Week 9: Prompt Engineering Fundamentals

Learning objectives

By the end of this week, you will be able to:

1. Write a clear, specific instruction for an AI tool instead of a vague one, and explain why the difference matters.
2. Provide useful context in a prompt, such as audience, purpose, and what you have already tried.
3. Use an example inside a prompt to show an AI tool the style, format, or tone you want.
4. Assign a role to an AI tool and explain why doing so can shift its tone and focus usefully.
5. Request a specific output format, such as a list, a short paragraph, or a set word count.
6. Iterate on a prompt with follow-up requests instead of starting over from scratch each time.
7. Ask an AI tool specifically for accessible output, such as plain text instead of a table or clear structured headings.
8. Combine several of these techniques into a single, well-formed prompt.

Lesson

A prompt is the text you type or say to instruct an AI system, and a token is a small piece of text the AI processes one piece at a time. Nearly everything you do with an AI tool, no matter which tool, involves writing a prompt. This skill has a name: prompt fluency, the ability to write clear, effective instructions and get a useful result efficiently, without a lot of trial and error.

This week breaks prompt fluency down into six concrete building blocks you can practice one at a time: clear instructions, context, examples, role assignment, output format requests, and iteration. It closes with a section on a building block that matters especially for you as a screen reader user: prompting specifically for accessible output. None of these building blocks require memorizing special code or secret phrases. They are habits of communication, the same kind of habits that make a request to a helpful coworker easier to act on than a vague one.

Clear instructions: the difference between vague and specific

A vague prompt leaves an AI tool guessing about what you actually want, and it will guess, filling in the gaps with something generic. Here is a familiar example of this idea: "write something about our product" tends to produce an unfocused result that needs heavy rewriting, while "write a three-sentence product description for a first-time customer, in a friendly tone, mentioning our free trial" tends to produce something close to usable right away. This lesson takes that same idea and applies it far beyond product descriptions, to nearly everything you might ask an AI tool to do.

Think about the difference between asking a coworker "can you help with the report?" and asking "can you check the third paragraph of the report for any spelling errors before three o'clock?" The second version tells your coworker exactly what to do, which part to focus on, and when you need it. An AI tool needs the same level of specificity, because it cannot read your mind, ask a clarifying question the way a coworker might, or notice your tone of voice to guess what you really meant.

Here is a second before-and-after pair. A vague prompt: "Explain AI to me." This could produce anything from a one-sentence definition to a college lecture, and you have no way to predict which one you will get. A specific prompt: "Explain what a large language model is, in three sentences, for someone who has never used AI before." This tells the AI tool the topic, the length, and the audience, all in one sentence, and it will almost always produce something close to what you actually wanted on the first try.

Being specific does not mean writing a long prompt. A short prompt can be perfectly specific, and a long prompt can still be vague if it never states what you actually want. The goal is precision, not length: name the topic, name what you want done with it, and name any limit that matters, such as a length, an audience, or a deadline.

A practical habit worth building is to pause before you send a prompt and ask yourself one question: if a new coworker who knew nothing about my situation read only this sentence, could they do exactly what I want on the first try? If the answer is no, something is missing, usually the topic, the desired action, or a limit such as length or audience. This one habit, checked silently before every prompt, catches most vagueness before it ever costs you a wasted round of editing.

Context: telling the AI what it needs to know

Context means background information relevant to your request: who the response is for, what its purpose is, and what you have already tried. An AI tool has no memory of your life, your workplace, or your goals unless you tell it, in that conversation, what matters. Two people can ask an AI tool the exact same question and need two completely different answers, simply because their situations are different, and context is what lets the AI tool tell those situations apart.

Consider a request to "help me write a cover letter." Without context, the AI tool has no idea which job you are applying for, what your background is, or what tone the employer expects. With context added, such as "help me write a cover letter for an administrative assistant position at a small nonprofit, mentioning my three years of customer service experience and my comfort with AI tools," the AI tool has almost everything it needs to produce something genuinely useful on the first attempt.

Context also includes what you have already tried, especially when you are troubleshooting or refining something. Telling an AI tool "I already tried turning it off and on again, and that did not fix it" saves it from suggesting the same first step you already ruled out. Telling an AI tool "I already asked for a shorter version and it was still too formal" saves you a wasted round of back-and-forth. Good context is not about writing a long biography before every request; it is about naming the two or three facts that would actually change the answer.

A simple way to check whether you have given enough context is to imagine handing your exact prompt to a stranger who knows nothing about your situation. If that stranger could not produce a useful answer from your prompt alone, an AI tool probably cannot either.

