Reviewing and rewriting policies

Every service has policies that are older than the person responsible for them. Reviewing them is important, unglamorous, and permanently near the bottom of the list.

AI helps with this, with one significant caveat that applies more here than anywhere else.

The caveat: it does not know the rules

Asked about a regulation, a care standard or an inspection requirement, an AI assistant will answer confidently. It may cite a regulation number, quote a clause, or describe a requirement in detail.

It may be right. It may be describing superseded guidance, or guidance from another country, or something it has constructed because your question implied such a rule should exist.

There is no way to tell from the output, because it is equally fluent either way. CQC’s own risk list notes AI “can hallucinate and present untrue information, authoritatively as fact”.

So: supply the guidance rather than asking for it. Paste in the actual standard you are checking against. Used that way, the tool compares two documents you control, which is a task it does well. Used the other way, it is inventing the standard you are measuring yourself against, which is worse than not checking at all.

Where it helps

Consistency checking. “Does this eleven-page policy contradict itself anywhere?” is a genuinely useful question and a tedious one to answer by hand.

Comparing against a supplied checklist. Give it the policy and the standard, and ask which parts of the standard are not addressed. This is the strongest use on this page.

Plainer wording. Policies are written in a register nobody reads. A policy staff do not understand is not a control, whatever it says.

Spotting stale references. Mentions of superseded bodies, old job titles, a form that no longer exists. Easy to miss, easy for a tool to flag.

Producing the staff-facing version. A one-page summary of an eleven-page policy, for the people expected to follow it.

Where it goes wrong

Generic policy, real inspection. The failure that matters. An AI-written policy reads well and describes a service that is not quite yours. At inspection the question is not whether you have a policy but whether it matches what people actually do. A polished policy describing practice you do not follow is evidence against you.

It strips the local detail. The sentence about the one flat with the awkward key safe is exactly the specific, unusual detail these tools smooth away — and exactly the detail a new member of staff needs.

Invented authority. Covered above, and worth repeating because it is the one that will embarrass you in front of an inspector.

A sensible method

1. Supply the current guidance. Do not ask the tool to recall it.
2. Ask for gaps, not for a rewrite. “Which parts of this standard does this policy not cover?” is safer and more useful than “update this policy”.
3. Check every citation. If the output names a regulation, verify it exists and says what the tool claims.
4. Keep your specifics. Before accepting a rewrite, check that the local detail survived.
5. Have someone who knows the rules sign it off. That has not changed.
6. Check it against practice. Ask two staff whether the policy describes what they actually do. If not, fix the policy or fix the practice, but do not leave the gap.

Start with your AI policy

If you are looking for a first policy to work on, the obvious one is the policy about AI — which half the sector does not have. Our one-page template is a starting point, and writing a one-page AI use policy explains what it needs to cover.