Building AI Literacy — Security Medic Consulting

Students & Early Career / Skills & Practice

12th Grade · First-Year College · Job Seekers

The skill separating people who get real value from AI from people who get generic slop isn’t intelligence or technical background — it’s practice. This is a workshop, not a lecture. Bring a real task. By the end, you’ll have worked.

§01 What “AI literacy” actually means

Literacy isn’t a feeling. It’s a set of habits.

“AI literacy” is one of those phrases everyone uses and almost no one defines. Job postings list it as a requirement. Schools claim to teach it. Keynote speakers at conferences say we all need more of it. Meanwhile, most people pointing at it mean something different from the next person pointing at it.

Here’s a working definition that will actually help you. AI literacy is the ability to do five specific things, reliably, across different situations:

Pick the right tool for a task. Not every problem needs an AI. Not every AI is right for every problem.
Write prompts that get useful output. Bad input produces generic output. Specific, well-framed input produces work you can actually use.
Evaluate what comes back critically. Recognize when output is good, when it’s wrong, when it’s subtly biased, and when it’s bullshit dressed as confidence.
Iterate rather than give up or settle. First drafts from AI are rarely the finished thing. Knowing how to push, refine, and redirect is the core loop.
Know when to stop using it. The best AI-literate people put the tool down when it’s the wrong tool, and don’t feel guilty about it.

If you can do these five things, you’re AI-literate. If you can’t yet, you’re going to be able to by the end of this piece. This is a workshop, not a philosophy lecture.

The core skill

AI literacy isn’t about knowing what AI is. It’s about having good instincts for when to use it, how to direct it, whether to trust what it gives you back, and when to close the tab.

Every technique in this piece is a way of building those instincts through practice.

§02 The mental model that changes everything

Stop thinking of AI as a search engine. Start thinking of it as a brilliant intern.

Almost everyone’s first instinct, when they start using AI, is to treat it like Google. Type a question, expect an answer. This is the wrong mental model, and almost every AI mistake people make flows from it.

A better mental model: AI is a very capable, very fast, very inexperienced intern who has read an enormous amount but has never actually done your job.

That framing changes everything. Consider what you do differently when you’re working with a talented but green intern:

You explain the context of the task, not just the task.
You specify what the finished work should look like.
You give examples of what “good” means in your world.
You tell them who the work is for.
You don’t assume they know which parts are important.
You check their work before passing it on.
When it’s off, you redirect — clearly, without being mean about it.
You don’t fire them the first time they make a mistake. You teach them.

Every one of these translates directly to getting useful output from AI. The people who treat AI like a brilliant intern get useful, specific, tailored work. The people who treat it like a search engine get generic, hollow, frustrating slop — and then they blame the AI, when the mismatch was on their side.

You’re not issuing commands. You’re managing a collaborator. The quality of your output is a function of how well you do the managing.

§03 Prompt craft — the skill that compounds the fastest

Ten seconds of better prompting saves an hour of editing.

The difference between a bad prompt and a good one isn’t length. It isn’t complexity. It’s specificity about four things: the task, the context, the audience, and the form. If your prompt gives the AI all four, you’ll get work you can use. If it gives the AI only one or two, you’ll get something vaguely related to what you wanted.

Here’s the same task, prompted badly and then well.

✕ Weak prompt

Write me an essay about climate change.

No task specifics. No context. No audience. No format. The AI will produce a generic 5-paragraph structure you’ve already read a hundred times.

✓ Stronger prompt

I’m writing a 600-word op-ed for my college newspaper. The thesis: my generation is being sold climate despair when incremental policy wins are actually happening. Write three draft opening paragraphs in different tones — urgent, contrarian, and measured — each citing one specific recent development.

Defines the task, the context, the audience, the length, and the angle. The AI now has enough to produce drafts you can actually choose between.

The stronger prompt takes fifteen seconds longer to type and saves forty-five minutes of editing. This is what prompt craft really is — not a mysterious art form, just the practice of giving the AI enough to work with.

