Using AI WellLearning the Basics · Lesson 5
The Human Infinite Project · Learning the Basics · Lesson 5

Using AI Well: Practical Skills

AI is like a very capable new assistant on their first day: the clearer your instructions, the better the work.

The first four lessons explained what AI is, how it learns, how chatbots work, and what other kinds of AI can do. This lesson is about you: how to actually get good results, how to know when to trust what you get, and when to put the tool down.

None of this requires technical skill. The people who get the most out of AI aren't programmers; they're people who explain clearly, ask good follow-up questions, and check what matters. Those are skills you've been practicing your whole life with coworkers, family, doctors, and repair people.

You'll find real examples throughout: weak requests next to strong ones, so you can see the difference instead of just reading about it.

How to use this page

  • There are seven sections. Each takes about 8 to 10 minutes.
  • Start anywhere. If you only read one, read Section 1 (writing a good prompt). It improves everything else.
  • Keep a chatbot open in another window and try the examples as you go. You'll learn faster by doing.
  • Press Focus at the top to show just one section at a time. Your checkmarks are saved on this device.

Focus mode is on: one section at a time. Press Focus again to see everything.

A prompt is simply what you type to the AI, and the single biggest improvement most people can make is to write it the way you'd brief a capable new helper: say what you want, who it's for, what they should know, and what a good result looks like.

Analogy: the first-day assistant

Imagine a brilliant new assistant on their first day. They're well-read, fast, and eager, but they know nothing about you, your family, your business, or what you like.

Say "Write a letter" and they'll produce a letter, a generic one. Say "Write a short, warm thank-you letter to my neighbor Carol, who watered my garden for two weeks while I was in the hospital; mention her tomatoes, which she's proud of; keep it under 150 words," and you'll get something you can actually send.

The assistant didn't get smarter. Your instructions did.

See the difference

Vague

Tell me about Medicare.

Clear

I'm turning 65 in March and I'm still working, with health insurance through my employer. In plain language, explain what I need to decide about Medicare and by when, and what happens if I miss a deadline. Keep it to one page, and list the questions I should ask my HR department.

The first gets a general encyclopedia entry. The second gets advice shaped to a real situation, at the right length, with a useful next step.

The five ingredients of a good prompt

You don't need all five every time. For anything that matters, run through them:

IngredientAsk yourselfExample phrase
1. TaskWhat exactly do I want done?"Summarize…", "Draft…", "Compare…", "Explain…"
2. ContextWhat does it need to know about me or the situation?"I'm a retired teacher…", "This is for my church newsletter…"
3. AudienceWho will read or use this?"…for my 10-year-old grandson", "…for a bank manager"
4. FormatHow long, what shape, what tone?"Three bullet points", "under 200 words", "friendly but professional"
5. ExampleCan I show what good looks like?"Here's a letter I wrote last year that I liked. Match its style."

Showing examples is a superpower

Describing a style is hard. Showing it is easy. If you want something written "in my voice," paste in a few things you've written. If you want a list formatted a certain way, show one line done right. AI is extremely good at copying a pattern it can see. (Specialists call this "few-shot prompting," meaning a few examples.)

With an example

I'm making labels for my spice jars. Here's the format I want:

CUMIN · Earthy, warm · Chili, tacos, curry

Now do the same for: paprika, oregano, turmeric, cinnamon, and bay leaves.

More habits that help

  • Give it a role when useful: "Act as a patient computer teacher for someone who's never used email."
  • Say what to avoid: "No jargon." "Don't recommend specific brands." "Don't make up statistics."
  • Invite questions: "Before you start, ask me any questions you need answered." This one line often transforms the result.
  • Paste in the real material: the actual email, the actual recipe, the actual error message, instead of describing it.
  • Break big jobs into steps: outline first, then one section at a time.
  • Talk instead of typing if that's easier. Many apps let you speak your request, and a rambling, detailed spoken request often works better than a short typed one.