Context becomes especially important once a conversation runs long. A context window is the amount of text an AI system can consider at one time. In a very long conversation, early details can fall outside that window and effectively be forgotten. If you notice a response that seems to ignore something you established earlier, it is often worth restating the key piece of context directly in your next prompt, rather than assuming the AI tool still remembers it.

Examples: showing the AI what you want

Sometimes the easiest way to describe what you want is not to explain it in words at all, but to show a sample of it. Including one example of the style, format, or tone you want inside your prompt is one of the most reliable ways to get a matching result. This technique has a formal name, "few-shot prompting"; for now, just think of it as showing, not only telling.

For example, instead of asking "write a friendly reminder email," you could write: "write a friendly reminder email in this style: 'Hi Sam, just a quick nudge that your library book is due Friday. No rush, just didn't want it to slip past you. Thanks!' Write a similar reminder about an upcoming dentist appointment." The AI tool now has a concrete sample of the tone, sentence length, and level of formality you want, which is far more reliable than trying to describe "friendly but not too casual" in the abstract.

Examples are especially useful when a tone or format is hard to put into words but easy to recognize once you see it. If you have ever struggled to describe exactly how formal or casual you want something to sound, try pasting in one sentence that already sounds right, and ask the AI tool to match that style for the rest. This works well for emails, social media posts, and any writing task where "you'll know it when you hear it" describes your own standard better than any adjective would.

Role assignment: asking the AI to respond as a particular expert

Role assignment means asking an AI tool to respond as though it were a particular kind of expert or persona, such as "respond as a patient teacher explaining this to a beginner" or "respond as a professional editor reviewing this document." This framing can shift the tone, focus, and level of detail in a response in genuinely useful ways. A prompt that says "explain this as a patient teacher" tends to produce a slower, more encouraging, more step-by-step explanation than the same question asked plainly. A prompt that says "review this as a professional editor" tends to focus more tightly on clarity, grammar, and structure than a general "look this over" request would.

It helps to understand why this works. An AI tool has been trained on huge amounts of text written by real teachers, real editors, and real experts of every kind, so asking it to adopt one of those roles nudges it toward the patterns of language, tone, and priorities associated with that role in its training data. It is a framing tool, a way of steering the response, not a guarantee that the AI tool suddenly has real expert-level knowledge or credentials it did not have a moment before. The verification habits covered earlier in this course still apply in full: cross-check anything important, especially medical, legal, or financial claims, no matter what role you assigned.

Useful role assignments tend to be specific rather than generic. "Respond as an expert" adds little on its own, since "expert" does not point toward any particular tone or priority. "Respond as a patient teacher explaining this to someone brand new to computers" or "respond as a strict but fair editor who catches every typo" gives the AI tool something concrete to imitate, and the more concrete the role, the more noticeable the shift in the response tends to be.

Role assignment also pairs naturally with context. Telling an AI tool "respond as a patient teacher, and remember I am completely new to spreadsheets" combines a role with a fact about your situation, and the two together tend to produce a far gentler, more step-by-step explanation than either one alone would. As with every technique in this lesson, the building blocks are meant to be combined, not used one at a time in isolation.

Output format requests: telling the AI exactly how to shape its answer

An output format request tells the AI tool exactly how you want the response structured before it writes a single word, rather than leaving the shape of the answer up to chance. This might mean asking for a numbered list instead of a paragraph, a short paragraph instead of a long list, an exact word count, or a specific structure such as "a heading, followed by three short paragraphs." Being explicit about format up front almost always saves you editing time later, because reformatting a response after the fact, especially with a screen reader, takes real effort: reflowing a wall of text into a list, or condensing five paragraphs down to three, is slower than simply asking for the right shape from the start.

Some useful format requests you will use constantly: "give me a numbered list of five steps," "answer in one short paragraph, no more than sixty words," "use a heading before each section," and "give me exactly three bullet points, not a paragraph." Each one removes a decision the AI tool would otherwise make on its own, and it tends to make that decision differently every time, which is part of why an unstructured request can produce wildly different-looking answers from one attempt to the next.

Format requests work well combined with everything else in this lesson. You can assign a role and request a format in the very same prompt, for example: "as a patient teacher, explain how compound interest works, in exactly three short paragraphs, with a heading before each one." Later exercises in this week will have you practice exactly that combination.

Iteration: treating the first response as a draft

Iteration means treating an AI tool's first response as a starting draft you can refine with follow-up prompts, rather than as a finished answer you must either accept or throw away and start over. If the first response is too long, you do not need to rewrite your entire original prompt from scratch; you can simply reply "make it shorter." If the wording feels too technical, reply "use simpler words." If it feels stiff or full of jargon, reply "try again without jargon." Each of these short follow-ups builds on everything the AI tool already produced, refining it step by step instead of starting over.

