I keep getting bombarded with ads for speaking courses. And while I do genuinely need to get better at speaking, something about these ones felt a little… off. So I sent a few screenshots to my AI and asked it what the course was really teaching.
The AI traced the lessons back to four kinds of teaching: Dale Carnegie's advice about making people feel important, Chris Voss's negotiation tactics (labeling a feeling and mirroring a few words), improv exercises for responding in the moment, and 2012-era research about body language.
In Issue 19, I wrote about making your AI research instead of guessing. This article builds on that: once you understand what a course is teaching, you can use the research
to build a course around your own goals, schedule, and practice.
1. Find out what the course is really teaching.
That is where I started. I gave my AI screenshots of the syllabus on the sales page, and asked it to audit what the course was promising. I wanted to know what the page explicitly promised, what books, research, or teaching methods it appeared to draw from, and which methods were well supported or needed more scrutiny.
The AI flagged a micro-expression lesson from the syllabus as being poorly supported by research, so I looked closer. I found a 2018 critique of the research behind it, which made me question whether the course was worth pursuing.
This step tells you what the course is selling. And I’m not at all against paying for courses. I want to learn from someone who has done exceptional work in the field, who
can show me how they think and give me feedback on my own work. But if it’s public ideas arranged into a generic syllabus, without clear evidence that the instructor has deeper expertise to offer, I’m going to ask my AI instead.
Here's a course I'm considering. Search the web before answering. Do not fill gaps from memory. If you cannot browse, say so plainly. Give a direct link inline for every source you use.
1. What does the course page explicitly say it teaches? What books, research, teachers, or teaching methods does it appear to draw from? Keep stated facts separate from your inferences, and point to the relevant wording on the course page.
2. Which claims or methods are well supported, and which have weak, mixed, or no evidence? For each, link to the strongest available source and explain how it supports your assessment. Do not manufacture two sides where credible evidence is lopsided.
3. Based on what the course page promises, what would I be paying for: material I could study elsewhere, instructor feedback or access, community, accountability, production quality, or something else? State what the page includes rather than assuming.
Keep the answer concise.
2. Research the subject more deeply.
Now move from auditing the course to researching the subject you want to learn. Start with the source list from the audit, but do not treat it as complete. Look for primary sources where they exist, current guidance, and experienced practitioners. Find out how the subject is taught or practiced, where credible people disagree, and which parts of the course are worth keeping, dropping, or replacing. The course you audited is a starting point, not a curriculum you need to follow.
The course audit gave us a starting list, not a complete curriculum. Now do a deeper web-research pass on [SUBJECT] before we build anything. Treat the audit list as a starting point, not the full scope of the research.
Search beyond the course materials. Find primary sources where they exist, plus current guidance and knowledgeable practitioner sources. Distinguish evidence from practitioner preference. Find how people build competence in this subject,
where credible experts disagree, and which popular methods have weak or mixed support.
Open the original page before citing it. Label each source as a primary or original source, a reliable secondary source, a practitioner source, or a weaker lead. For each finding, give me the direct source link, publication date or say if it is undated, what it changes about the course, and any uncertainty or disagreement.
At the end, show me what to keep, drop, or replace from the course we audited. Do not write the syllabus yet.Before you build the course, open the sources that would change your plan. You do not need to read everything, but check the research you are relying on to keep, drop, or replace part of the course.
3. Turn the research into a course you can finish.
Keep the same chat open, or paste the research into a new one. Tell your AI how
good you want to get, how long you have, and how many minutes you can give it
each day. Describe the finish line as a task someone could see you do, not
"understand negotiation" or "be better at speaking." "Handle a difficult
renewal call" or "start a conversation without freezing" gives the course a
real target.
I made my own four-week speaking course with a daily limit of ten minutes.
Day 1 is easy on purpose: make eye contact and nod once. That’s it. It takes
under thirty seconds. The harder sections arrive in the final week.
Now build me a course from the deeper research you just did.
Subject: [what I want to learn]
By the end, I want to be able to: [a specific, observable task]
Time available: [two weeks / a month / three months], with up to [N] minutes a day.
If that goal, timeframe, and daily time limit do not fit together, say so plainly and propose a narrower goal or longer plan.
Use the sources you linked, not a generic curriculum. Build the plan around the practice methods the research supports. Leave out material you found weakly supported. If the evidence is unclear or credible practitioners disagree, say where.
Give me one specific action per day, sized to the time I gave you. Group the plan into stages, and briefly explain why each stage comes when it does. For each practice task, say how I will know I completed it and where feedback will come from. If a task needs another person, a coach, equipment, or any other resource, say that plainly.
Show which source or research finding shaped each stage.
Before you accept the plan, read Day 1, a day in the middle, and the last day. Each should ask you to do something, not just name a topic. If Day 1 feels too big, or the last day asks you to do something the earlier days have not prepared you for, send it back and ask for a smaller start, more practice, or a later challenge.
4. Make it fit your life and save it outside the chat.
Tell it what you already know, what you have tried, where you get stuck, and what you avoid. Be specific and unflattering. Leave out confidential work, health information, and identifying details unless you are comfortable sharing them with the service you are using.
Then save the course somewhere outside the chat. Ask your AI to create three plain-text files:
COURSE.mdfor the course itself, including the goal, schedule, and source linksPROGRESS.mdfor a running log of what you did each day and how it wentINSTRUCTIONS.mdfor a short note telling the AI how to pick up where you left
off next time
Those files become the record. The .md ending means it is a Markdown file, which is just a normal text file with simple formatting.
Different apps handle memory and files differently, so open or attach them when
you come back. If your workspace lets the AI update the log, use that. If it
does not, ask it to give you the exact line to paste.
Customize this course for me. Here is what I already know, what I have tried, where I get stuck, and what I avoid: [be specific and unflattering, without including confidential or identifying details].
If you have relevant context about me from earlier chats or saved memory, you may use it. List the details you relied on so I can correct anything stale or wrong.
Tell me which days you changed and why. Then make Day 1 small enough that I am likely to start.
Create the contents of three files: COURSE.md with the plan and sources, PROGRESS.md with one line per day in this format: `Day N - date - what I did - what happened`, and INSTRUCTIONS.md for the next session to read first.
`INSTRUCTIONS.md` should tell a future AI session to read `PROGRESS.md`, find my most recent completed day, and give me only the next day's assignment.
Add a short weekly-review instruction to `INSTRUCTIONS.md`: every seven days, read `PROGRESS.md` and adjust the plan if needed.
If you can edit files in this workspace, update PROGRESS.md when I report back. If you cannot, give me the exact line to paste into it.Run the course one day at a time. Start each session by opening the progress log. If you miss a day, record the skip, then continue with the next scheduled day. If you skip three days in a row, treat it as a sign that the next step needs resizing or rethinking. Do not start over from Day 1.
At each weekly review, look at the log: what did you do, what did you avoid, and why? Use those answers to adjust the next week. You might make the assignments smaller, change the situation where you practice, or focus on a different part of the subject.
The 30-second version
Have your AI evaluate a course before you buy it. Ask it to separate what the course says from what it infers, then look at the underlying sources.
Decide whether the price buys public material or something harder to replace: genuine expertise, feedback, community, or accountability.
Research the subject beyond the course before you build a plan. Keep the useful material and replace what is weak or missing.
Build around a clear finish line, the time you have, a daily limit, and a way to get feedback on your practice.
Keep the plan, progress log, and next-session instructions in files you control.
A course is worth paying for when it offers more than information. If it only packages information, AI can turn it into a course that fits you better.
me+machine