Try this: the one-line upgrade

Take any request you'd normally make in five words and add one sentence each of who it's for and what shape you want. Compare the answers. Most people are surprised how much difference two sentences make.

Watch out for

  • "Magic prompt" formulas sold online. There's no secret password. Clear, specific, honest instructions beat tricks.
  • Leading questions. "Explain why X is true" pushes the AI to agree. "Is X true? Give me the evidence on both sides" gets a more honest answer. (See Lesson 3, Section 3.)
  • Private details in your context. Give the AI what it needs to help, not your account numbers or passwords.

The first answer is a first draft, not a final verdict; the real value comes from the back-and-forth: asking it to shorten, simplify, explain, push back, or try again until the result fits what you actually need.

Analogy: working with a tailor

A good tailor doesn't hand you a suit and walk away. You try it on. "A little shorter in the sleeves. Looser in the shoulders." They adjust, you try it again, and after a couple of fittings it fits perfectly.

AI works best the same way. Don't judge the tool by the first fitting.

Useful follow-ups to keep in your pocket

If the answer is…Try saying…
Too long"Cut this in half." "Give me just the three most important points."
Too complicated"Explain it like I'm new to this." "Give me an everyday analogy."
Too generic"Make this specific to my situation: [details]."
The wrong tone"Warmer." "More formal." "Less salesy." "Sound like a real person, not a brochure."
Confusing in one spot"What do you mean by 'that phrase'?"
Suspiciously sure"How confident are you? What might be wrong here?"
One-sided"What would someone who disagrees say?"
Almost right"Keep everything the same except the second paragraph. Make that one friendlier."
Just not workingStart a new conversation with a clearer prompt (see below).

Tell it what was wrong, specifically

"That's not right" gives the AI very little to work with. "That's too formal. My sister and I joke around a lot, so this should sound relaxed" gives it exactly what it needs. The more specific your feedback, the faster you get there.

Ask it to help you think, not just produce

  • "What questions should I be asking about this?"
  • "What am I missing?"
  • "Give me three different approaches and the trade-offs of each."
  • "Quiz me on this to see if I understood."
  • "Point out any weak spots in my plan."

When to start fresh

Long conversations can drift (Lesson 3, Section 5). If you've gone back and forth many times and it keeps missing, or you're switching to a different topic, start a new conversation. Take what you learned from the first attempt and write a better opening prompt. It's often faster than fixing a tangled conversation.

Try this: the three-round habit

For anything that matters, commit to at least three rounds: (1) your first request, (2) one round of specific feedback, (3) "What's wrong with this or what have I missed?" The third round is where hidden problems often surface.

Watch out for

  • Pushing it into agreeing with you. If you keep saying "Are you sure?", it may cave and change a correct answer. Ask for evidence instead of pressure.
  • Endless polishing. Sometimes "good enough" is good enough. Set a limit and move on.

How much you should check an AI answer depends on two things: how likely it is to be wrong, and how much it would cost you if it were, so a birthday-card verse needs no checking, while a medication dose, a legal deadline, or a fact you'll repeat publicly always does.

Analogy: the stakes decide the double-check

If a friend tells you a movie starts "around 7," you might just show up at 7. If a friend tells you your flight leaves "around 7," you check the ticket. Same friend, same confidence. Different consequences.

Use AI answers the same way: let the stakes decide how hard you check.

The trust-and-verify grid

Glance over Brainstorms, obscure trivia for fun Verify everything Doses, laws, deadlines, citations, recent news Use freely Card verses, recipe ideas, rewording your own email Spot-check key facts Summaries of your documents, explanations for a decision Cost of a mistake: low → high Chance it's wrong: low → high
The top-right corner is where AI answers need the most checking.

When AI is more likely to be wrong

(From Lesson 3, Section 4.) Specific numbers, dates, names, and quotes; citations and links; obscure topics; recent events; your specific local rules (state laws, your insurance plan, your county's deadlines); and long answers with lots of detail.

