Health, Cooking, and Home

By the end of this lesson, you will be able to use a free AI chatbot to plan a week of meals around what’s already in your fridge, get a plain-English explanation of a lab result or a medication label before your next doctor’s appointment, and talk through a home problem like a leaking tap or a stained carpet without paying for a call-out. You will also know the specific things AI gets wrong in these areas, and how to check its answers.

The idea in plain English

Most people meet AI at work first — writing an email, tidying up a spreadsheet. But the place it earns its keep fastest is the boring domestic stuff: the 6pm “what are we eating” panic, the letter from the clinic full of words you have never seen, the appliance making a noise it did not make last week.

Here is the useful way to think about it. A chatbot like ChatGPT, Claude, or Gemini is a very well-read helper who has never met you. It has read enormous amounts of text — recipes, medical explainers, repair manuals, forum threads from people with your exact problem. What it has not read is your kitchen, your body, or your house. So the quality of what you get back depends almost entirely on how much of your situation you hand over.

“Give me a healthy dinner recipe” gets you a generic chicken and broccoli. “I have half a cabbage, three eggs, a tin of chickpeas, and rice. No dairy. Twenty minutes. One pan” gets you something you can actually cook tonight. Same tool, completely different value. That gap — between a vague question and a specific one — is the whole skill.

One boundary before we go further, and it matters: AI is a translator and a first draft, not a professional. It can tell you what “elevated ALT” means so you walk into your appointment able to ask a real question. It cannot tell you whether you are sick. Use it to prepare for the expert, never to replace them.

How it actually works

These tools work by predicting text. They have absorbed patterns from a huge amount of writing, and when you ask something, they produce the words that most plausibly follow your question. That mechanism explains both why they are good at this and where they fail.

Why they are good at cooking: recipes are extremely patterned. Aromatics, then protein, then liquid, then simmer. The model has seen that shape thousands of times, so it can improvise a plausible version with your specific ingredients. Substitutions work the same way — it knows buttermilk and “milk plus a splash of vinegar” appear in similar contexts.

Why they are good at explaining health documents: medical language is a vocabulary problem. There are millions of pages explaining “hypertension” in simple terms. Asking a chatbot to rewrite your discharge notes at a reading level you are comfortable with is squarely what it does best.

Where it breaks: because it is predicting plausible text rather than looking things up, it will sometimes state something wrong with complete confidence. The industry word for this is hallucination — a confident, fluent, invented answer. In cooking that means an oven temperature that is a bit off, or an underestimated cooking time. Annoying, survivable. In health it means an invented drug interaction or a made-up “normal range” for a blood test. That is not survivable, which is why you verify anything with a health consequence against a real source.

Two practical notes. First, photos work. ChatGPT, Claude, and Gemini all accept images on their free tiers. Photograph the inside of your fridge, a plant that is dying, the label on a bottle of cleaning fluid, the model plate on the back of your washing machine. This is often faster and more accurate than describing something in words. Second, ask for sources on anything health-related. A tool like Perplexity is built around this — it answers with links to where each claim came from, which means you can click through and check. For “is this supplement safe with my blood pressure medication,” start there, then confirm with a pharmacist. Pharmacists answer that exact question for free, every day, and they know your actual prescription.

Try this now

Two exercises, about five minutes each. Do them in whichever free chatbot you already have open — ChatGPT, Claude, and Gemini all handle these fine.

1. Open your fridge and take one photo. Get the shelves and the door in frame. Open the cupboard where you keep tins, rice, and pasta and take a second photo. Do not tidy up first; the mess is the point.
2. Start a new chat and upload both photos using the paperclip or image icon near the message box. Then paste this, editing the bracketed parts to match your life:

   Here are photos of my fridge and my dry-goods cupboard. Based on what you can actually see plus normal pantry basics like salt, oil, and pepper, plan three dinners for [2] people this week.
   
   Rules:
   - No trips to the shop. Use what's in the photos.
   - Each one under [30] minutes, and I'm a nervous cook, so keep the techniques simple.
   - Avoid [dairy] — I can't eat it.
   - Use the [spinach] first, it's about to turn.
   
   For each dinner give me: the name, the ingredients with rough amounts, and numbered steps. Then at the end, list anything in the photos that's going to go off in the next two days and give me one idea for using it up.

3. Push back on the first answer. This is the step most beginners skip, and it is where the quality actually comes from. Reply with something honest: “The second one needs a pan I don’t own — redo it for one pot,” or “This is bland, add more spice,” or “My kids won’t eat mushrooms.” A chatbot remembers the conversation, so it adjusts rather than starting over. Two or three rounds of this and you have a real plan.
4. Now the health exercise. Find something you have been quietly confused by — a line from a blood test, a leaflet in a medicine box, a term a doctor used quickly. Type or photograph it, then use this:

   I'm not a medical professional and I don't want a diagnosis. I want to understand the words so I can ask my doctor better questions.
   
   Here's what I have: [paste the result, term, or label — for example "eGFR 58" or "take 500mg BD PRN"]
   
   Please:
   1. Explain what this means in plain English, as if to someone with no medical background.
   2. Explain what it's typically measuring or doing, and why a doctor would care.
   3. Give me four specific questions to ask my doctor about this at my next appointment.
   4. Tell me clearly which parts of your answer are general information and which would depend on my personal history, and flag anything I should verify with a professional rather than trust from you.

5. Read point 4 of the answer carefully. That is the honest part. Anything the AI flags as depending on your history is a question for a human, not a thing you have now learned.
6. Optional, two minutes: photograph one thing in your home that is broken, stained, or making a noise, and ask: “What is most likely causing this, what would I check first, and is this a DIY job or do I need a professional? Be honest if it’s a professional job.” That last sentence matters — without it, these tools tend to assume you want to attempt everything yourself.

Common mistakes

• Asking a vague question and concluding AI is useless. “Give me a healthy meal plan” produces something forgettable. The fix: always include four things — what you have, how much time, who is eating, and what you cannot or will not eat. Constraints are not limitations here, they are the instructions.
• Treating a health answer as a second opinion. A fluent, well-organised paragraph about your symptoms feels authoritative, and that feeling is the danger — it is generated text, not a judgement about you. The fix: use AI only to translate vocabulary and prepare questions. For anything about dosage, interactions, or whether to worry, ask a pharmacist or your doctor. Pharmacists are free and fast for medication questions.
• Trusting numbers without checking. Cooking temperatures, meat internal temps, calorie counts, and canning or preserving times are exactly the kind of specific detail these tools get slightly wrong. The fix: cross-check any number where being wrong has a consequence. Chicken to 74°C / 165°F is worth confirming; how long to soften an onion is not.
• Accepting the first answer. Most people ask once, get something mediocre, and quit. The fix: treat it as a conversation. “Too complicated.” “I don’t have that ingredient.” “My oven doesn’t have a grill setting.” Each correction gets you closer, and it takes seconds.
• Pasting in more personal information than you need to. Free chatbot conversations may be reviewed or used to improve the service. The fix: strip out your name, address, date of birth, and any patient or insurance ID before pasting from a document. “eGFR 58, 54-year-old man” is enough context — the rest adds nothing to the answer and everything to your exposure.

Key takeaways

• Specifics are the whole skill: what you have, how long you have, who is eating, what you must avoid.
• Use photos. Your fridge, a label, a dying plant, a model number — faster and more accurate than describing it.
• For health, AI is a translator that helps you ask better questions. It is never the answer. Pharmacists and doctors are.
• Verify any number that carries a consequence — cooking temperatures, doses, preserving times.
• Argue with the first draft. Two or three corrections turn a generic reply into something you can actually use.