🏠 All tools

← Using AI When It MattersEspaΓ±ol

What AI Is Great At, and Where It Quietly Fails You

It's genuinely brilliant at some things and quietly terrible at others β€” and it sounds exactly the same either way. Knowing the line is most of using it well.

General guidance for using any AI assistant β€” not a review of any specific tool, and not legal, medical, or financial advice. Written about AI in general.

The trap with AI isn't that it's bad β€” it's that it's uneven, and gives no sign of which mode it's in. The same calm, fluent voice that explains a contract clause perfectly will, a moment later, invent a court case that never existed and cite it just as smoothly. So the skill isn't trusting it or distrusting it across the board. It's knowing what it's genuinely good at, what it only looks good at, and treating each differently.

The one idea: AI is a superb explainer and drafter, and an unreliable source of fact and judgment. Lean on the first. Verify β€” or withhold trust on β€” the second.

What it's genuinely great at β€” lean in

These play to what it actually is: a tool that's read enormously and is fluent with language. Here it saves you real time and money.

Explaining things in plain words

A diagnosis, a legal clause, a bill, jargon β€” re-explained at your level, as many times and as simply as you need, with no one sighing at you. One of its best uses.

Drafting and rewriting

A complaint letter, an email, an appeal, a summary, a tough message β€” it gives you a strong first draft to edit, and fixes tone, length, and clarity. (Then you check the facts in it.)

Brainstorming and preparing

Questions to ask your doctor or lawyer, options you hadn't considered, pros and cons, a plan to react to. It's a tireless thinking partner for laying out the landscape.

Working with text you give it

Summarizing, comparing, or reformatting a document you paste in is far safer than asking it for facts from memory β€” it's grounded in your material, so it invents less (though still check).

Practice and role-play

Rehearse a hard conversation, a negotiation, an interview, or a difficult call β€” it'll play the other side and let you try again. Low stakes, high value.

Where it quietly fails β€” verify, or don't trust

These run against the grain of what it is. It will still answer β€” fluently, confidently β€” and that's exactly the danger.

Specific facts, numbers, names, citations

It fabricates these ("hallucinates") and states them as surely as real ones β€” invented court cases, wrong dosages, fake statistics, plausible names. Worst in law, medicine, and finance.

Anything recent or fast-changing

It may be out of date or simply guessing about current prices, rules, events, or releases. Check live, authoritative sources for anything time-sensitive.

Telling you hard truths

It leans toward agreeing with you and validating your plan (it's built to be agreeable). If it tells you exactly what you hoped, that's not confirmation β€” that's the default.

Deep or niche expertise

The further from common knowledge, the more confidently wrong it tends to be β€” and the harder it is for you to notice the error. Raise your guard as the topic gets specialized.

The judgment that's actually yours

What you should do β€” about your health, money, family, risk β€” depends on your values and your life, which it doesn't share or carry the consequences of. It can lay out options; the decision is yours.

Knowing when it's unsure

It rarely says "I don't know." It will produce a confident answer rather than admit a gap, so its certainty tells you almost nothing about whether it's right.

The one tell to remember: its tone is not evidence. AI sounds exactly as confident when it's fabricating as when it's correct. Judge how much to trust an answer from the task and the stakes, never from how sure it sounds.

Getting the best of it

Lean on it forNever take at face value
Explaining, simplifying, translatingFacts, figures, citations, names
First drafts and rewritesAnything recent or fast-moving
Brainstorming, prepping questionsSpecialized or niche topics
Summarizing text you provide"What should I do" β€” the call is yours
Practice and rehearsalAnything it sounds very sure about
  1. Ground it in your own material. "Summarize this," "explain this clause" beats "tell me about…" β€” it invents less when working from text you give it.
  2. Ask for sources, then check them. Make it cite, then confirm the sources exist and actually say what it claims. It sometimes invents citations too.
  3. Make it push back. "What would make this wrong? What am I missing? Argue the opposite." This counters its urge to please you.
  4. Keep the roles straight. Facts β†’ verify. Judgment β†’ yours. Drafting and explaining β†’ lean in. Run the trust check when the stakes are high.

The close

Used for what it's good at, AI is one of the most useful tools an ordinary person has ever had β€” it can demystify the systems that used to need an expensive professional just to translate. Used for what it only looks good at, it's a confident stranger guessing about your health, money, and rights. The whole skill is keeping those two straight, and never letting a fluent tone stand in for a checked fact. Lean in hard where it's strong; verify like your life depends on it where it isn't β€” because sometimes it does.

Free and public domain (CC0) β€” copy it, translate it, share it. No accounts, no tracking; it runs entirely in your browser. Practical judgment for using AI well, not professional advice; for high-stakes matters, confirm with a qualified human.

Last reviewed: June 2026. This is general information that can age β€” verify time-sensitive specifics (laws, numbers, programs, app menus) against current official sources.