Using AI for Regs Questions - What to Trust and What to Check

AI is on jobs already. Engineers use it to summarise documents, draft scope notes and write variation letters. Some also ask it regs questions between jobs.

Using it is a given. The harder part is using it without sending out a certificate built on a wrong answer.

AI is good at some things. It’s risky for others. Knowing the difference is most of the job.

What AI Is Useful For

AI is good at the parts of the job that eat time but don’t decide compliance. Pull a long email into a tidy summary. Rewrite a customer-facing note in plain English. Draft a method statement from a few bullet points. Build a first version of a quote or a scope of works so you have something to edit rather than something to write from scratch.

It is also useful as a starting point on a topic you don’t know well. Ask it to explain functional earthing or how SPDs work and you’ll get a clear, readable summary. Treat that as a way in, not a final answer.

AI saves you time on anything that isn’t the regulation itself. Drafting, summarising, tidying up.

Where It Falls Down

The thing to remember is what AI is and isn’t.

AI is like a well-read stranger. It might be right. It might sound right and be wrong. You can’t tell from the answer.

AI doesn’t know the difference between a correct answer and a confident-sounding wrong one. It is trained on what it has read, and what it has read can be out of date, partial, or just wrong. Forum posts. Old training notes. Articles written under earlier amendments. The answer comes back in the same tone whether it is right or not.

For regs questions, that is a problem. A Zs limit, a cable size, a disconnection time. These need to come from BS 7671, not from a confident paragraph that might be off.

One last trap. Don’t ask AI to check its own answer. Ask any AI ‘is this correct’ and it’ll usually tell you yes. Sounding sure is the default. That doesn’t mean it is.

The Amendment 4 Problem

Amendment 4 was published in April 2026. Most AI tools were trained months or years before that. They don’t know what is in the Orange Book. Some don’t even know it exists.

That means anything AI tells you right now about a regulation value, a section number or a cable rating is, at best, based on the previous edition. At worst it is older than that, mixed with forum advice it absorbed somewhere.

This will improve as AI tools update. For the rest of 2026, treat any AI answer on a BS 7671 detail as starting from the wrong book.

15 October sharpens this. The previous edition is withdrawn on that date, and from then on every new certificate has to be to Amendment 4. An AI giving you Brown Book values past 15 October is giving you values you can no longer use. There’s no guarantee the major AI tools will be updated to Amendment 4 by then. Their updates happen on their schedule, not yours.

What to Double-Check

A rough rule. If the AI answer is going to shape a certificate, check it. If it’s only going on an email or a draft, you can usually let it run.

Worth checking every time.

• Any number. Zs limits, voltage drop, cable ratings, disconnection times, fuse and MCB ratings. AI confidently quotes old figures.
• Any regulation reference. AI sometimes invents section numbers that sound right. If you can’t find it in the regs, it might not be there.
• Anything tied to a date or an amendment. AI may be answering as if it is 2023. Verify the edition it is referring to.
• Anything safety-critical. RCD ratings, earthing arrangements, special locations. If a mistake here ends up on a certificate, AI’s confidence won’t help you in front of a scheme assessor.

If you’re using AI to help you learn, none of this is a problem. If you’re using it to fill in something that has your name on it, the rule is simple. Check it against the book.

Where Our Software Sits in This

Shine Forms is bringing AI in where it earns its keep, on the parts of the job that don’t decide compliance. More on that side of things in the Chief AI Officer piece.

For regulation compliance, the software doesn’t lean on AI. The checks come from BS 7671 rules verified directly against BS 7671 by someone who helps write it. Enter a value, the software checks it against fixed limits. Not against a probable answer.

The combination is the point. AI where AI is useful. Verified rules where the answer needs to be right.

For more on how the regulation side works, read What We Mean By “Regulation Intelligence” (And Why It’s Not AI). And to meet the person leading the AI work at Shine Forms, see Meet Dr David Dickie, Our Chief AI Officer.