This connects to a broader point about AI drafting in general: AI can save real time by producing a fast first draft, but that draft always needs your judgment before it is finished. Iteration is where that judgment gets applied in practice. You read the first response, decide what is missing or wrong, and say so directly, in plain language, exactly the way you would give feedback to a coworker's first draft of a memo.

Iteration is often faster than trying to write the perfect prompt on the first attempt. Even experienced prompt writers rarely nail every detail in one try, and that is fine: a quick exchange of two or three short follow-ups usually gets you to a better result, faster, than spending five minutes trying to anticipate every possible issue in a single giant first prompt. Give yourself permission to start simple and refine, rather than feeling you must get everything right immediately.

A few phrases are worth having ready for iteration: "make it shorter," "make it longer," "use simpler words," "make the tone more formal," "make the tone more casual," "try again without jargon," and "keep the same content but reformat it as a list." Keeping a mental list like this turns editing an AI response into a quick, familiar routine instead of a fresh puzzle every time.

Prompting for accessible output

Everything so far in this lesson applies to any AI user. This section applies specifically to you, as someone who reads AI output with a screen reader, and it may be the single most practically useful section in this week's lesson. AI tools were mostly trained on how sighted readers process information, which means their default output sometimes leans on structure that is easy to see but slow or awkward to hear: a wide table where a short list would do, dense bold or color-based emphasis, or a wall of unbroken text with no headings to navigate by.

The good news is that you can simply ask for what works better for you, the same way you would ask for a specific word count or tone. If an AI tool responds with a table and you did not actually want one, say so directly: "please give me that same information as a plain list instead of a table." If a response comes back as one long, unbroken block of text, ask: "please break this into a numbered list" or "please add a heading before each section." These are not unusual or difficult requests. AI tools handle them easily and reliably, precisely because they are just another kind of output format request, the same building block covered earlier in this lesson.

It also helps to ask an AI tool up front, before it even produces a first response, to avoid formatting that depends on things a screen reader cannot convey well. A prompt like "please answer in plain text with clear headings, and do not rely on bold text, color, or spacing to show importance, since I am using a screen reader," sets an expectation from the very first response instead of requiring a correction afterward. Many learners find it worth adding a short version of this instruction as a standing habit whenever they open a new AI conversation for an important task, especially one they plan to keep or share.

Some concrete accessible-output requests worth having ready: "give me this as a plain bulleted list, not a table," "put a heading before each section instead of writing one long paragraph," "do not use tables for this," "write this so headings, not bold text or color, carry the structure," and "keep sentences short and avoid dense blocks of text." Each one asks for exactly the same underlying information, just organized in a way that reads well aloud, one item at a time, through a screen reader.

Accessible-output requests are worth building into your very first message in a new conversation whenever a task matters, rather than waiting to see what the AI tool produces by default and fixing it afterward. Setting the expectation up front, for example by opening with "please use plain text, clear headings, and no tables unless I specifically ask for one," tends to keep every later response in that same conversation consistent, since the AI tool generally continues following an instruction you gave earlier unless you tell it otherwise.

One caution matters here: asking an AI tool to follow an accessible format is not the same as confirming that it actually did. AI tools sometimes say a response is now "in plain text" or "no longer a table" while the underlying HTML or document structure still contains a table, or while headings were only described in words rather than actually marked as headings. Before you trust an AI-generated document's or webpage's structure, especially one you plan to keep, share, or submit somewhere, check it directly with your screen reader's list of links or interactive elements (sometimes called an elements list), or its landmark and region navigation. If your screen reader's tool offers a Tables view, use it to confirm no unwanted table exists, and use your screen reader's heading navigation to confirm real headings are present. Trust, but verify, applies to formatting requests exactly as much as it applies to factual claims.

Putting it together: a worked example

Here is a single example that shows every building block from this lesson working together. Start with a vague prompt: "Tell me about time management." This prompt gives no context, no example, no role, no format, and no sense of what "good" would even look like, so the response could be almost anything: a short tip, a long essay, a list, a lecture, all equally likely.

Now compare a revised prompt that applies several techniques at once: "As a patient teacher, explain three simple time management tips for someone who works from home and gets distracted easily. Give me a heading before each tip, followed by one short paragraph of no more than three sentences. Keep the language simple, avoid jargon, and do not use a table." This revised prompt assigns a role (patient teacher), gives context (works from home, gets distracted easily), sets a clear instruction (three tips), requests a specific output format (heading plus short paragraph per tip), and builds in an accessible-output request (no table, no jargon) all in one pass.