The four questions to ask yourself before every prompt

When you’re about to ask AI for anything non-trivial, pause and answer these four questions in your head. After a few weeks of practice, this becomes automatic.

What exactly am I trying to produce? Not “write about X” — what specifically needs to exist at the end?
What does the AI need to know about the context? The setting, the purpose, the constraints.
Who is the audience? A professor, a hiring manager, a friend, yourself. The audience changes everything about tone and substance.
What’s the form? Length, structure, format. A tweet and a term paper require completely different outputs.

Most AI literacy is just consistently asking these four questions before you press enter.

§04 Five techniques worth practicing until they’re reflex

The moves that separate fluent users from beginners.

These are specific, repeatable patterns you can start using tomorrow. They don’t require memorizing anything. They just require practice.

Technique 1

Give the AI a role

Start your prompt by telling the AI who it’s pretending to be. Expert feedback from a “senior editor with 20 years at The Atlantic” sounds very different from generic AI feedback. A “patient college tutor” gives different explanations than a “professional researcher.” You’re not lying to the AI. You’re framing what kind of response it should generate.

You: “You’re a college admissions officer who has read 10,000 essays. Read my draft and tell me three specific weaknesses you’d flag if this came across your desk.”

AI: Returns feedback that sounds like an admissions officer, not like generic writing advice.

Technique 2

Show an example of what “good” looks like

AI is exceptionally good at pattern-matching. If you show it one or two examples of the output you want — even rough ones — it calibrates quickly. This is vastly more effective than describing the output in the abstract. The technical name is “few-shot prompting,” but in practice it just means: show, don’t tell.

You: “Here’s a product description I like: [paste]. Here’s another: [paste]. Now write one for this product in the same voice: [paste].”

AI: Matches the voice of the examples, because it now knows what you mean.

Technique 3

Ask for multiple versions, then pick

When you want AI to help with something creative or consequential, don’t ask for one draft. Ask for three or five, in different styles. It takes the AI no longer to produce multiple versions, and having options to compare is much better than having one draft to either accept or reject. You also learn what you actually want by seeing what you don’t.

You: “Draft three different opening paragraphs for this cover letter — one direct, one story-driven, one reflective. Don’t polish them. I want to choose a direction first.”

AI: Three distinct approaches. You pick the one that resonates, then iterate on just that one.

Technique 4

Make the AI show its reasoning

For tasks where you care whether the logic is right — analysis, math, arguments, advice — ask the AI to explain its thinking before giving you the conclusion. This catches errors that the polished final answer would have hidden, and it lets you spot places where you disagree with the reasoning. Researchers call this “chain-of-thought prompting.” In practice, just add: “Before you answer, think through this step by step.”

You: “Walk me through your reasoning step by step before giving the final recommendation.”

AI: Shows the logic, which you can now evaluate — instead of just accepting or rejecting a final answer that might have been wrong in an invisible way.

Technique 5

Iterate with small, specific redirects

When AI gives you something close but not right, don’t start over. Redirect. Specifically. “Shorten by half.” “Make the tone more casual.” “Rewrite the second paragraph — it’s too abstract.” “The opening feels generic. Start with a concrete scene instead.” The skill is giving feedback the way you’d give it to a capable collaborator, not scrolling back up and typing a whole new prompt from scratch.

You: “The third paragraph is great. The first and fourth feel like filler. Rewrite those two without losing the idea in the third.”

AI: Targeted revision. You get closer each round.

Practice these five until they feel automatic. After a month of real use, you’ll find yourself reaching for them without thinking — and the quality of what you get back will climb dramatically.

§05 Evaluating output — the skill that will save your career

The ability to tell when AI is wrong is worth more than the ability to make it work.

Generating output with AI is easy. Evaluating that output — knowing when it’s right, when it’s subtly wrong, when it’s confidently making things up, when it’s reflecting bias rather than reality — is the skill that actually separates literate users from amateur ones. It’s also the skill employers care about most.

Here’s a practical framework for evaluating AI output, no matter what form it takes.

The five-question evaluation checklist

When you get meaningful output from AI, run through these five questions before you trust it.

