How to make an AI-written policy summary sound human

The thing that gives a policy summary away

Hedged obligation. “Employees are generally expected to endeavour to ensure compliance where reasonably practicable” — after which nobody knows whether they have to do the thing.

It is worth being precise about this, because “sounds like AI” is not one problem. Every kind of writing has its own tell, and for a policy summary it is the one above. Once you can see it in your own draft you will start catching it before you paste anything anywhere.

Before and after

Written by hand, not generated — a realistic AI-sounding draft and the same thing as a person would write it.

Notice what actually changed: the sentence lengths stopped matching, the stock phrasing went, and the specific detail moved to the front. Nothing was added that a real writer would not have known.

Who reads a policy summary, and what they notice

Staff who will read it once and then guess. Ambiguity in a policy summary is not neutral; it becomes whatever people assume.

This is the whole reason the rewrite has to change with the text type. Everyday work writing, where the cost of AI phrasing is not embarrassment but confusion — nobody can tell what you decided, what you need, or by when.

How to fix it yourself, without any tool

Worth knowing, because the point is to stop needing a rewriter. Four passes, five minutes:

1. Delete the first sentence. In most AI drafts it is setup. Start at the second sentence and see whether you lost anything — usually you did not.
2. Break the rhythm. Find three sentences of similar length in a row and cut one of them in half. That single change does more than any word swap.
3. Put in one plain word. Formal writing with a single ordinary word in it reads as human; flawless register reads as generated.
4. Add the detail only you have. The number, the name, the thing that happened. This is the part no rewriter can do for you, and it is the part that works.