№ 00cFor non-technical users · Past the first chat
Everyday AI.
Recipes for using AI in real life. Letters you didn't want to write, statements you didn't want to read, doctor's appointments you'd rather walk into prepared. You've chatted with one of these tools a few times. Now here's what to actually do with it.
The mindset shift
The single most useful thing to internalize is that AI is an assistant, not a search engine. Google looks things up. AI helps you write, summarize, plan, and think out loud. Confusing the two is why people get frustrated.
The frame I keep coming back to is this: treat it like a very sharp friend who has read everything, can write in any voice you ask for, and will sometimes confidently get a fact wrong. You wouldn't ask that friend for your child's medication dose without checking. You also wouldn't write a thank-you note from scratch when they're sitting right there. Same instincts apply.
Writing letters and emails
This is where most people first feel the click. The blank page is the hardest part of any letter. AI is good at first drafts. You'll never send the first draft as-is, but you'll save the half-hour you would have spent staring at the cursor.
The pattern is always the same. Tell it the audience, the goal, and the tone. Three short clauses. Then read what it gives you and adjust.
1. A thank-you note
Short, warm, specific. Your goal is to sound like yourself, not like a Hallmark card.
Help me write a thank-you note to my neighbor Sandra, who watered our plants and brought in the mail while we were at my mother's funeral. Two short paragraphs, warm but not gushing, and please mention that the lemon tree is still alive thanks to her.
You'll get something workable. If it sounds too formal, say "more like how I'd actually talk." If it's too casual, say "a little more grown-up." Keep nudging until it sounds like you.
2. An insurance complaint
The kind of letter where being polite but firm matters more than being clever.
Help me write an email to my health insurance company. They denied a claim for an MRI my doctor said was medically necessary. Claim number 4429-X. I want them to reconsider. Polite, firm, and clear. Mention that I've been a member since 2012 and that the denial letter cited the wrong CPT code.
Read what it produces with a careful eye. Insurance language is legalese-adjacent, and the model can occasionally invent a process step that doesn't exist. Verify any procedural claim ("you may file a Level 2 appeal within 60 days") against the actual denial letter before sending.
3. The awkward decline
The email you've been putting off for a week. Wedding you can't go to. Dinner with the in-laws you need to skip. Cousin's MLM pitch you're not interested in.
Help me write a short email declining my sister-in-law's invitation to her essential-oils launch party. I want to be kind and brief. I do not want to lie about being busy. I want to leave the door open for actual family events. Three or four sentences.
Decline emails are where the AI's tendency toward warmth actually helps. It tends to soften the blow without being dishonest. If a draft feels too soft for your situation, ask for "a touch firmer."
Summarizing the firehose
Modern life is mostly things that are too long. HOA notices that bury the one fee change on page four. Insurance booklets. Articles you saved last Tuesday. The forty-message family group chat about Thanksgiving logistics. A meeting you couldn't attend whose transcript is now in your inbox.
AI is excellent at this. Paste the text into the chat box and ask for a summary. The version of the prompt I keep coming back to is the one that asks for two things at once: the gist, and the action.
The summary prompt that works
Use this almost verbatim. Paste the long thing first, then ask:
Give me three bullet points capturing what this is about, and then tell me the one thing I should actually do or decide because of it. Plain language.
The "one thing I should actually do" clause is the key. It forces the model past summary into action. Most things you skim do not require action. The few that do, you'll now spot immediately.
Once you have the summary, the conversation isn't over. Ask follow-ups in plain language:
- "Is there anything in here that contradicts what you just told me in the summary?"
- "What's the part that's likely to be controversial or worth pushback?"
- "What did the author skip over or hand-wave?"
- "If I only had time to read one paragraph, which one?"
That last one is underrated. Models are better at picking the load-bearing paragraph than they are at writing one.
Learning anything
This is the use case where AI quietly outperforms most other tools, and almost no one talks about it. Three patterns to keep in your back pocket.
Explain it like I'm 12
Use it for medical terms in a doctor's letter. Legal language in a contract. The financial jargon in your 401(k) statement. Anything you've nodded along to without quite getting it.
