None of this needs a policy document to start. A one-page record for each client and a review checklist will do, and they are exactly what any later AI system will depend on.