You know the material cold. You've taught it for years, in workshops or one to one, and people have told you to your face that it changed how they work. The course should be the easy part. Instead it's been sat in a "someday" folder for eighteen months, because turning what's in your head into a structured, sellable course is a different job from knowing the subject. It's instructional design, scriptwriting, assessment building, and sales copy. Most experts are good at exactly none of those.
Claude is genuinely good at three of the four, in my experience. The fourth still needs you. What follows is a walkthrough of where it does the work, where it can't, and a worked example you can copy.
Start with the outcome, not the outline
The mistake almost everyone makes is starting from "what do I know" and dumping it into modules. That produces a course that's comprehensive and unsellable, because nobody buys comprehensiveness. They buy a transformation: who they are at the end that they weren't at the start.
So your first conversation with Claude isn't "outline a course on X." It's a working session to pin down the outcome. Try this:
I teach [subject] to [audience]. I want to build a course. Before we outline anything, interview me: ask me what specific result a student should be able to get by the end, what they can currently do, what's stopping them, and what they've already tried that didn't work. One question at a time.
That one-question-at-a-time instruction matters. It stops Claude generating a plausible answer to its own question and forces it to use yours. By the end you'll have a single sentence: "By the end, a [freelance designer] can [price a project with confidence and defend the number] even if [they currently undercharge out of fear]." Every module then earns its place by moving the student toward that sentence, or it gets cut.
This part is pure judgement, and it's yours. Claude can pressure-test the outcome — ask it "is this specific enough to be measurable? what's vague here?" — but it can't decide what's worth teaching. You lived the problem. It didn't.
Curriculum design: where Claude saves you a fortnight
Once the outcome is fixed, structuring the path to it is where AI earns its keep. Give Claude your outcome sentence, your audience, and a brain-dump of everything you'd want to cover — bullet points, voice notes transcribed, whatever you've got — and ask it to sequence that into modules and lessons that build on each other.
Two things it does well here that are tedious by hand. First, it spots prerequisite ordering. It'll flag that you've put "advanced negotiation" before "knowing your own numbers," which is the wrong way round. Second, it right-sizes each lesson, so a module isn't three sprawling hours followed by a forty-second afterthought but a consistent rhythm of digestible chunks.
A prompt that works:
Here's my outcome and my raw material [paste]. Design a 5-module curriculum. For each module: the one job it does toward the outcome, 3-5 lessons, and for each lesson a single learning objective phrased as "the student can…". Order everything by prerequisite. Tell me what I've included that doesn't serve the outcome.
That last line is the useful bit. A good instructional pass cuts as much as it adds. Claude is unsentimental about your favourite tangent in a way you won't be.
What it can't do: it doesn't know which of your war stories land, which exercise once made a workshop room go quiet because it hit a nerve. Mark those in your raw material. The structure is the skeleton. Your specific examples are why anyone stays.
Lesson scripts: a first draft in your voice, not a finished one
Set your expectations here. Claude will write you a full lesson script — intro hook, teaching, example, exercise, recap — in a couple of minutes. It'll be coherent, correctly structured, and about 70% of the way there. The missing 30% is the part only you can supply: the real client name (anonymised), the exact figure, the mistake you made in 2019 that taught you the lesson you're now teaching.
The workflow that produces something genuinely good:
- Give Claude the lesson objective and your bullet points for that lesson.
- Ask for a script "in a spoken, plain style — short sentences, no corporate filler, written to be said out loud, not read." Spoken cadence isn't the same as written, and if you don't ask for it you get an essay.
- Read the draft aloud. Where you stumble or cringe, that's where it sounds like AI and not you. Rewrite those lines yourself, or feed Claude a voice note of how you'd actually say it and ask it to match.
Hand-edit the first script fully, then tell Claude "that's my voice — match it for the rest." It learns the pattern from your correction and the next four lessons land closer. This is the single biggest quality lever and almost nobody uses it. They accept the first draft and wonder why the course feels generic.
One warning. Don't let it invent statistics, studies, or quotes to make a point sound authoritative. It will, cheerfully, if you don't tell it not to. Add "use only facts I've given you; if you need a stat to make a point, flag it and I'll provide one" to your script prompt. Your credibility is the whole product. One made-up figure a student fact-checks is worse than no figure at all.
Quizzes and exercises: better than you'd expect
Assessment is the bit course creators skimp on most, because writing good questions is hard and dull. Claude is strong here. Give it a lesson's learning objective and it'll generate multiple-choice questions where the wrong answers are plausible (that's the hard part — bad distractors make a quiz pointless), short-answer prompts, and a practical exercise that applies the lesson to the student's own situation.