The "smart twelve-year-old" prompt
Explain "subordinated debt" to me like I'm a smart twelve-year-old. Use a regular-life analogy. Then tell me why someone might care about it in an ordinary household context.
The "smart twelve-year-old" framing works better than "explain simply" because it gives the model a specific reading level and an implicit permission to use analogies. "Then tell me why someone might care" forces relevance.
Quiz me on this
If you're trying to actually learn something (a topic for work, a hobby you're picking up, the vocabulary your new cardiologist keeps using), reading is passive. Being quizzed is not.
The quiz pattern
I just read about how the Federal Reserve sets interest rates. Quiz me on the basics, one question at a time. Wait for my answer. Tell me if I got it right, give a one-sentence explanation, then ask the next question. Five questions total, getting harder.
This works astonishingly well, and it's one of the few uses where the conversational format genuinely beats reading.
The professor walk-through
For when you want depth, not breadth. The persona prompt is silly-sounding but works.
The professor pattern
Pretend you're a friendly professor of personal finance walking me through how mortgage amortization actually works. Start from the absolute basics, build up, and stop after each major idea to check whether I'm with you.
The "stop and check" instruction is what makes this useful. Without it the model lectures. With it, the model paces itself to your understanding.
Money decisions and bills
This is one of the highest-value categories, and it comes with the strongest caveats. AI is extraordinarily good at the "decoding" half of money: reading a statement, comparing two plans side by side, telling you what questions you should be asking your advisor or your HR rep. It is not a substitute for a person whose job it is to give you advice.
Comparing two insurance plans
Here are two health insurance plans my employer is offering for next year. I'm 52, generally healthy, two prescriptions, expecting one minor surgery. Compare them. Tell me three scenarios where Plan A wins, three where Plan B wins, and the questions I should ask HR before deciding. Plain language.
The "three scenarios each, then questions" structure is what makes this actionable. A plain comparison is information; a scenario-and-question list is a decision aid.
Decoding a statement you don't understand
I'm looking at my brokerage statement. There's a line called "wash sale adjustment" for $412. I do not know what that is. Explain what it means, why it might be on my statement, and whether it's something I need to do anything about.
The model can identify the term and explain the typical meaning. It cannot tell you what's actually true about your specific statement. For that you call your broker, but now you can call them with informed questions instead of confusion.
The "what should I be asking?" prompt
I'm meeting with a financial advisor next week to talk about rolling over an old 401(k). I've never done this before. What are the five most important questions I should ask, and what are the answers I should be skeptical of?
This is the highest-leverage version of money prompts. It turns AI into a coach for the conversation you're about to have, not a replacement for it.
Healthcare conversations
I want to be careful here, because the temptation with AI and medicine is to ask it for a diagnosis. Don't. That's not what it's good at, and the failure mode is bad.
What it is very good at is helping you be a more informed patient. The fifteen minutes you get with a specialist are worth more if you walk in prepared. AI can help you walk in prepared.
Preparing for an appointment
I have a first appointment with a rheumatologist next Tuesday for joint pain in my hands and fingers. I'm 58. Help me prepare. What information will the doctor likely want from me, what questions should I have ready to ask, and what should I be paying attention to in the days before so I have good answers?
The "what should I be paying attention to" line is the gold. It often surfaces things that would not have occurred to you (when does the pain peak, what makes it worse, family history specifics) that the doctor will absolutely ask about.
Decoding a lab or diagnostic letter
My doctor's portal sent me a note that says "thyroid panel within normal range, TSH 4.1, recommend recheck in six months." I do not know what TSH is or whether 4.1 is good. Explain what this means in plain language, and what I should ask about at the recheck.
You'll get a useful explanation. You'll also get a list of follow-up questions you can bring to the recheck. That's the whole win.
Questions for a specialist
I've been referred to a cardiologist for a possible workup. I am healthy and active but my GP heard a heart murmur. What are the most important questions to ask the cardiologist, and what tests are typical at a first visit so I'm not surprised?