Before you trust AI output, ask:

Q1Is this factually verifiable?

Especially for names, dates, statistics, citations, quotes, and specific claims. AI hallucinates confidently. If something looks like a fact, check it before relying on it. If you can’t verify it, treat it as potentially fabricated.

Q2Does the reasoning hold up?

Read the logic from beginning to end. Does each step follow from the previous one? Are there hidden assumptions? Would a skeptical expert in this field push back on any of the conclusions?

Q3Whose perspective is reflected here?

Does the output reflect only one framing of a contested issue? Is it showing patterns it learned from a narrow slice of training data? If representation or viewpoint diversity matters for this task, is it sufficient?

Q4Does this sound like AI slop?

Generic structure (“In conclusion…” “It’s important to note…”). Hedging where clarity is needed. Parallel constructions that don’t actually say anything. If it reads like the average of everything ever written on the topic, it probably is — and your audience will notice.

Q5Would I be comfortable defending this as mine?

The final check. If a skeptical teacher, employer, or friend pushed you on any specific claim or word choice, could you defend it? If not, either learn it so you can, or cut it.

Signs of AI slop

AI-generated text has patterns. Learning to recognize them is a skill that protects you in two ways: you’ll stop submitting work that screams “AI wrote this,” and you’ll spot AI-generated content when you’re reading it from others.

Empty transitional phrases. “It’s important to note.” “In today’s world.” “One must consider.” Real writers don’t use these unless they’re genuinely setting up something important.
The everything sandwich. Every paragraph presents a “this, and also this, and also this” structure, giving equal weight to every aspect. Real thinking prioritizes.
Synonym overload. Unnecessarily fancy words where simple ones would do. “Utilize” instead of “use.” “Individuals” instead of “people.”
Generic examples. “For example, consider Apple or Google.” Real writers use specific, surprising examples, not the most-cited ones.
The five-paragraph tell. Opening summary, three body paragraphs of equal length, closing summary. High school essay structure, scaled up.
Safety hedging on benign topics. “While opinions vary, it’s worth acknowledging…” where no hedging is needed.
Perfectly balanced opinions. “There are arguments on both sides” when one side is clearly stronger.

The citation trap, one more time

This comes up in card #3, but it’s worth repeating because it keeps ending careers. AI will generate citations that look real and are completely fabricated. Fake authors. Fake journal titles. Fake DOIs. Before submitting anything with a citation the AI provided, verify that the source exists. Use Google Scholar, your library database, or the actual publisher. If you can’t find it, it doesn’t exist. No exceptions.

§06 When not to use AI

The most AI-literate people are the ones who know when to close the tab.

There’s a cultural pressure to use AI for everything. In some contexts, it’s the right default. In others, it actively harms you. The signal that you’re becoming genuinely literate is that you start developing instincts for when the tool is wrong for the job.

Don’t use AI when the point is your own thinking

Some tasks exist to build your brain. Math problems in a class designed to teach you math. An essay assigned to make you wrestle with a hard idea. A journaling practice. A first draft of something you want to say, before you contaminate your own voice with someone else’s. Outsourcing these tasks to AI makes them useless — you’ve traded the growth for the shortcut, and the shortcut is why the task existed.

Don’t use AI when the stakes are high and verification is hard

Medical questions about your own health. Legal questions where the answer depends on jurisdiction. Financial decisions with long-term consequences. Technical questions where being wrong could hurt someone. These are situations where “plausible-sounding answer” can lead you badly astray — and where the cost of being wrong is not recoverable. Use experts, not chatbots.

Don’t use AI when emotional presence is the point

Talking to AI is not the same as talking to a friend who knows you, a parent who loves you, or a therapist trained to help. The AI can say comforting things, and sometimes that’s fine. But real human connection — the kind that helps you through a hard night, a grief, a crisis — cannot be outsourced to a language model. If you catch yourself preferring AI conversation to human conversation about things that matter, that’s a sign to notice, not a habit to lean into.