The likely improvement is significant. The vague version might return a long, generic essay covering a dozen loosely related tips with no clear structure, forcing you to skim or listen through all of it to extract anything useful. The revised version is far more likely to return exactly three headed sections, each short enough to read quickly with Say All, already in a format you can use without further editing. That gap between the two results is the entire point of this week's lesson: the same AI tool, the same underlying capability, produces a dramatically more useful result once you know how to ask.

Notice, too, that the revised prompt did not require anything technical or unusual, no special software and no secret commands. It required only plain language, applied deliberately. That is exactly what prompt fluency means, and it is a skill that gets faster and more automatic every time you practice it, which is exactly what this week's exercises and portfolio project are designed to help you do.

Key terms from this week

Use this list to review the vocabulary introduced in this lesson before you start the exercises and the test.

Role assignment

Asking an AI tool to respond as though it were a particular kind of expert or persona, such as a patient teacher or a professional editor, in order to shift its tone and focus.

Output format request

Telling an AI tool exactly how you want its response structured, such as a numbered list, a short paragraph, a set word count, or headings, before it writes the response.

Iteration

Treating an AI tool's first response as a draft and refining it with short follow-up prompts, such as "make it shorter," instead of starting over from scratch.

Keyboard-only exercises

These exercises use only your keyboard. Screen reader commands are described in general terms below; substitute your own screen reader's equivalent command where needed.

Exercise 1: Revise a vague prompt into a specific one

1. Open your AI chatbot of choice and locate its message box. In most browser-based tools, press Tab until you hear "edit," "message," or the tool's name for its text entry field, or use your screen reader's browse mode navigation to jump directly to an edit field.
2. Type a deliberately vague prompt, such as "tell me about resumes," and send it by pressing Enter.
3. Wait a few seconds for the response to begin appearing, then use Say All, or your screen reader's continuous reading command, to read the full reply. Check your own screen reader's shortcuts if you are not sure which key combination starts it.
4. Return to the message box and type a specific, revised version of the same request, naming a clear instruction, context, and an output format, such as "give me three tips for improving a resume for an administrative job, as a numbered list, one short sentence each." Send it.
5. Read the second response with Say All and compare it to the first. Notice which one was more immediately useful and why.

Exercise 2: Turn a table into an accessible list

1. In the same AI chatbot, type a prompt that is likely to produce a table, such as "give me a comparison of three note-taking apps in table form," and send it.
2. Open your screen reader's list of links or interactive elements (sometimes called an elements list), or use its landmark and region navigation to explore the response. Check your own screen reader's documentation for the exact shortcut.
3. If your screen reader offers a Tables view, switch to it and confirm a table is present in the response, then close the list.
4. Return to the message box and type "please give me that same comparison again as a plain accessible list instead of a table." Send it.
5. Open your screen reader's elements list again, check the Tables view if available, and confirm the second response contains no table. Then check the Headings or Links view as needed to review the list's structure.

Exercise 3: Combine role assignment and output format in one prompt

1. Choose a topic you would like explained simply, such as how a savings account earns interest, or how to back up files.
2. In the message box, type a single prompt combining a role and a format request, for example: "as a patient teacher, explain how a savings account earns interest, in exactly three short paragraphs, with a heading before each one." Send it.
3. Read the response with Say All, or your screen reader's continuous reading command.
4. Use your screen reader's heading navigation to jump heading by heading through the response, confirming three headed sections are present as requested.
5. If the response did not match your request exactly, practice one round of iteration by replying with a short follow-up, such as "make each paragraph shorter," and read the revised response.

Portfolio project: My Prompt Before and After

This week's portfolio piece is a personal document showing how applying this week's techniques improves a prompt. You will build on this habit throughout the rest of this course.

1. Open Notepad or your word processor of choice.
2. Write a heading at the top: "My Prompt Before and After."
3. Write three vague prompts of your own, each on a topic that matters to you, such as work, a hobby, or something you want to learn.
4. Underneath each vague prompt, write a revised version applying at least two techniques from this week's lesson, such as adding context and requesting a specific output format, or assigning a role and requesting accessible output.
5. After each pair, write one sentence explaining what specifically improved between the vague version and the revised version.
6. Try at least one of your revised prompts in an actual AI tool, and add a short note about how the real response compared to what you expected.
7. Save the file with the name my-prompt-before-and-after, in your portfolio folder.

Weekly test

Answer key

Each answer below includes a one-sentence explanation drawn from this week's lesson.