How to check, from quickest to most thorough

  1. Ask it to show its sources, then open them. Does the source exist? Does it actually say that?
  2. Search one key fact yourself, ideally on an official or well-known site (a government agency, a major hospital, your insurer, a reputable news outlet).
  3. Ask a second AI the same question. Disagreement is a signal to dig deeper. Agreement is not proof, since both may have learned the same mistake.
  4. Check against the original when it summarized something: read the parts that matter most to you.
  5. Ask a qualified human for anything in the top-right corner of the grid.

Reliable signs vs. misleading signs

Actually reassuring

It cites a source you opened, and the source says the same thing.

It answered from a document you gave it, and you spot-checked.

It admits uncertainty or says "check with your doctor about your case."

Not reassuring on their own

It sounds confident and polished.

It includes precise-looking numbers and names.

It agreed with what you already believed.

Watch out for

  • Checking fatigue. After AI is right twenty times, it's tempting to stop checking. The twenty-first time might be the one that matters. Keep checking in the high-stakes corner.
  • Only checking what surprises you. Mistakes that sound plausible slip through precisely because they don't surprise you.

AI shines when it helps you get started, understand something, or see options, especially on tasks where you can easily judge the result yourself and where a first draft saves real time.

Analogy: the power tool, not the carpenter

A power drill makes a carpenter faster, but it doesn't decide where the shelves go or whether the wall can hold them. The carpenter still measures, chooses, and checks.

AI is the power tool. You're still the carpenter.

1. Drafting

Blank pages are hard. AI makes a starting point you can then shape.

  • Emails you've been putting off: a complaint, a polite "no," a condolence note.
  • Letters to a landlord, insurance company, or city office.
  • Toasts, eulogies, newsletter articles, social media posts.
  • Cover letters and résumé updates.
Try

Draft a polite but firm email to my internet provider. I've been billed $15 extra for three months for a service I cancelled on June 3. I want a refund of $45 and confirmation it's removed. Keep it short and factual.

Your job: make it sound like you, and check every fact and name.

2. Summarizing

  • Long articles, reports, or instruction manuals.
  • Terms of service or a lease: "What are the three things I'm agreeing to that I should know about?"
  • A long email thread: "What's been decided and what's still open?"
  • A transcript of a meeting or a video.

Your job: for anything important, read the original parts that matter to you. Summaries can drop or distort details.

3. Learning

This may be AI's most underrated use. It's a patient tutor that never gets tired of your questions.

  • "Explain how a heat pump works, with an analogy."
  • "I just read this article about interest rates. Explain it at a level a curious non-expert can follow."
  • "Teach me the basics of watercolor in five short lessons, one at a time."
  • "Quiz me on what I just learned."
  • "I don't understand this paragraph from my doctor's report. What do these terms mean?" (Understanding terms is fine; the medical decisions belong with your doctor.)

Your job: for facts you'll rely on, confirm with a reputable source.

4. Brainstorming

Here AI's mistakes matter least, because you're choosing from options, not trusting a single answer.

  • Gift ideas, party themes, names for a pet or a small business.
  • Ways to use leftovers, new twists on a family recipe.
  • Topics for a talk, questions for an interview.
  • "Twenty ideas for keeping my grandkids busy on a rainy day with no screens."

Tip: ask for more ideas than you need, including a few "unusual" ones, then pick.

5. Planning

  • Trip itineraries, packing lists.
  • Weekly meal plans and shopping lists.
  • Breaking a big project into steps: cleaning out a garage, organizing paperwork, planning a reunion.
  • Preparing for appointments: "Help me organize my questions for my accountant."

Your job: check hours, prices, and availability directly. They change, and AI may be out of date.

