Past the easy wins

You use one most days. Small things — tidy this up, draft that, what’s another word for. It’s useful. You also have the nagging sense that people getting far more out of this know something you don’t.

They mostly know five things. None of them are tricks.

Give it standing context, not context per message

The single biggest difference between casual and serious use is that serious users stop re-explaining themselves every time.

Every major assistant now has somewhere to put material that persists across conversations — reference documents, a description of your business, how you want things written, what your products are called. The names differ and will keep changing; the capability is standard. Find it in whatever you use.

Put in the things you find yourself typing repeatedly: what your company does, who your customers are, the words you use and the words you avoid, an example of writing you were happy with. Ten minutes, once.

The change is immediate and larger than any phrasing improvement. Most “AI writing sounds generic” complaints are really “the tool has never been told anything specific,” and no amount of clever prompting substitutes for it knowing what you sell.

Have longer conversations — and know when to abandon one

Casual use is one message, one answer, mild disappointment. Serious use looks like a conversation with a capable colleague who cannot read your mind: closer, but too formal. Cut the second paragraph. Now say why it matters to them specifically.

Three or four rounds of that beats any single well-crafted instruction, and it is faster than writing the thing yourself.

But know the failure mode, because it is not obvious: a conversation that has gone wrong stays wrong. Once it has produced a bad answer, that answer is sitting in the context influencing everything after it. If you find yourself correcting the same mistake twice, stop. Start a new conversation and put the good version of your request in first. You will get there faster than by arguing.

Rough rule: keep going while you are refining one thing. Start fresh when you switch tasks, or when it has become stubborn.

Ask it to argue with you

This is the capability most people never touch, and it is where the value stops being about typing speed.

“What’s the strongest objection to this?”
“What did I not ask that I should have?”
“Read this as the customer receiving it. What would put you off?”
“Where is this argument weakest?”

You are not asking it to be right. You are using it as a first reader who is never too busy and never worried about your feelings. Half of what comes back will be generic. The other half will catch the thing you would have found out about later, from someone who mattered.

This works far better than asking it to improve something, because “improve” gets you rewording and “what’s wrong with this” gets you thinking.

Learn what competent-but-wrong looks like

At this level you are past being impressed by fluency, and the skill you need next is telling good from merely well-written.

The reliable signal is the same one from what breaks: whether the answer is reshaping what you gave it, or producing something you didn’t.

Two habits that cost seconds and catch most of it:

Check the part you can verify cheapest. If a summary gets the one detail you already know slightly wrong, do not trust the rest of it. Errors are not usually isolated — a confident mistake about the easy part means it was not working from the source as carefully as it sounded.

Ask where something came from. “Which part of the document says that?” If it can quote the passage, it read it. If it paraphrases the answer back at you, or the quote does not appear in your document, it invented it. This takes one message and settles the question.

Try the tasks you’re not trying

Most people plateau because they only reach for it for writing. The higher-value uses are transformations:

Turn a mess into a structure. A rambling call transcript into a list of decisions and who owns each. This is close to the best thing it does.
Rewrite for a different reader. The same update for your team, your accountant, and your biggest customer.
Explain someone else’s document back to you — a contract, a policy, a technical proposal — then ask what it would want changed if the document were arguing against you.
Compare things against criteria you set. Three quotes, three suppliers, three candidates’ cover letters, scored on the things you actually care about rather than the things the documents emphasise.
Interrogate a decision you’ve already made. Describe it, then ask what you would need to believe for it to be wrong.

Notice what these have in common: you supply the material and the judgement, it does the transformation. That is where the reliability is, and it is also where the hours are.

Know when to close the tab

Getting good at this includes recognising when it is the wrong tool:

Arithmetic across a lot of rows. Unless your tool can actually run code, it is pattern-matching numbers, and it will be confidently close.
Anything that depends on current facts — prices, availability, who holds an office, what a company announced. It does not know, and it will not reliably say so.
Anything where you can’t judge the output. If you could not tell a good answer from a plausible one, you have not saved work — you have moved the risk somewhere you cannot see it.
Tasks where explaining takes longer than doing. Sometimes the honest answer is that it is a three-minute job.

The people who get the most out of these tools are not the ones who use them for everything. They are the ones who know what they are for.

Before you go further

Everything on this page is you asking and checking. The moment you start letting a tool act — sending, filing, deleting, posting — the arithmetic changes, and it changes more than people expect.

→ What breaks, and how far it reaches

Found something that belongs here, or disagree with a call? Tell us.