Most of the difference between a mediocre AI result and a genuinely useful one comes down to a handful of habits. These are the ones that matter most.

Give it context before instructions

The assistant works from what it can actually see. A board that already holds the brief, the transcript, the reference images and the spreadsheet will produce far better output than a long prompt describing those things from memory.

Put the material in the workspace first, then ask.

Select before you ask

Whatever you have selected on the canvas is what "this", "these" and "it" refer to in your next message. This is the difference between the assistant editing the three cards you meant and it creating three new ones beside them.

The same applies in documents: highlight a paragraph and ask for a rewrite, and it rewrites that paragraph rather than the whole page.

Ask for research on anything factual

Anything with a date on it (prices, tools, events, what is currently popular) is worth researching before building. Content built from research carries names, numbers and specifics. Content built from memory reads generic and slightly dated, which is the most common complaint about AI output anywhere.

Say "research this first" and it will, and it will cite what it used.

Say what good looks like

Four things are worth stating up front: length, tone, audience and format. "Three bullets, plain language, for a client who has not read the brief" gets a usable answer immediately. "Write something about the project" does not.

Correct the thing, do not restart

When output is close but not right, respond to it directly: "shorten this to two sentences", "make this friendlier", "swap the third one for something cheaper". The assistant edits what is there.

Retyping the whole original prompt with one word changed usually produces a second version of the same problem.

Work in one thread

The conversation is context. A long thread where you refine, correct and build gets better as it goes, because the assistant knows what you have already rejected. Opening a fresh chat for every idea throws that away.

Set your preferences once

Anything you find yourself repeating in every prompt belongs in AI settings instead: who you are, what you work on, how you want to be addressed, how you like answers written, and reference facts it should treat as given. Set once, applied everywhere.

Tips and tricks

Ask what is missing. With a board attached, "what am I missing here?" is often more valuable than asking for more content. It is good at spotting gaps in work you have already done.

Ask it to argue against you. When a plan feels too comfortable, asking for the strongest case against it surfaces the risks faster than asking for more detail.

Build in stages. For anything large, ask for the structure first, correct the structure, then ask it to fill in. Fixing a wrong skeleton is much cheaper than fixing a finished board built on one.

Good to know

Everything the assistant makes is ordinary content: editable by hand, reversible with undo, and recoverable from the board's trash for 7 days. You do not need to get the prompt perfect, because nothing it does is hard to walk back.