A few more

  • Rewriting: "Make this easier to read." "Fix the grammar but keep my voice."
  • Translating a letter or a menu, with a native speaker checking anything important.
  • Practicing conversations: rehearse a job interview or a hard talk with a family member.
  • Troubleshooting: "My printer shows this error message. What do I try first?"

Try this: your first week with AI

  1. Day 1: Draft an email you've been avoiding.
  2. Day 2: Summarize a long article, then read the original and compare.
  3. Day 3: Learn something you've always wondered about. Ask for analogies.
  4. Day 4: Brainstorm twenty ideas for something coming up.
  5. Day 5: Plan a meal, a trip, or a project with it.

Watch out for

  • Sending AI drafts unread. They can include wrong details, odd phrasing, or placeholder text like "[Your Name]."
  • Losing your own voice. People who know you can often tell. Edit until it sounds like you.

AI can help you understand and prepare for health, legal, and money matters, but it should never be the one making the final call, because it doesn't know your full situation, can be confidently wrong, and can't be held responsible when things go badly.

Analogy: the medical encyclopedia on your shelf

Families used to keep a thick home medical guide on the shelf. It was genuinely useful for understanding symptoms and knowing what to ask. But nobody sensible used it to decide on surgery or change their heart medication without a doctor.

AI is a much smarter, more conversational version of that book. The same rule applies: great for understanding, not for final decisions.

Why these areas are different

  • The stakes are high and mistakes can be hard or impossible to undo.
  • Details matter enormously: your other medications, your state's laws, your tax situation, the exact wording of your contract.
  • AI only knows what you tell it, and people often don't know which details matter.
  • Accountability: doctors, lawyers, and licensed financial advisors have professional duties, licenses, and insurance. A chatbot has none of these.
  • Rules change: tax limits, drug guidance, and laws are updated, and AI may be working from old information.

The line: preparing vs. deciding

AreaGood: helping you prepareRisky: letting AI decide
HealthExplaining a diagnosis in plain words; listing questions for your doctor; understanding a test report's terms; organizing your medication list to bring to an appointmentChanging or stopping a medication; choosing a dose; deciding a symptom isn't serious; picking a treatment
LegalExplaining what a legal term means; summarizing a document so you know what to ask; drafting questions for a lawyerSigning a contract based on its reading; filing court papers it wrote without review; relying on it for deadlines
MoneyExplaining how an annuity or Roth conversion works; building a simple budget; comparing conceptsMoving retirement savings; buying investments it suggested; tax filing decisions without checking

Emergencies: skip AI entirely

For chest pain, stroke signs, trouble breathing, severe bleeding, thoughts of harming yourself, or any other emergency: call 911 (or your local emergency number). For mental health crises in the U.S., call or text 988. Don't spend precious minutes typing to a chatbot.

A special warning: AI and scams

Scammers increasingly use AI-written messages that look professional: fake "investment opportunities," fake IRS or Medicare notices, fake bank alerts. And some scams claim that an AI has found a guaranteed investment. No AI can guarantee returns. Anyone who says so is selling something.

Do this: the "bring it to the expert" method

  1. Use AI to understand your situation in plain language.
  2. Ask it: "What questions should I ask my doctor / lawyer / advisor about this?"
  3. Write the questions down, and bring them to your appointment.
  4. Let the professional, who knows your full situation, make or confirm the decision with you.

This way AI makes your expensive time with experts more productive, instead of replacing it.

Watch out for

  • "It's just a quick question." Quick questions about doses, deadlines, and signatures are exactly where errors are costliest.
  • Using AI to talk yourself out of seeing someone. If you're asking a chatbot whether you really need a doctor, the safer answer is usually yes.

For people whose eyes, ears, memory, attention, hands, or energy make ordinary tasks harder, AI can be one of the most useful tools ever made, reading aloud, describing the world, captioning speech, breaking tasks into steps, and turning speech into text and back.

Analogy: eyeglasses for the mind and senses

Nobody thinks eyeglasses are cheating. They simply close the gap between what your eyes do and what the world asks of them.

