Most AI content assumes you’re trying to level up your career or automate a business. Fair enough. But some of the most consistently useful things I do with AI are stupidly domestic — meal planning, arguing with a confusing appliance manual, figuring out why a plant looks sad. None of it is glamorous. All of it saves actual time.
Here’s what I actually use it for, with enough specificity to be useful rather than just inspirational.
1. Decoding appliance manuals and error codes
This is the one I’d recommend to literally anyone. An appliance throws an error code, the manual says something like “E3 — contact service,” and that’s the entire entry. Paste the model number, the error code, and whatever context you have into ChatGPT or Claude, and you’ll usually get a real explanation of what the error actually means — and whether it’s something a non-expert can fix (a clogged filter, a stuck float switch) or genuinely needs a technician.
I’m not saying trust it blindly on anything electrical. But for “why does my dishwasher think there’s a water leak when there isn’t one,” it’s faster and more useful than most manufacturer support lines.
2. Meal planning against whatever is actually in the fridge
The prompt format that works: photograph or just type out what’s in your fridge, add any dietary constraints, and ask for three dinner options using what you already have. This sounds trivial until you realize how often the alternative is ordering delivery because you can’t think of what to do with half a butternut squash, some leftover rice, and an open can of coconut milk.
The model is better at this than recipe sites, which require you to already know what you’re making before you search. Here you’re starting from ingredients, not a dish name — and that turns out to be the harder direction to search in.
3. Translating what a tradesperson just said
A plumber tells you your “expansion vessel has lost charge” and you need to “repressurise the system.” Do you agree, push back, look for a second opinion? Hard to know when you don’t know what those words mean in context.
Ask Claude to explain what the thing is, what it does, why it fails, and what the fix involves. The goal isn’t to do the job yourself — it’s to have an informed conversation, know whether the quoted price is in the right ballpark, and understand whether it’s urgent. AI is actually pretty good at that specific problem.
4. Figuring out what’s wrong with a plant
Describe the symptoms specifically: yellowing from the bottom up versus the top, brown edges versus brown spots, soggy soil versus bone dry, how long since you watered, which direction the window faces. That specificity matters more than you’d think. “Yellow leaves” has about eight different causes; “yellowing from the oldest leaves upward with the soil staying wet for two weeks” narrows it to two or three.
ChatGPT-4o with vision is better here if you can upload a photo. But even text descriptions get you somewhere useful faster than most gardening forums, which tend to rabbit-hole into competing advice with no resolution.
5. Writing complaint letters that actually work
There’s a specific register that gets results from companies: firm, factual, references the relevant consumer protection framework without being threatening, asks for a concrete resolution rather than just venting. Most people don’t write this way naturally — they either go too emotional or too passive, especially when they’re frustrated at the point of writing.
Give the AI the facts and ask it to write a formal complaint in that register. Read it, correct anything that isn’t accurate, and send it. The draft takes 30 seconds. One caveat: if you’re dealing with anything involving real legal rights — a landlord deposit dispute, a significant financial fraud — get actual advice from a consumer rights organization or solicitor rather than relying on the letter alone. Useful for the draft. Not a lawyer.
6. Gift research for people who are hard to buy for
The model can’t shop for you, but it’s good at generating ideas you’d never have thought of. The trick is being specific about the person rather than the occasion. Not “gifts for my dad’s birthday” but “gifts for someone who is retired, spends most of their time gardening and doing crosswords, has no interest in technology, doesn’t drink, and lives alone — budget around £50.”
That level of specificity gets you off the generic gift-guide lists that every e-commerce site generates. You’re using AI to think, then going to find the actual product yourself.
7. Making sense of a dense document you have to sign
Tenancy agreements, insurance policy exclusions, the terms on a subscription you’re about to cancel. These documents are long, deliberately unreadable, and written by lawyers for lawyers. Paste a section — or the whole thing if it fits in the context window — and ask Claude to explain what it actually means in plain language, what your key obligations are, and whether anything looks unusual or worth questioning.
Not a replacement for legal advice when the stakes are high. But for “is this standard” or “what does this clause actually mean,” it’s genuinely useful and much faster than trying to parse legalese yourself.
Why these work when plenty of AI use cases don’t
None of these tasks require AI to be brilliant. They require it to be fast, patient, and specific — which it is. Most of them involve converting a messy, context-rich situation into a useful explanation or a first draft, and that’s exactly where language models are strongest: they’ve seen thousands of versions of your dishwasher error, your complaint letter, your sad-looking plant.
They’re weakest when the answer depends on something they can’t verify — whether that plumber’s diagnosis is actually correct, whether a clause in your lease is enforceable in your jurisdiction. That’s where you stop and check with a human who can be accountable for the answer. The model can tell you what an expansion vessel is; it can’t tell you if your specific one actually needs replacing.
If you’ve been mostly using AI for work tasks, try running one genuinely domestic problem through it this week — ideally something where you’d normally Google for 20 minutes and end up more confused. Pick the appliance error code or the complaint letter. The bar for useful is lower than you’d expect, and the time cost of trying is basically zero.