The point is not to second-guess the doctor. It's to walk in with the questions you'd otherwise think of in the parking lot afterward.
Planning trips and events
Travel is the use case where the conversational format earns its keep. Plans change. Restaurants close. The kid gets a cold the day before. AI can replan as fast as the situation does.
An itinerary you can actually use
Plan a four-day trip to Lisbon for two adults in their 60s. We walk fine but not for hours uphill. We like food, history, and quiet neighborhoods more than nightclubs. We're staying near Rossio. Give me a day-by-day with two or three things per day, lunch and dinner ideas, and a rough sense of distances. Leave one afternoon mostly open.
The "leave one afternoon open" instruction is what separates a usable itinerary from an aspirational one. Cross-check restaurant names and opening hours before you book; that's the part most likely to be stale or invented.
Plan B for when X is closed
We're in Porto and the Livraria Lello bookshop is closed for a private event today. We had two hours blocked for it. What's nearby that fits the same vibe (charming, walkable, ideally indoors because it's raining), and could we reasonably move our 1pm lunch reservation to fit?
This is where a chat in your pocket beats a paper itinerary every time.
Packing list with constraints
Help me pack for ten days in Iceland in late September, mostly day hikes and one fancy dinner. I'm bringing one carry-on and one personal item, no checked bag. List by category, and call out the three items most people forget.
The "three items most people forget" trick reliably surfaces the headlamp, the second pair of liner socks, the small dry bag.
Dietary restrictions in another country
I have celiac disease and we're going to Tokyo for a week. I do not speak Japanese. Help me. What foods are usually safe, what's hidden gluten I should know about, and write me a short card I can hand to a server in polite Japanese explaining my situation.
The translation card is one of those small things that makes the whole trip easier. Have a native speaker confirm the card before you rely on it for anything serious.
Cooking and meal planning
Cooking might be the most underrated everyday use. The chat in your kitchen replaces ten different recipe-site searches and most of the back-and-forth with whoever in your house cooks the most.
The fridge inventory prompt
Here's what I have: half a rotisserie chicken, a bag of spinach that's about to turn, half a block of feta, a lemon, an onion, garlic, eggs, rice, a little leftover white wine. What can I make for two people in 30 minutes? Give me three options at different effort levels.
"Three options at different effort levels" is the move. You'll pick the medium one, but knowing the lazy and the ambitious versions makes the medium one feel right.
Substitutions on the fly
The recipe calls for buttermilk and I don't have any. I have whole milk, half a lemon, plain yogurt, and heavy cream. What's the best substitute given what I'm making (a quick cornbread for chili tonight)?
Telling it what you're making changes the answer. The buttermilk substitute for cornbread is different from the one for fried chicken brine.
Dietary scaling
I'm making a lasagna for eight, but my daughter-in-law is gluten-free and my brother-in-law is vegetarian. Help me adapt one base recipe so I'm not making three lasagnas. Tell me where the swaps go and what to watch out for.
The "watch out for" part catches the things you'd otherwise discover at 6pm when guests arrive.
A week of dinners on a budget
Plan five weeknight dinners for a family of four. Budget is around $90 for the week. We're not picky but we don't love mushrooms. I want one make-ahead, one slow-cooker, three quick. Give me a single shopping list at the end, organized by section of the store.
The grocery-list-by-section step is the part that'll change your relationship with the supermarket.
Parenting and kids
This category needs the most care, because the temptation is to use AI to short-cut the part of parenting that's actually the parenting. Don't have it write your kid's essay. Don't have it answer their homework. Use it to help you help them.
Homework, the right way
My eighth-grader has written a draft essay arguing that social media is bad for teens. I don't want you to rewrite it. I want you to read it and give me, the parent, three questions I could ask her that would help her see where the argument is weak, without me telling her the answer.
This is the prompt that actually helps a kid grow as a writer. The model gives you Socratic questions, not the kid a finished paragraph.
