Keep a person in charge

Please note: This site is only a guide to help people understand. Consult your IT, security and/or privacy department for clarification.

What it means

Being a human in the loop means a person stays in the process. The AI drafts. A person reviews. A person approves.

Nothing important should go from the tool straight to a donor, a funder or a client.

Reading the draft closely counts. A quick glance that approves everything is not a review.

Who reviews what

Match the reviewer to the risk. A brainstormed list of event ideas needs a quick look from you. A public post needs a careful read by the person who owns that channel.

Anything for the board, a funder or the public gets a second reviewer. Two sets of eyes catch more than one.

A simple review checklist

Read the draft slowly with these questions in mind.

Facts: is each claim true? Did you check it against a record?
Names: are people, places and programs spelled right and described right?
Numbers: do dollar amounts, dates and counts match your records?
Tone: does it sound like your organization? Is it kind and respectful?
Promises: does it promise anything your team cannot deliver?
Private details: does it include anything about a person that should stay private?

Where AI must not decide

A person decides; AI can only help draft.

That is the rule for any choice that affects someone's life. AI must not:

decide who is eligible for services or help
rank or prioritize clients, families or beneficiaries
screen, rank, hire, promote, discipline or fire people
decide anything about someone's benefits, housing or legal status

Why those lines exist

AI can be wrong. It can also be unfair in ways that are hard to see. And people deserve to know that an accountable human made the call and can explain it.

AI can help you draft the policy or the criteria document, such as a written list of how your team decides who gets a spot. People then apply those criteria and make the decisions.

Keep a simple note

For important documents, write down where AI helped, who reviewed it and the date. A line at the bottom of your working file is enough.

If someone asks later how a document was made, you can answer.

When a reviewer finds a problem

Fix it first. Then note what went wrong, for example a wrong name or an invented number.

Last, change the prompt or the workflow so it is less likely to happen again. Add the missing facts, or add a check step.