Used well, AI can do the same for many kinds of challenges: not replacing your abilities, but meeting you where you are.

Vision

  • Describe what the camera sees: read mail, labels, and menus; identify products, currency, and colors.
  • Read documents aloud in a natural voice, and answer questions about them.
  • Describe photos shared by family, including who's in them and what's happening.
  • Enlarge and simplify: "Rewrite this in large-print-friendly short paragraphs."

Hearing

  • Live captions on phones and computers for calls, videos, and in-person conversations.
  • Transcripts of meetings, doctor's appointments (with permission), and voicemails.
  • Summaries of long audio, like a church service or lecture recording.

Memory

  • A patient re-explainer: ask the same question as many times as you need, with no judgment.
  • Turn talk into lists: speak your thoughts and ask for a clean to-do or shopping list.
  • Summarize conversations so you have notes afterward.
  • Reminders and routines through phone assistants.
  • Saved memory features (Lesson 3, Section 6) mean you don't have to re-explain your preferences each time.

Attention and focus

  • Chunk it: "Break this into five-minute steps." "Give me just the first step."
  • Shorten it: "Summarize this in three sentences before I read the whole thing."
  • Get started: a rough first draft removes the hardest part, the blank page.
  • Get back on track: "Remind me where I was and what I decided."
  • Make it interesting: "Explain this with an analogy," or "Quiz me as we go."
Try

I have trouble staying focused on long tasks. Help me clean out my hall closet. Give me one small step at a time, and wait for me to say "done" before giving the next one.

Hands, speech, and energy

  • Voice typing for people for whom typing is painful or slow.
  • Speech tools that understand atypical speech better than older systems, and voice banking (Lesson 4, Section 2) for people who may lose their voice.
  • Doing more with less effort for people with chronic fatigue or pain: one request instead of many searches and forms.

Reading and language

  • Plain-language rewrites of dense letters from insurers, government offices, or employers.
  • Translation for people navigating a second language.
  • Dyslexia support: listening instead of reading, and spelling and grammar help while writing.

Try this: tell it how you work best

Most chatbots let you save standing instructions ("custom instructions" or "personal preferences" in settings). Tell it once, for example: "I like short answers with analogies. Break long tasks into small steps. Use large, simple formatting." It will apply that to every conversation.

Watch out for

  • Critical readings. For medication labels, expiration dates, and anything safety-related, double-check with a person or a dedicated, tested tool when you can.
  • Privacy of what the camera sees. Descriptions of mail, screens, and documents mean sending images of them to a company. Check the app's privacy settings.

Some things are better done without AI: when you're sharing private information, when the point is your own effort or voice, when a human connection matters, when you're trying to build a skill, or when simply looking it up or doing it yourself is faster and more reliable.

Analogy: the car and the walk

A car is a wonderful invention. But you don't drive to the mailbox, and you don't drive when the whole point is the walk: the fresh air, the exercise, the neighbor you stop to chat with.

AI is a wonderful tool too. Knowing when to leave it parked is part of using it well.

Seven times to leave AI parked

When it's private.

Passwords, account numbers, Social Security numbers, confidential work documents, other people's secrets. If you wouldn't write it on a postcard, don't paste it into a chatbot.

When the point is that it's from you.

A love letter, an apology to your child, a eulogy for your best friend. AI can help you find words, but people value these because they come from you. Use it as a sounding board at most, and let the final words be yours.

When you're trying to learn a skill.

If you want to become a better writer, cook, or problem-solver, having AI do the work skips the part that builds the skill. Use it as a tutor ("give me a hint," "check my work") rather than a substitute. Researchers are actively studying whether heavy AI reliance weakens the skills people stop practicing.

When you need a human.

Loneliness, grief, conflict, a big life decision. AI can be a comfort in the moment, but it isn't a friend, a counselor, or a community. Call someone.

When a direct source is quicker and surer.