The hard conversation
Our golden retriever is going to be put down this Friday. We have a six-year-old and a nine-year-old. Help me think through how to talk to each of them about it. What words to use, what to avoid, what to expect after, and what's actually age-appropriate at six versus nine.
You're not outsourcing the conversation. You're rehearsing for it. Big difference.
Conversation starters with a teenager
My 15-year-old son barely talks to me. He's not in trouble; he's just teenage. Give me ten low-stakes conversation openers that aren't about school, aren't about chores, and aren't transparently parental "are you okay" attempts. Things he might actually answer.
Most of them won't work on your particular kid. Two or three will. That's a good rate.
Age-appropriate explanations
My seven-year-old asked why some kids at her school have two moms. Help me with a short, accurate, kind explanation that fits a seven-year-old's worldview, doesn't oversimplify, and gives her room to ask more.
For any topic where you want to get the words right the first time, this is the move.
The patterns that keep working
If you only remember a handful of things from this whole guide, remember these. They are the habits I keep watching pay off across every category above.
- Specific beats vague. "Help me write an email" gets a generic email. "Help me write a short, warm email to my pediatrician's office thanking the receptionist who fit us in last Tuesday" gets you something usable. Specifics are not extra work; they're the work.
- Audience plus goal plus tone. Three short clauses, every time. Who is it for, what should it accomplish, how should it feel. If your draft comes back wrong, the missing one is almost always tone.
- Iterate; the first answer is rarely the final one. The model produces a starting point. The good results come from one or two follow-up nudges. Plan for that. You're not failing if you're iterating.
- Edit and resend, don't argue. If a draft is wrong, hover over your previous message and edit it instead of writing a corrective new one. Cleaner result, less back-and-forth, and the chat doesn't get clogged with rejected versions.
- Cross-check anything load-bearing. A name, a date, a number, a procedure, a citation. If you're going to repeat it to a person or use it to make a decision, verify it somewhere else. Two minutes of checking prevents most of the bad outcomes.
The patterns that don't work
The flip side. These are the recurring ways people get burned, and they're avoidable once you've named them.
- Trusting facts you can't verify. Fluent does not mean correct. The most dangerous output is the one that sounds polished and authoritative and is wrong about a name or a number. If you can't verify a load-bearing fact, treat it as a hypothesis, not an answer.
- Pasting personal information. Account numbers, full medical records with your name on them, Social Security numbers, passwords. The chat is convenient. It is not a vault. Strip identifying details before pasting; the model doesn't need your address to interpret your statement.
- Treating it like Google for current events. Most chat models have a knowledge cutoff and will sometimes confabulate recent news. For "what happened this week," use a real news source. For "help me understand this story I just read," paste the article and ask.
- Asking the same chat for a second opinion. The model has an institutional bias toward agreeing with what it just said. Asking "are you sure?" will get you a confident "yes." For a real second opinion, open a fresh chat with no prior context, or paste your question into a different tool entirely.
- Mega-prompts when a conversation would do. Trying to specify every constraint up front in one giant prompt usually goes worse than asking, getting a draft, and refining. The model is built for conversation. Use it that way.
When you're ready for more
If the recipes above are clicking and you find yourself reaching for the chat box on a regular basis, you've crossed the line that most people never cross. From here, the next steps are about depth and discernment.
- The Best Practices primer. Once you've used these tools enough to have your own friction points, the primer for practitioners is worth a real read. It collects the cross-cutting habits that work across every tool you'll ever use, and it's still readable for non-developers.
- The AI Safety guide. If you've started to wonder what's actually private, what gets used to train the next model, and what to do about it, the AI Safety guide is the practical version of that conversation.
- A tool-specific guide. If you've settled on one tool, go deeper. ChatGPT and Claude have their own guides on this site. They cover the features that matter once you're past the basics: voice, projects, custom instructions, file uploads, image generation, and the rest.
The point of getting good at this is not to use AI more. It's to spend less time on the parts of life that always felt like friction, so you have more time for the parts that don't. — TWD
Try one recipe a day for a week. Pick the categories that match the actual shape of your life. The skill is small and stackable, and there's no race.