How to use these prompts

The mistake is going in cold: asking for the result right away without having given the AI the picture. It answers generically because it doesn't know who it's writing for or what matters to you. The prompts below are given once at the start of the conversation (or saved in your assistant's personal instructions, if it has them): from there on, every answer is calibrated to your project, not to the average.

A note on where to put them: the most common assistants have a "personal instructions" area or "projects," where these settings stay valid across all chats. If you can't find that area (menus change often), paste the starter prompt as the first message of the conversation: the effect is the same, it lasts for that session.

The prompt library

The complete starter kit

Before we begin, here's the context. Keep it in mind in all your
answers in this conversation:
- Who I am: [role, field, level of expertise on the topic]
- The project: [what I'm doing, what the final goal is]
- Who it's for: [who will read or use the result]
- How I want answers: [length, tone, with or without examples]
- What NOT to do: [things to avoid absolutely]
When you've understood, just say "ready" and wait for the first request.

Define the tone and style once and for all

For this whole project, write in this style, without my having to
repeat it: [e.g. plain English, short sentences, no over-the-top
enthusiasm, no needless jargon]. I'll paste an example of how I
write: study it and keep this voice.

Example of my style:

Have the AI ask for clarification before starting

I'm about to hand you a project. Before giving me any result,
ask me the questions you need so you're not working blind: what
are you missing to do a good job? Ask me at most five questions, the
most important ones, then wait for my answers before starting.

The project:

Set how results should be delivered

For this project I want results always in this format, so I don't
have to reformat every time: [e.g. title, three key points, then the
text; or a table with these columns]. Apply it to every answer until
I tell you to change. Confirm you've understood the format.

Create the "context document" to reuse

Help me write a context document about this project, that I can
paste to you at the start of each new conversation to pick up where
we left off. Include: goal, decisions already made, constraints,
agreed tone, what's left to do. Keep it under one page.

What we've established so far:

A real example

Riccardo has to write the copy for his garage's website, and every time he opens the AI he gets bloated, Americanized paragraphs he has to deflate by hand. He decides to set up the project. He fills in the starter kit: who he is ("mechanic, I run a neighborhood garage, I talk the way I talk to customers"), the project ("website copy, it has to make people feel they can trust us without overdoing it"), who it's for ("ordinary drivers, not enthusiasts"), how ("short, concrete sentences, no superlatives"), what not to do ("no 'excellence' and 'passion,' no promises I can't keep").

Then he pastes two of his own sentences into the style prompt, taken from an old flyer. From that moment, every piece of copy he asks for comes out already in his voice: lean, direct, without the catalog makeover. At the end he uses the last prompt to have it create a half-page context document, which he saves and re-pastes the next week when he picks things back up. He doesn't start from scratch: he restarts from where he was. The five initial minutes saved him an hour of polishing on every piece of copy.

When it does NOT work (and how to fix it)

If the AI "forgets" the instructions halfway through the conversation

In long chats the initial settings get diluted and the AI reverts to its default tone. Re-paste the starter kit or recall the specific rule: "remember: short sentences, no superlatives." Better still, save the settings in the personal-instructions area, if your assistant has one: there they don't get lost between one message and the next.

If there are too many instructions and the AI ignores some

A starter kit with twenty rules gets five followed. Keep the essential constraints, three or four, the ones that really matter. Enforce the minor rules as needed, in the moment. A few strong rules beat many weak ones that cancel each other out.

If every new conversation starts from scratch

Memory between different chats is limited and not always on: don't take it for granted that the AI remembers yesterday's project. The remedy is the context document: keep it yourself in a file, and paste it at the start of every new session. True continuity is a document you own, not the internal memory of a tool.

A tip from someone who actually uses it

Build your starter kit once for recurring projects and save it as text to copy. Don't rewrite it every time: keep it in a note, improve it when you notice the AI always getting the same thing wrong (add the rule that prevents it). Within a few weeks you have a road-tested starter prompt that gets every project off the ground already aligned. It's the investment that pays back the most, because it amortizes over every single later request.

Frequently asked questions

Do I have to give all the context every time, or does the AI remember it?

It depends on the tool and where you put the instructions. In the personal-instructions area or in projects, the settings stay valid across chats. In a normal conversation, they apply only there and it's worth re-pasting them. The safe rule: don't rely on automatic memory for the things that matter, keep them in a document of your own to re-paste.

How much context is too much?

When you start seeing the AI ignore some instructions, you've given too many. Useful context is the kind that actually changes the answer: who you are, who you're writing for, what to avoid. Details that don't affect the result are noise that dilutes the important rules. Better half a page on target than three exhaustive pages.

Isn't setting all this up at the start a waste of time?

It's the opposite, and you see it on the second use. Five minutes of setup pay for themselves by the third request, when you get answers you don't have to correct. Whoever skips the setup redoes the same polishing work on every answer, for the whole length of the project. You don't save the time by skipping the setup: you spend it anyway, in installments, correcting.