← Back to all free guides
A free guide by Mariah Brunner ↗ INSTAGRAM ↗ TIKTOK

Train Your Brain
To Get Good At AI

Smooth practice feels productive and teaches you almost nothing. Five ways to practice so it sticks, plus the 30-day plan.

Here’s the finding that should change how you practice: how well you do while you’re learning and how much you actually learn are two different things. They can even move in opposite directions.

That comes from decades of work by Robert and Elizabeth Bjork at UCLA on desirable difficulties. Make practice smooth and you look great in the moment but forget it later. Make it harder in specific ways and you look worse today, but you keep it.

You’ve probably felt this. You watch a tutorial, copy the prompt, get a great result and feel like you learned something. A week later you can’t rebuild it. You watched someone else learn it.

The Five Moves

01

Train one thing, not everything

Your brain adapts to what you ask it to do over and over. Ask it to do twelve things and you build twelve shallow paths.

How to do it: pick one capability and commit to twenty reps. Writing ads. Reading ad data. Building one agent. Turning a call into a plan. If you can’t name your one thing in four words, you haven’t picked one yet.

02

Keep the struggle in

This is the generation effect. Trying something before you see the answer builds a stronger memory than reading the answer first, even when your attempt is wrong.

How to do it: before you copy anyone’s prompt or skill, write your own version. Badly is fine. Then compare. Seeing why yours fell short teaches you more than the finished prompt ever could.

03

Space the reps

The spacing effect. The same practice time spread across days beats the same hours crammed into one sitting. It’s one of the most replicated findings in learning research.

How to do it: thirty minutes a day beats a four-hour Saturday binge. The little bit of forgetting between sessions is the point. Pulling back something you half lost is what makes it stronger.

04

Force retrieval

The testing effect. Pulling something out of your head builds the memory. Rereading mostly builds confidence.

How to do it: close the tutorial, open a blank chat and rebuild the thing from memory. You’ll get it wrong and it will feel like failing. That’s the mechanism working. A meta-analysis of 29 studies found that spacing your retrieval practice out, instead of cramming it, boosted retention with an effect size of 0.74, which is large for this field.

05

Turn the gap into the lesson

Reps without review just make you fluent at your own bad habits.

How to do it: when you finally get a result you like, put your first attempt next to your final one and make the AI name what changed. That difference is the lesson, and it’s the only part worth writing down.

The Gap Finder
You are my practice coach. I am learning [MY ONE THING] and I want to understand exactly why my final version beat my first one, so I can skip the bad version next time. INPUTS - My first attempt: [PASTE FIRST PROMPT + ITS OUTPUT] - My final attempt: [PASTE FINAL PROMPT + ITS OUTPUT] STEPS 1. Compare the two prompts line by line. List every meaningful change I made (added context, constraints, examples, format, tone, order). 2. For each change, say what it did to the output, pointing to specific lines in the two outputs. 3. Rank the changes by impact. Tell me which one or two did most of the work. 4. Name the habit behind my first attempt that caused the gap (for example: no audience, no example, vague goal). OUTPUT - A table: Change | Effect on output | Impact (high / medium / low) - One sentence: "Next time, start with ___." - One line I can paste into my reps-log. RULES - Be direct. Do not praise the final version, explain it. - If a change made no real difference, say so. - Only use what is in my two attempts. Do not invent context.

The Trap

Everything on this list feels worse than watching a tutorial. That’s how you know it’s the right list.

The 30-Day Plan

Thirty minutes a day, five days a week, four weeks. One capability the whole time. Each week changes the kind of hard on purpose.

The 30-Day AI Practice Plan
# The 30-Day AI Practice Plan 30 minutes a day, 5 days a week, 4 weeks. One capability the whole time. ## DAY 0: PICK YOUR ONE THING (10 minutes) Finish this sentence in four words or less: "I am getting good at ______." Good: writing cold emails. Reading ad data. Building one agent. Turning calls into plans. Too broad: using AI. Getting better at prompting. Automating my business. Then write down where you are today, honestly: - What can I do right now without help? - What do I always need a tutorial or a copied prompt for? - What would "good at this" look like in 30 days, in one observable sentence? Keep one file called reps-log. Every session gets three lines: what I tried, what broke, what I would do differently. That file is the whole system. ## WEEK 1: ATTEMPT BEFORE YOU LOOK (the generation effect) Rule: you can't look at anyone else's prompt or skill until you've written your own version first. Daily, 30 minutes: 1. Pick one real task from your actual work, not a practice exercise. 2. Write your prompt cold. No searching, no examples. 10 minutes, then stop. 3. Run it. Save the output. 4. Now ask AI: "Here is my prompt and the output it produced. Rewrite the prompt the way an expert at [YOUR ONE THING] would write it. Then list the specific differences between mine and yours, and for each difference explain what it changes about the output. Be blunt." 5. Three lines in the reps-log. End of week: read your 5 comparisons in one sitting and write down the three mistakes you made more than once. ## WEEK 2: BUILD IT FROM MEMORY (the testing effect) Rule: nothing open. No tabs, no notes, no old chats. Daily, 30 minutes: 1. Open a blank chat. 2. Rebuild last week's best prompt or workflow from memory. It will be worse. That's the point. 3. Only after you finish, open the original and compare. 4. Ask: "What did I leave out of my rebuild, and which of those omissions actually mattered to the output?" 5. Three lines in the reps-log. Twice this week, do a cold recall with no building: "Quiz me on [MY ONE THING]. Ask one question at a time, wait for my answer, and tell me what I got wrong before the next question. Ask what a person would actually need to know to do this well, not trivia. After 8 questions, tell me my two weakest areas." ## WEEK 3: CHANGE THE CONTEXT (interleaving and varied practice) Rule: same skill, different situation every day. Daily, 30 minutes: 1. Same capability, new context. For cold emails: a different industry, a different length, a hostile audience, a follow-up, a version for someone who already said no. 2. Halfway through the week, remove your crutch. Half the words. No examples. A tool you don't usually use. 3. Once this week, ask: "Give me five situations where the approach I've been using for [MY ONE THING] would fail. Then have me try the hardest one and critique my attempt." 4. Three lines in the reps-log. ## WEEK 4: SHIP IT AND TEACH IT Rule: it has to leave your computer. Days 1 to 3: use it on something real with a consequence. Send the email. Run the analysis your team sees. Put the agent on a real task. Day 4: teach it. Write the explainer, record a voice memo, or walk a coworker through it. Teaching is retrieval with stakes, and it exposes every gap you've been sliding past. Day 5: the review. Paste your whole reps-log into a chat with this: "Here is my full practice log from 20 sessions learning [MY ONE THING]. Compare my earliest attempts to my most recent ones. Tell me: 1) what specifically improved, with evidence from the log, 2) what mistake I'm STILL making that shows up more than twice, 3) the one habit that produced the biggest jump, and 4) what my next 30 days should focus on, and why. Be direct. Quote the log. I want the pattern I can't see." ## THE RULE THAT MAKES OR BREAKS IT A day where you copied someone's prompt and it worked is not a rep. A rep is a day where you tried, it fell short, and you found out why. Twenty of those beat a hundred tutorials.
What This Won’t Do

This won’t make you good at AI in general. It makes you good at one thing, and one real capability is worth more at work than a light familiarity with nine tools.

It also won’t feel good for the first two weeks. Week two is where most people quit, because rebuilding from memory feels like going backwards and copying a working prompt feels like progress. Knowing that ahead of time is most of what it takes to push through.