Chapter 17: Use AI to Learn Any AI Tool
The skill that keeps this book useful
Everything up to here has a shelf life. The tool chapters are already drifting out of date. The interfaces will be redesigned. The plans will be renamed.
This chapter is the answer to that, and it is the reason I wrote the book. By the end of it you will be able to teach yourself any AI tool (including ones announced after this book was printed) without waiting for anyone to write a guide.
It is also, as a bonus, how you learn any other piece of software. More on that at the end.
A more reliable way to learn
The natural instinct is to ask the tool about itself.
How do I use your projects feature?
Try it. You will get a clear, confident, well-organized answer that may describe a version of the feature that no longer exists, may invent a menu that was never there, and may miss the capability you actually needed.
This is not a malfunction. A model has no privileged knowledge of itself. It answers from what was written about products like it before its training cutoff, plus whatever instructions its developers supplied. Product documentation changes weekly; training data does not. And self-description is exactly the kind of thin, fast-moving material where confident errors tend to concentrate.
So the circularity in “use AI to learn AI” is real, and LEARN exists to break it.
The principle underneath the whole method:
Use official documentation as the factual source and AI as the instructor.
Keep those roles separate and this works reliably. Blur them and you get confident instructions for a button that does not exist.
The LEARN method
L: Locate current, official information
E: Explain your objective and experience level
A: Ask AI to teach one step at a time
R: Repeat the task with guided practice
N: Note changes and verify important instructions
L: Locate current, official information
Before you ask the model anything, get it something current to work from.
Go to the source. Every major tool publishes a help center, documentation, and release notes. Release notes are the most underused resource in software: a dated list of exactly what changed, written by the people who changed it.
Get that material into the conversation. Three ways, in order of reliability:
1. Paste it. Copy the relevant help page directly into the conversation. Now the model is working from the actual current text rather than its memory of an older version.
2. Give it the link and require retrieval. Ask it to read the page; then confirm it actually did, rather than answering from memory. The tell is specificity: retrieved answers reference the page’s actual wording and structure.
3. Ask it to search, then check the sources it used.
The instruction that makes this work:
Use only the documentation I’ve given you. If something I ask isn’t covered there, say so rather than filling the gap from memory.
That single sentence converts the model from an unreliable narrator into a competent reader of a reliable document, which is a role it is genuinely excellent at.
E: Explain your objective and experience level
Two pieces of information, both usually omitted.
What you are actually trying to accomplish. Not “teach me projects” but “I manage six clients and I’m re-explaining their background every time I start a conversation. I want that to stop.”
The difference is that a stated objective lets the model tell you the feature is wrong for your purpose, or that a different one fits better. A request to be taught feature X can only be answered by teaching feature X.
What you already know. Your comfort with software generally, whether you have used anything similar, and (importantly) where you get lost. “I’m fine until something asks me to configure permissions” saves an enormous amount of wasted explanation.
Combined:
I’m comfortable with everyday software but I’ve never used a tool like this. I manage six clients and I’m re-explaining their background in every conversation. Using only the documentation above, tell me whether this feature solves that, and what it cannot do.
A: Ask AI to teach one step at a time
The default failure of AI instruction is the wall of text: eleven numbered steps delivered at once, which you read, half-follow, and abandon at step four when your screen does not match.
Prevent it explicitly:
Teach me one step at a time. Give me a single step, then stop and wait. If I say I’m stuck, help with that step before moving on. Don’t give me the whole sequence.
Models comply with this well, and it changes the experience completely. You are now in a guided session rather than reading a manual.
Two supporting moves.
When your screen does not match the instructions, photograph it. Take a screenshot, upload it, and say this is what I’m looking at, where is the thing you just described? This resolves the single most common point of failure in every software tutorial ever written, and most people never think to do it.
Ask what you should be seeing. Before I click that: what should appear afterward? Now you know immediately whether it worked, instead of discovering three steps later that you were in the wrong place.
R: Repeat the task with guided practice
Being walked through something once produces the feeling of understanding and very little of the substance. You need to do it yourself.
