Starvard University · AI Prompting
Get useful work out of AI instead of fighting it. This is the highest-leverage skill of the decade, and it is learnable in a week.
The whole course in brief
Twelve lessons, distilled. Read this and you are already ahead of most people.
No sign-up, no prerequisites. Just you and any AI chat tool.
The twelve lessons
Lesson 1
Forget the sci-fi. A chatbot is a prediction machine trained on a mountain of human writing. When you type a prompt, it does not think or look things up the way you do. It calculates the most likely sequence of words that should follow. That sounds limited, but "most likely next words" on top of that much training turns out to be astonishingly capable at drafting, summarizing, explaining, brainstorming, and translating.
Why does this matter for prompting? Because the model has no fixed knowledge of you, your situation, or the truth. It plays the odds. Prompts that narrow the odds, by giving context, examples, and constraints, get dramatically better answers. You are not commanding a genius. You are steering a very talented improviser.
One more mental model that helps: the AI is confident by default. It does not flag its own uncertainty the way a thoughtful human would. It will invent a citation, a statistic, or a historical fact with the same smooth tone it uses for real ones. Keep that in the back of your mind for every lesson that follows, especially the one about verifying.
Lesson 2
Most bad AI answers come from bad prompts, and most bad prompts share one flaw: they assume the AI knows your situation. "Write me a cover letter" could mean a thousand things. "Write me a cover letter for a warehouse supervisor role at a Denver logistics company. I have four years of forklift experience and I am applying because I want shift work. Keep it under 200 words" is a prompt with a fighting chance.
Use the context sandwich for anything that matters. Three layers. First, the goal: what you are trying to accomplish and why. Second, the context: who it is for, what has already happened, the facts the AI could not possibly guess. Third, the format: how long, what shape, what tone. You can write this in three sentences and it will beat most prompts twice as long.
Here is the muscle to build: before you hit send, ask yourself, "What would a smart human friend need to know to do this well?" Then paste that into the prompt. The AI is not smart about your life. You have to be smart about telling it.
Lesson 3
The most common prompting mistake after vagueness is cramming. People ask the AI to research a topic, write a post about it, translate the post, and make it funny, all in one message. The result is a mushy compromise where nothing gets full attention. The model splits its effort across the tasks and every one of them comes out weaker.
Instead, chain small prompts. Research first, then draft from the research, then edit the draft, then shorten it. Each step gives the model one clear job, and each step gives you a checkpoint to correct course. This feels slower. It is faster, because you stop getting outputs you throw away.
The same rule applies inside a single prompt. Pick the one outcome that matters most and name it first. "The most important thing is accuracy" or "The most important thing is that a 12-year-old understands it." When the model has to trade off, it will trade toward what you named.
Lesson 4
If you want output in a particular style, the fastest way is to show it what good looks like. Paste in an example: a headline you love, a paragraph in the voice you want, a table with the columns you need. One example calibrates the model better than ten adjectives. Adjectives are arguable. Examples are not.
This works for structure too. Want a weekly report in a specific format? Paste last week's report and say "same format, this week's numbers." Want customer replies that sound like you? Paste two of your best replies and say "match this voice." The model imitates patterns superbly, so give it the pattern.
Pro move: keep a folder of examples you admire, from emails to summaries to bios. When you need something similar, drop the example into the prompt. Your taste becomes a reusable asset, and the AI becomes a pretty good ghostwriter for it.
Lesson 5
Starting a prompt with a role, "You are an experienced hiring manager reviewing resumes," is the simplest trick in this course and one of the most powerful. It pulls the model toward a specific corner of its training: the vocabulary, priorities, and standards of that profession. A hiring manager reads a resume differently than a career coach, and you can pick which lens you want.
Make the role specific and relevant. "You are a friendly tutor who never gives the answer directly, only hints" beats "You are a tutor." "You are a senior copywriter who writes punchy, plain-English headlines" beats "You are a writer." The more concrete the role, the more the model leans into its training about how such a person actually works.
You can also stack roles for a second opinion. Get the marketing expert's take, then ask the skeptical customer what they think of it. Nobody knows you did this. You just got a roundtable.
Lesson 6
Beginners judge AI by the first answer. Professionals judge it by the fifth. The first draft is raw material, and your follow-ups are where the quality lives. "Make it shorter." "Less formal." "The second paragraph is good, rewrite the rest to match it." Each round sharpens the result, and it is usually faster than writing from scratch.
The skill here is giving useful feedback. Vague feedback gets vague improvement. Point at specifics: the exact sentence that sounds wrong, the missing fact, the tone that is off. The model cannot read your mind, but it is excellent at applying precise corrections to existing text.