Don’t use AI when your voice is the point

Personal essays. Wedding toasts. Letters of condolence. A message to someone you love. The reason these land is that they sound like you. AI-polished versions of them don’t. They read as performances of emotions, not expressions of them. If the purpose is for someone to feel seen by you specifically, write it yourself — even if it’s less polished.

Don’t use AI when the work is meant to certify a skill

Coding interviews that test your fundamentals. Exams in a class teaching you a foundational skill. Certifications meant to verify your own knowledge. Using AI here defeats the point of the test, and puts you in the position of having a credential you didn’t actually earn — which catches up with you when you need to use the skill on the job.

Don’t use AI when the context requires confidentiality you can’t guarantee

As covered in card #3, some information doesn’t belong in consumer AI tools. If you wouldn’t be comfortable seeing the data on a billboard, don’t paste it in. The convenience isn’t worth the privacy violation or the legal exposure.

Every time you reach for AI, there’s a silent question embedded in the decision: is this the right tool for this moment? Literacy is the habit of actually asking it.

§07 Building the habits — a 30-day practice plan

What to actually do, starting this week.

AI literacy isn’t built by reading about it. It’s built by using the tools on real tasks, making real mistakes, and learning from them. Here’s a practice plan that will get you from “I’ve heard of AI” to “I’m comfortably fluent” in about thirty days, at roughly fifteen to thirty minutes a day.

A 30-day literacy ramp

Week 1Pick one tool and use it daily on real tasks

ChatGPT, Claude, Gemini — pick one. Use it every day for a week on actual work you’d be doing anyway: studying, research, emails, brainstorming, debugging, drafting. Don’t treat it as a demo. Use it for real things, and notice when it helps and when it doesn’t.

Week 2Practice prompt craft deliberately

This week, every time you use AI, take a moment to write a better prompt than you otherwise would. Use the four-question check from §03 (task, context, audience, form). Try giving the AI a role. Try showing it an example. Try asking for multiple versions. The goal is to make deliberate practice feel natural.

Week 3Focus on evaluation

This week, don’t accept AI output at face value. For every meaningful response, run through the five evaluation questions from §05. Spot-check facts. Read for AI slop patterns. Notice where the output is genuinely useful and where it’s generic filler. You’re training your eye to tell the difference.

Week 4Build something you can show

Use AI to build one concrete thing — a study tool, a personal website, a short analysis, a small automation, a research summary, an app prototype. It doesn’t need to be impressive. It needs to be a specific artifact you built, using AI, that demonstrates you know how to apply these skills. This is what you show in interviews. This is what makes the skills visible.

After thirty days of deliberate practice, you’ll be more AI-literate than most adults in your life. That’s not an exaggeration. Most people pick up AI through casual exposure and never develop the habits — so a month of deliberate use puts you ahead of almost everyone who hasn’t done the work.

§08 Myths about AI literacy

Bad ideas that make the skill harder to build.

Literacy myths, honestly corrected

Myth “AI literacy is about technical skills.”

Reality It’s about judgment and communication. The most AI-literate people I know are often not the most technical. They’re the ones who ask the right questions, spot bad answers, and write clear prompts. Technical skills help; they’re not the bottleneck.

Myth “The best way to learn AI is to read about it.”

Reality Reading helps, up to a point. After that point, it becomes a form of productive procrastination. Fluency comes from use. Pick a real task today, open an AI tool, and work on it. You’ll learn more in thirty minutes of real use than in three hours of articles.

Myth “I should wait for the AI to get better before I learn it.”

Reality Waiting is the most expensive strategy. The fundamentals you build now — prompt craft, evaluation, knowing when not to use it — will still matter when the AI improves. The people who wait for “AI to mature” keep finding new reasons to wait, while the people who started earlier are steadily more effective.

Myth “Using AI makes you dumber.”

Reality Using AI badly makes you dumber — the same way using a calculator badly in math class prevents you from learning arithmetic. Using AI well makes you more capable. The question isn’t whether to use it; the question is whether you’re using it in a way that builds your thinking or substitutes for it.

Myth “I’ll learn AI when I need it for a job.”