Ask specifically for application, not recall:
For this lesson objective, write one practical exercise where the student applies it to their own [project/business], plus 3 multiple-choice checks. Make every wrong option a realistic mistake someone actually makes, not an obvious throwaway.
That "realistic mistake" instruction is what separates a quiz that teaches from one that's a box-tick. You can still eyeball the output — you'll occasionally spot a question that's ambiguous — but the lift is real and the editing is light.
The sales page: the bit that decides whether any of this matters
You can build the best course on the internet and sell none of it behind a weak page. Sales copy is a craft, and "I'm not a copywriter" is the most common thing course creators say right before they publish three paragraphs about features nobody asked about.
Claude is a strong drafting partner here, on one condition: it can only work from truth you give it. The structure of a course sales page is well understood. The transformation up top, who it's for and who it isn't, what's inside, who you are and why you're credible, proof, the offer, objection handling, a clear call to action. Claude knows that structure. What it doesn't know is your actual student results, your real testimonials, your specific guarantee. Feed it those. Never let it manufacture social proof. A fabricated testimonial isn't just slop, it's the kind of thing that ends a personal brand.
A good prompt anchors on the transformation and the objections:
Draft a sales page for this course [paste curriculum + outcome]. Lead with the transformation, not the modules. Here are three real objections my audience has [list them] — handle each one. Here are two genuine testimonials [paste]. British English, no hype words, sound like a person who's confident in the work, not a marketer shouting.
Then you do the judgement pass. Is every claim true. Is the price right. Does the guarantee match what you'll actually honour.
What the whole thing looks like end to end
Say you're a nutrition coach building "Meal Prep for People Who Hate Cooking." A realistic week. Monday: the outcome interview and a cut-down 4-module curriculum. Tuesday and Wednesday: lesson scripts, where you hand-edit module one into your voice, Claude matches the rest, and you record. Thursday: quizzes and a "plan your week" exercise per module. Friday: the sales page draft, with your real before-and-after client photos and quotes dropped in and your honest 14-day guarantee. What used to be an eighteen-month "someday" becomes a first version you can put in front of ten people and improve from their feedback. The course doesn't have to be perfect. It has to exist and be honest.
Hold onto the division of labour. Claude does the instructional structuring, the first-draft scripting, the assessment writing, and the sales-copy scaffolding — the craft work that isn't your craft. You supply the subject, the stories, the truth, and the call on what's worth teaching and what to cut. That split is why the output is yours and not generic. Hand the whole thing to AI and it shows. Use it to remove the parts that were stopping you from starting, and you ship.
Where Skill Locker fits
I build Skill Locker, and the Course Creation pillar is a set of these workflows packaged so you don't have to engineer the prompts yourself. A curriculum builder, a lesson-script writer tuned for spoken delivery, a quiz and assessment creator, and a sales-page writer that refuses to invent proof. Each one encodes the instructions above, so you just bring your material. It's my product, so take that as you will. But if you've got the expertise and not the eighteen months, that's the gap it's meant to close. New to Claude entirely? Start with the non-developer on-ramp first.
FAQ
Can Claude build an entire online course on its own?
No, and you wouldn't want it to. It does the structuring, first-draft scripts, quizzes, and sales-copy scaffolding well. The subject expertise, real examples, student results, and the call on what to teach versus cut are yours. The good courses come from that split, not from automation.
Will an AI-written course sound generic?
It will if you accept the first draft. The fix is to hand-edit one lesson fully into your own voice, then tell Claude to match that pattern for the rest. The difference between a generic course and one that sounds like you is almost entirely whether you do that step.
How long does it take to build a course this way?
A focused week gets a solo expert to a sellable first version — curriculum, scripts, assessments, and a sales page — versus the months it usually stalls for. You're meant to launch that to a small group and refine, not to perfect it before anyone sees it.
Is it safe to use AI for the sales page?
Yes, for the drafting and structure. Not for inventing testimonials, results, or statistics. Give it only true material and have it handle the wording. Tell it explicitly not to manufacture proof. Fabricated social proof is both dishonest and easy for buyers to catch.
Do I need to know how to code or use the terminal?
No. This is plain-English conversation with Claude. The course-creation work described here is writing and structuring, not programming.
Build the course you've been putting off
The Course Creation pillar packages the curriculum, scripting, assessment, and sales-page workflows into ready-to-run skills, so you supply the expertise and skip the prompt engineering. There are free skills to test the quality before you commit.