Think of it like directing, not typing. A director does not reshoot the whole movie when one scene is off. They say "again, but angrier." Keep the conversation going in the same thread so the context builds, and do not be afraid to ask for five variations of one line. Iteration is the job.
Lesson 7
Freedom makes AI rambling. Constraints make it sharp. Tell it the length ("in three bullet points," "under 100 words"), the format ("as a table with these columns"), the tone ("warm but professional"), and the audience ("for someone who has never heard of this"). Each constraint removes a guess the model would otherwise make poorly.
The audience constraint deserves special attention. "Explain this to a smart 10-year-old" and "Explain this to my CFO" produce completely different explanations from the same facts. Most of the time your prompt fails, the model was writing for the wrong audience, and you never told it who that was.
Watch out for over-constraining, though. Ten requirements can fight each other, and the model will quietly satisfy half of them. Pick the three constraints that matter most. If you need more, that is a sign you should split the job into smaller prompts, per Lesson 3.
Lesson 8
AI hallucinates. It invents facts, quotes, citations, statistics, and dates with total confidence, because confidence is just tone and tone is free. For brainstorming and drafting, this does not matter. For anything you will act on, publish, or send to a boss, it matters completely.
The rule is simple: the more it matters, the more you verify. Check numbers, names, and dates against a real source. Click the links it gives you; some will not exist. Ask it to show its reasoning or sources, but then still check those, because it can hallucinate sources too. Treat AI output like a talented intern's first draft: promising, and not to be trusted unsupervised.
A useful habit: ask "What are you least sure about in this answer?" Models will often flag the shaky parts when asked directly. It is not foolproof, but it catches things a straight read-through misses.
Lesson 9
AI is genuinely bad at a few things, and knowing them saves you from expensive mistakes. Arithmetic and precise counting: it can botch simple math, so double-check calculations. Very recent events: its training has a cutoff, and it may confidently describe a world that has changed. Judgment calls: it can list pros and cons of firing someone, but it cannot feel the weight of the decision the way you can.
It also struggles with anything requiring exact recall. Asking it to quote a document from memory invites invention. Instead, paste the document into the chat and ask it to work from what you provided. This one technique, giving it the source material instead of asking it to remember, eliminates a huge class of errors.
Finally, it mirrors you. If your prompt is sloppy, biased, or confused, the output will be too. Garbage in does not just produce garbage out. It produces polished, confident garbage, which is worse.
Lesson 10
Treat every AI chat box like a postcard, not a sealed letter. Do not paste passwords, full account numbers, your Social Security number, a client's private data, or anything you would not want on a company bulletin board. Most services use conversations to improve their models, and even with opt-outs, the safest policy is simple: secrets do not go in the box.
This does not mean you cannot use AI for work documents. It means you redact first. Replace real names with placeholders, strip account numbers, remove anything identifying. The AI works just as well on a redacted draft, and your compliance team stays happy.
Same goes for other people's information. Your friend's medical details, your customer's address, a colleague's performance review: not yours to share, not the AI's to see. The convenience is never worth the exposure.
Lesson 11
The moment a prompt works beautifully, save it. Most people regenerate the same good prompt from memory every time, which means they regenerate a worse version every time. A prompt library, a simple note or document with your best prompts, turns one good session into a permanent tool.
Organize it by job: email drafts, meeting summaries, learning new topics, trip planning, tough conversations. For each entry, keep the full prompt, one line about when to use it, and any tweaks you discovered. When you improve a prompt, update the saved version. This is a living document, not a trophy case.
Within a month, you will have ten or twenty prompts that reliably do real work for you. That collection is worth more than any single clever trick, because it compounds. Every new task starts from your best previous thinking instead of a blank box.
Lesson 12
Reading about prompting is nearly useless. Doing it for a week changes how you work. Here is the plan:
Graduation
You know what AI is and is not. You write specific prompts with goals, context, and format. You give it one job, show it examples, assign roles, iterate like a director, set constraints, verify what matters, respect the blind spots, protect your secrets, and save what works. That is the whole skill, and it is genuinely one of the highest-leverage abilities you can own right now. Use it daily and it compounds.
Keep learning
This course is part of Starvard University, striving for greatness.
A note on accuracy: Much of what you read at Starvard was created with help from AI agents, working alongside people. We check what we can, but we cannot guarantee that everything here is true, current, or complete. If something really matters, your health, your money, a grade, a job, please verify it with a trusted source before you rely on it.