Reality By the time you need it for a job, the people who started earlier will be much better at it than you, and hiring managers will notice. Also: you’ll need it sooner than you think. College already assumes some AI literacy in many classes. Internships assume it. Start now, on low-stakes tasks, so you’re fluent when the stakes rise.

§09 Practice-level terms worth knowing

A glossary for the practical craft.

Prompt

The input you give to an AI. The craft of writing good prompts is often called prompt engineering, but for most purposes, it’s just clear thinking about what you want.

System prompt

An instruction that sits above the conversation, setting the AI’s role or constraints. Not always visible to you, but it shapes how the AI responds to everything in that session.

Few-shot prompting

Giving the AI one or two examples of the output you want, so it can match the pattern. The practical version of “show, don’t tell.”

Zero-shot prompting

Asking the AI to do something without giving examples. Works for common tasks; less effective when the output needs to match a specific style or voice.

Chain-of-thought

Asking the AI to reason step by step before giving a final answer. Improves accuracy on complex tasks and lets you check the logic.

Context window

How much text the AI can “see” at once. Longer windows let you paste in larger documents for analysis. Running out of context is why the AI sometimes forgets the start of a long conversation.

Temperature

A setting (in some tools) that controls how random or predictable the output is. Low temperature = conservative, consistent output. High temperature = more creative, less predictable output. Most consumer tools don’t expose this directly.

Hallucination

When the AI generates confident output that’s factually wrong. Especially common with citations, statistics, historical facts, and niche claims. Always verify.

AI slop

Generic, low-effort AI-generated content that lacks voice, specificity, or originality. A term that appeared in 2024 and has stuck because the phenomenon is so recognizable.

Grounding

Giving the AI specific source material to work from (a document, dataset, or set of facts), so its output is based on that material rather than whatever it remembers from training. Usually reduces hallucination.

Retrieval-Augmented Generation (RAG)

A technique where the AI first looks up relevant documents, then uses them to answer. It’s why some AI tools cite real sources while others make them up.

Model comparison

The practice of trying the same prompt across multiple models (GPT, Claude, Gemini, etc.) and comparing outputs. Teaches you the strengths and weaknesses of each, and protects against overfitting your judgment to one tool.

Agent

An AI system that can take actions (browse the web, run code, use tools) rather than just responding to prompts. A newer frontier. Practicing with simple agents now will matter when they’re more capable.

Fine-tuning vs. prompting

Two ways to make an AI do something specific. Fine-tuning modifies the model itself (expensive, requires data). Prompting just changes the input (cheap, immediate). For most users, prompt-level skills are what matter.

§10 What to carry forward

Seven principles of the literate user.

Literacy is habits, not credentials. You become AI-literate by practicing, not by taking a course. Use the tool on real things.
Treat AI like a brilliant intern, not a search engine. Context, examples, specificity. You’re managing, not commanding.
Prompt craft compounds fast. Fifteen seconds of better prompting saves an hour of editing. Learn it once; benefit for years.
Evaluate everything important. The skill that will save your career — professionally and academically — is being able to tell when AI is confidently wrong.
Know when to close the tab. The most AI-literate people are the ones who can put the tool down when it’s the wrong tool.
Build something you can show. A project, a portfolio, a specific artifact. Evidence beats assertions, especially in interviews.
Stay calibrated. The tools change. Your fundamentals (prompt craft, evaluation, judgment) don’t. Keep using them, and keep learning what’s new without panicking.

AI literacy is not a certificate. It’s not a mindset. It’s not a worldview. It’s a pile of small habits you practice until they’re invisible — and then they become the thing that makes your work better than everyone else’s.

The final card in this series applies everything you’ve learned so far to the specific moment most of you will face next: using AI to find, apply for, and land a job. It’s where all this theory meets the most important real-world test of your AI literacy. And it’s where the people who’ve built these habits pull dramatically ahead of the people who haven’t.

Next in this series

Job Seeking with AI

How to use AI tools to research roles, tailor applications, and prepare for interviews — while understanding where the line is between assistance and misrepresentation.