How to train your AI

The setup that makes your AI actually sound like you

Most people start with a prompt. That's already the wrong move.

Everyone wants to know the best prompt. The magic sentence that makes AI produce great content instantly. There is no magic sentence. There is only configuration.

→ A prompt is the last step, not the first.

Before you type a single prompt, you need to do the work that makes the AI output worth reading. It means teaching your AI who you are, how you think, what you stand for, and your point of view. So AI can actually help you, instead of just producing something that sounds vaguely like marketing.

And before you say it, the tool is no excuse. Claude, Gemini, ChatGPT, Copilot. All of them can be trained. All of them respond to good configuration. Been there, done it.

Here's how to do it.

Step 1. Configuration is your foundation, not the prompt

Out of the box, every AI tool is generic. It has no idea what you sound like, what you care about, or who your readers are. So it defaults to average. Polished, confident, completely interchangeable average.

The fix is configuration. Feed AI what it needs to perform.

That means settings and documents. See this as the onboarding file for a new team member who is very fast, very literal, and has no common sense yet.

Configuration works in two ways. First, you set up your profile inside the tool, so it knows how to behave. Then you upload your files: brand docs, tone of voice guidelines, and writing samples.

Most AI tools have one or two profile fields in the settings. Use them.

Your profile settings cover two things.

Who you are. Your name, role, experience, and what you work on. Include your goals, how you communicate, and what you care about. Add your point of view. Treat this as a brief bio. Not a CV, not a corporate summary. The real you, in plain language.
How you want it to respond. Define the tone, the length, the reading level. Tell it to treat you as an expert and skip the basics. Tell it to have an opinion.

Tell it what to avoid. Jargon, passive voice, moral lectures. The more specific, the better.

Your files cover the rest.

Your brand identity. What you stand for, how you position yourself, and what you don't.
Your tone of voice, for yourself or your brand. Specific rules. Short sentences or longer? Do you use "I" or "we"? What words make you cringe?
Your audience. Who reads your content, what they already know, and what they don't have time for. For example, use ‘Writing for busy readers’ to enhance readability of your copy.
Your content structure. How you open, the order, and an indication of the length. For example, I always open with ‘the bottom line first’.
Your lexicon. Words you use. Words you never use. Terms so overused they've lost all meaning.
Your examples. Upload pieces that sound unmistakably like you. Not as templates to copy. As a reference for how you think and how you move through an argument.

This is not a one-hour job. A solid configuration takes days, sometimes weeks. For my client work, I typically build 15 to 20 documents before I let AI near a single piece of content.

Step 2. Vague prompts are where time goes to die

Most people prompt badly. Not because they can't write, but because they're too vague. And vague prompts are where time goes to die.

Once your AI is properly onboarded, the quality of your output depends on what you ask for. A good prompt tells AI exactly what to do, in what format, for whom, and why.

Be specific about the task. Not "write a social post," but “write a LinkedIn post of 150 words that leads with a provocation, aimed at a marketing director under pressure to prove ROI, and ends with a question.”
Set the constraints. Reading level, length, tone, structure. If you don't set them, AI picks them. And AI's defaults are rarely yours. Tell it what to avoid: no passive voice, no bullet points when prose works better, no sentences that start with “As the industry continues to evolve.”
Break the work into small steps. Don't try to get everything done in one prompt.

Good prompting is a skill. The more time you spend on your prompt, the better the response. Your prompts will get sharper with practice. And if you've set your configuration well, you can get away with simpler prompts.

Learn to prompt well. It is a skill. Treat it like one.

Step 3. No feedback loop means correcting the same mistake over and over again

This is the step most teams skip. They configure, they prompt, they publish. And they assume that if the output is good enough, the job is done. No good hire improves without feedback. AI is no different.

AI doesn't learn from your corrections automatically. If you edit something and don't feed that feedback back into your configuration, you'll be correcting the same mistake next time. And the time after that.

Feedback loops are how you raise the floor, over time.

Update AI's memory on the spot. It can be as simple as telling Claude: “Put this in your memory so I don't need to correct it next time.”
Build your review agent. My own agent checks for reading level, scanability, and whether the copy sounds like me. Another checks if my content resonates with a CEO or CMO. Rigorous, blunt, and worth every minute of the setup.
Update regularly. AI tools change. Your brand evolves. Your audience shifts. A configuration that worked six months ago may feel slightly off today. A monthly review of your core settings keeps you ahead of that drift.

The bottom line

Training your AI is not a technical task. It's a strategic one.

It requires the same clarity of thinking you'd bring to briefing a new agency, onboarding a senior hire, or writing a brand strategy. The difference is that AI has infinite patience, zero ego, and will follow your instructions to the letter, which means your instructions need to be very, very good.

→ Most people skip training their AI. They expect AI to figure it out. Then they wonder why the output feels flat and generic.

Get the foundation right, and AI becomes a genuine asset.

Skip it, and you'll spend more time correcting than creating.