Drafting anything
Emails, letters, descriptions, applications, listings, notes to the landlord. A first draft in your hands beats a blank page every time, even when you rewrite most of it.
An honest map.
Where this genuinely helps, where it’s mediocre, and where using it is an actively bad idea. Organised by life, not by feature list.
Green light
Where it earns its keep.
The common thread: language-shaped work, where you can look at the result and immediately know whether it’s good.
Making long things short
Paste in a document, a contract, a wall of meeting notes, a dense article. Ask for the gist, the decisions, or what affects you specifically. You can see the source, so you can check the answer.
Explaining things at your level
“Explain this like I know nothing.” “Explain it like I already know the basics.” Being able to set the level and ask the same dumb question four times without anyone sighing is genuinely valuable.
Organising a mess
A shapeless pile of notes, ideas, or tasks turned into categories. It is very good at finding structure that was already there and that you were too close to see.
Getting unstuck
Twenty names when you need one. Fifteen angles on a problem. The obvious approach you were too tired to think of. Quantity on demand, from which you pick.
Being a patient thinking partner
Talking a decision through out loud, at 2am, with something that never gets bored and never mentions it later. This one surprises people.
Amber light
Useful, with a hand on the wheel.
Real value here, but the failure modes are subtle enough that you need to stay involved.
01
Research and background reading
Excellent for orienting yourself in an unfamiliar subject and learning what questions to ask. Unreliable for specific facts, figures, dates, and citations, which it will occasionally invent with total confidence. Use it to get oriented, then verify anything you intend to rely on.
02
Anything in your own voice
It has a default register — slightly eager, slightly corporate, fond of tricolons. Left unchecked it will sand your writing down into something smooth and anonymous. Give it samples of how you actually write, and keep editing after.
03
Learning a skill
A patient tutor that never judges you is a real gift. But it can also let you feel like you understand something you have only had explained to you. Make it ask you questions, not just answer yours.
04
Images and visual work
Genuinely capable and improving fast. Also where the ethical questions are sharpest, and where the gap between “impressive” and “actually right for this project” is widest. Fine for exploration and reference; think hard before it becomes the finished thing.
Red light
Where I’d genuinely tell you not to.
Not squeamishness. These are the places where the specific way this technology fails causes real harm.
Medical, legal, financial decisions
Use it to understand terminology and prepare better questions for a professional. Do not use it to make the decision. It does not know your situation, it cannot be held responsible, and it will sound equally certain either way.
Facts you will state publicly
If you are going to repeat a number, date, quotation, or claim where being wrong would embarrass you or mislead someone, verify it at the source. Every single time.
Anything you’d hate to see leaked
Treat a mainstream assistant roughly like a work email: probably fine, occasionally reviewed, not the place for someone else’s private information. See the privacy guide for the practical version.
Replacing a person who cares about you
It is a genuinely good thinking partner. It is not a friend, a therapist, or a substitute for someone who will notice if you go quiet for a week. It is designed to be agreeable, which is precisely the wrong quality in the moments that matter most.
Straight answers
What is AI genuinely best at for everyday people?
Anything involving language where you can judge the result: drafting, rewriting, summarising documents you can see, organising messy notes, explaining unfamiliar things at your level, and being a patient thinking partner. These are low-risk and high-value because you can immediately tell whether the output is good.
What should I not use AI for?
Anything where being confidently wrong is costly and you cannot easily check: specific medical, legal, or financial decisions, precise facts and figures you will repeat publicly, and anything requiring current information it may not have. Also avoid it as a replacement for professional judgment or for genuine human contact.
Will AI replace creative work?
It will not replace taste, judgment, or having something to say, which is most of what creative work actually is. It is genuinely useful for the surrounding labour — variations, first drafts, admin, research, naming things, getting unstuck. The concern is legitimate; the practical answer is to use it where the craft isn't.