First pass: do it with the model watching. Narrate what you are doing and let it correct you.
Second pass: do it cold. A similar task, no help. If you get stuck, note exactly where — that is the part you did not learn, and it is a much more useful thing to know than a general sense of not being confident.
Then ask for the compressed version:
Write me a short reference for this task: the steps, the two mistakes I’m most likely to make, and how to tell if it worked.
Save it. Ten of those and you have a personal manual for the tools you use, written at your level, which is a thing that does not otherwise exist.
N: Note changes and verify important instructions
Two habits, one small and one non-negotiable.
Note what you learned, with the date. Software changes underneath you. A dated note tells you whether your knowledge is six weeks old or two years old, and it is the difference between “I don’t remember how this works” and “this used to work this way; what changed?”
Check release notes occasionally for the tools you rely on. Fifteen minutes a quarter keeps you genuinely current.
Verify anything consequential at the official source. Do not let a model walk you through, unverified:
• Anything involving billing: upgrades, cancellations, refunds
• Anything irreversible: deleting data, closing accounts, clearing history
• Anything about who can see your data: sharing, permissions, connectors, training settings
• Anything about security: authentication, access, recovery
These are exactly the areas where the interface changes often, where a plausible-sounding wrong instruction does real damage, and where the model’s confidence is least correlated with accuracy. Use the AI to understand what the setting means. Use the official documentation for where to click and what happens when you do.
A worked example
The situation. You want to stop your work conversations being used to train a model, and you are not sure whether your setting is correct.
L: Locate. Go to the tool’s help center, find the article on data controls, and paste it into a conversation. Note the date on the page.
E: Explain. I use this for client work. I need to be certain my conversations aren’t used for training and I want to know what’s retained regardless. I’m comfortable with software but I don’t want to guess about this. Use only the documentation I pasted.
A: Ask. One step at a time. Start by telling me which setting controls this and what it does and doesn’t cover. Screenshot your settings screen when the description does not match.
R: Repeat. Change the setting. Then go back and confirm it held: settings occasionally revert or apply only to a scope you did not expect. Ask for a two-line summary of what is now true.
N: Note. Write down what you changed, on what date, based on which help article. Then, because this is a data-handling setting squarely in the non-negotiable category, confirm it against the official page yourself rather than relying on the summary.
Fifteen minutes, and you actually know rather than assume.
The same method works on everything else
Nothing in LEARN is specific to AI tools. It works on any software with documentation and a learning curve.
Spreadsheet functions you have avoided for a decade. The features of your accounting package you have never touched. Your camera’s manual mode. Your practice management system. Tax software. A new phone.
Paste the relevant manual page, state your objective and your level, demand one step at a time, practice twice, and note what you learned. This is, for many readers, more valuable than anything else in this book — because the tools you already own and half-use represent more unclaimed capability than any new tool you might adopt.
The fifteen-minute version
When you just need to get something done:
1. Find the official help page. Paste it.
2. Using only this, and one step at a time: I want to [objective]. I’ve never done it before.
3. Screenshot anything that does not match.
4. Ask for a short reference at the end. Save it.
5. If it touches billing, data, permissions, or deletion: verify at the source.
That is the whole method, and it will still work when every tool in this book has been replaced.
Exercise 17.1: Learn something you have been avoiding
Pick one feature in a tool you already use that you have never understood. Run full LEARN on it.
Time it. Most people are surprised it takes under twenty minutes for something they have avoided for a year.
Exercise 17.2: Prove the L step matters
Ask an AI tool to explain one of its own features from memory. Write the answer down.
Then paste the current official documentation for that feature and ask again.
Compare. The gap between those two answers is the entire justification for step L, and you will not need convincing again.
Exercise 17.3: Use it on something that isn’t AI
Apply LEARN to a non-AI tool you use badly: a spreadsheet function, a feature of your work software, a setting on your phone.
Applying the method outside AI often changes how people approach the software they already own.
Learn More About Artificial Intelligence
Learn more about the CLEAR, TRUST and LEARN methods for Artificial Intelligence