Store hours, your bank balance, the weather, today's game score, your flight status: the official site or app is faster and always current.

When the rules say no.

Some employers, schools, courts, and contests forbid or limit AI use, or require you to disclose it. Know the rules before you start.

When you can't check the result, and it matters.

If you have no way to tell whether the answer is right, and being wrong would hurt, that's a job for a qualified person (Sections 3 and 5).

A question worth asking yourself

Before you open a chatbot, try asking: "What am I hoping to get out of this, and is AI the best way to get it?" Often it is. Sometimes the answer is a phone call, a library, a walk, or ten quiet minutes with a pencil.

Healthy habits

  • Try first, then ask. For things you want to stay sharp at, make your own attempt before asking AI.
  • Notice dependence. If you feel anxious making small decisions without asking AI, that's a signal to step back.
  • Mind the time. Chatting with AI can be absorbing. If an hour disappeared, ask whether it was well spent.
  • Be honest about AI help where it matters: work, school, public writing.

Watch out for

  • AI as a replacement for people. It can be a useful helper. It can't show up at your door with soup.
  • Guilt about using it. Knowing when not to use AI doesn't mean using it is wrong. For countless everyday tasks, it's simply a good tool.

Putting it together

Seven habits of people who use AI well:

  1. They brief it like a capable new assistant: task, context, audience, format, and an example when they can.
  2. They treat the first answer as a first draft and refine it with specific feedback.
  3. They let the stakes decide how hard to check, and always verify doses, laws, deadlines, and citations.
  4. They use it where it shines: drafting, summarizing, learning, brainstorming, and planning.
  5. They use it to prepare for experts, not replace them, on health, legal, and money decisions. In emergencies, they call 911.
  6. They shape it to how they work best, especially if vision, hearing, memory, or attention make tasks harder.
  7. They know when to leave it parked: private matters, personal words, learning a skill, and moments that need a human.

Check yourself

Try to answer each question in your head, then tap to see a sample answer.

1. Name the five ingredients of a good prompt.

Task, context, audience, format, and an example.

2. What single sentence can you add to a prompt to get better results on complex tasks?

"Before you start, ask me any questions you need answered."

3. Your first answer is too formal. What's better than saying "That's not right"?

Specific feedback, like "Make it warmer and more relaxed. My sister and I joke around a lot."

4. What two things decide how carefully you should check an AI answer?

How likely it is to be wrong, and how costly a mistake would be.

5. If two different chatbots give the same answer, is it proven correct?

No. Agreement isn't proof; both may have learned the same mistake. Check an authoritative source for anything important.

6. What's the difference between a good and a risky use of AI for health questions?

Good: understanding terms and preparing questions for your doctor. Risky: letting AI decide doses, treatments, or whether a symptom is serious.

7. Give two ways AI can help someone who has trouble focusing.

Any two of: breaking tasks into small steps, short summaries before long reading, a first draft to get started, reminders of where they left off, analogies to make things interesting.

8. Name three situations where it's better not to use AI.

Any three of: private information, personal words that should come from you, learning a skill, moments that need a human, when an official source is quicker, when rules forbid it, when you can't check the result and it matters.

Glossary

Custom instructions
Standing preferences you save in a chatbot's settings so it applies them to every conversation.
Few-shot prompting
Including a few examples in your prompt so the AI can copy the pattern.
Follow-up
A later message that refines, questions, or redirects an AI's answer.
Hallucination
A confident but false AI output (see Lesson 3).
Leading question
A question worded to push toward a particular answer.
Live captions
Real-time text of spoken words, shown on screen.
Plain language
Writing that is clear and easy to understand on first reading.
Prompt
What you type or say to an AI: your request and instructions.
Role prompt
Asking the AI to act as a particular kind of helper, like "a patient teacher."
Verification
Checking an AI's answer against a reliable source or a qualified person.
Voice typing
Speaking instead of typing, with speech turned into text.