How to Talk to AI: The T.A.L.K. Method

Most people treat AI like Google. They type “write my cover letter,” get something generic, and decide AI “just isn’t that good.” They’re typing at it instead of talking to it.

Think about how you’d brief a teammate on a group project. You wouldn’t text three words and hope for the best. You’d explain the assignment, say who it’s for, show what a good version looks like, and point out what to avoid. The conversation is the briefing. AI works the same way.

You don’t need 47 prompts or a prompt-engineering course. You need four letters: T.A.L.K.

The T.A.L.K. method at a glance

Step 0: Pick your AI

ChatGPT, Claude, and Gemini are all good. The one that wins isn't the one that benchmarks highest. It's the one you'll actually open tomorrow. Google person? Gemini. Live in Microsoft 365? Copilot. Want the best writing partner? Claude. Easiest start? ChatGPT. Pick one and stick with it for a month.

T: Talk, don't type

This is the single biggest shift you can make. Typing feels like search. Talking feels like a conversation. You riff, push back, and it keeps up in real time.

You can start free today:

• ChatGPT: the mic icon dictates your words into text; the waveform icon starts a full spoken conversation. About 2 hours a day on the free plan.
• Claude: voice mode is free in the phone app (roughly 20–30 conversations a day). Great for thinking out loud and working through a problem.
• Windows: press Win+H in any text box and start talking. Free, built in, works everywhere. (On a Mac, Dictation is built in too.)

A: Add your context

Most prompts fail because the AI doesn't know who you are or who you're writing for, so it guesses, and those guesses are generic. Give it three things up front:

• Who you are: "COO at ScaledOn."
• Who it's for: "a recruiter skimming 200+ applications."
• Your rules: "no buzzwords, keep it to one page, no em dash, sound like a real person," plus brand voice rules, style references, output conventions, and review checklists once you have them.

Context is the difference between internet generic and yours.

L: Lead with what "good" looks like

Describing a task isn't the same as defining a good result. Three moves:

1. Show an example. Paste something you liked and say "match this." AI is a great mimic and a poor mind-reader.
2. Name the outcome. Not "write a blog post," but "the best listicle on budget places to stay around Portsmouth, NH for car camping families."
3. Make it flag uncertainty. Add "mark anything you're not sure about with [VERIFY]." AI states wrong facts as confidently as right ones.

K: Keep what works

Here's what separates people who get better at this every week from people who start from scratch every time: they save the instructions, not just the answer.

When something works, keep the recipe, not just the dish:

• Easy: a notes doc with your best context blurbs, ready to paste.
• Better: your tool's built-in memory (ChatGPT Projects, Claude Projects, Gemini Gems) via a project. Set it up once with 10-15 examples and it loads every time, so you explain yourself once instead of every session.
• Best: once you're hooked, turn a task you repeat into a saved "skill," a short file that tells the AI how to do it every time.

Try it in five minutes

Go deeper