Can You Trust an AI Chatbot on Noise Nuisance Law? What the Evidence Shows

Which?, the UK’s consumer rights organisation, issued a direct warning that AI chatbots “make up fake laws” when people use them to help with formal complaints — in their case, bank complaints, but the underlying failure is the model, not the subject. A chatbot asked about banking regulation and a chatbot asked about statutory nuisance are running on exactly the same tendency to generate a plausible, confidently-worded answer whether or not the underlying provision is real.

No published study has specifically tested ChatGPT, Claude or Perplexity against the Environmental Protection Act 1990’s statutory nuisance provisions, or the “clear days” rules that govern a valid notice. We haven’t found one, and we’re not going to claim one exists. What we can say honestly is narrower, and still matters: the documented failure is general — it’s about how these tools work, not about banking law specifically — so there’s no real reason to assume noise nuisance is the one area immune to it. Given what’s actually at stake — a case a magistrates’ court either accepts or doesn’t — that’s not a gap worth being the test case for.

A general-purpose chatbot has read enormous amounts of UK legal writing, which is exactly why its answers sound right — the sentence structure, the section-number format, the measured tone are all genuinely drawn from real legal text. What it isn’t doing is looking anything up in the moment. If the specific provision, time limit or procedural detail your situation turns on doesn’t land correctly in that generated text, nothing in how the answer reads will tell you that’s happened. The separate, documented problem of AI-drafted submissions arriving longer and harder for a council to act on compounds this: a confidently wrong complaint doesn’t just risk being wrong, it risks being slower too.

For statutory nuisance specifically, the provision that matters is Section 82 of the Environmental Protection Act 1990 — worth reading in its actual, current form rather than a generated summary of it, precisely because a generated summary is the thing that can quietly drift from what the Act says.

None of this means avoid AI tools entirely — it means not trusting one, unverified, with the legal basis a case depends on. The Case Toolkit doesn’t generate a fresh answer about the law every time you ask it something; it’s built on the Act’s actual text, cited correctly, every time — the same letters, the same statutory references, the same structure an officer or a magistrate expects to see, not a new roll of the dice per prompt.

• The correct statutory basis, cited the same way every time — not regenerated and re-worded per request.
• A dated, structured diary an officer can assess quickly, not a paragraph they have to extract the facts from.
• Letters to your council that quote the Environmental Protection Act 1990 correctly, in the right order.
• A path to the magistrates’ court under Section 82 if the council still won’t act — built once, not improvised.

That’s not a reason to start again from nothing — it’s a reason to check it. Read it against the real Section 82 text, and see our guide on what actually counts as proof of a statutory nuisance before you send anything citing law you haven’t verified yourself.