Five Moves That Actually Work

Core Prompting Techniques

Most professionals are prompting AI completely wrong, and getting mediocre results because of it

Ask ten managers how they use ChatGPT and nine will describe the same pattern: type a quick question, get a vague answer, feel mildly disappointed, and quietly wonder if AI is overhyped. The problem isn't the tool. The problem is a set of deeply held beliefs about how prompting works that are simply incorrect. These beliefs are understandable, they come from how we use search engines, how we talk to assistants, and how we were taught to ask questions. But they produce weak results. This part of the lesson names three of the most common misconceptions, shows you exactly why they fail, and replaces each one with a mental model that actually works. By the end, you'll understand why two people can use the exact same AI tool and get wildly different outputs.

Myth 1: The AI Knows What You Mean

The most common assumption professionals make is that AI understands context the way a colleague does. You've worked with your marketing director for three years. When you say 'clean up the proposal,' she knows your brand voice, your client's industry, your preferred tone, and what 'clean up' actually means to you. AI has none of that unless you provide it. Every prompt is a cold start. ChatGPT, Claude, and Gemini don't carry memory of your preferences between sessions by default. They don't know your company, your role, your audience, or your standards. They generate the most statistically likely response to the exact words you typed, nothing more, nothing less.

This gap explains why a prompt like 'write a proposal for a new client' produces something embarrassingly generic. The AI isn't being lazy. It's doing exactly what you asked, filling in every blank with the most average, middle-of-the-road interpretation possible. 'New client' could mean a Fortune 500 procurement team or a local bakery owner. 'Proposal' could mean a two-page executive summary or a 20-slide deck. Without specifics, the model hedges. It writes for everyone, which means it writes for no one. Professionals who understand this stop blaming the tool and start providing context deliberately.

The corrected mental model is this: treat every AI prompt like a briefing to a brilliant new contractor who started today. They are highly capable but know nothing about your world. A good briefing covers who you are, what you're making, who it's for, what tone is appropriate, and what success looks like. When you give that briefing in your prompt, the output quality jumps immediately. This isn't a minor improvement, professionals who add context consistently report outputs that need 60-70% less editing. That's the difference between a useful tool and a frustrating one.

Myth 2: Longer Prompts Are Always Better

Once professionals learn that context matters, many swing to the opposite extreme: writing prompts that read like legal briefs. Pages of background, bullet points of requirements, exhaustive lists of things to avoid. The assumption is that more information always produces better output. This is partly right and mostly wrong. More relevant context helps. More irrelevant context hurts. When you bury a simple request under paragraphs of tangential detail, the model has to decide what matters most, and it frequently picks wrong. Precision beats volume every time.

Think about how you'd brief a graphic designer for a presentation cover slide. You wouldn't hand them 40 pages of company history. You'd say: 'Dark blue background, white sans-serif font, our logo top-left, the title centered, professional but not stuffy, here's the brand guide.' Specific, structured, and tight. That's exactly how effective prompts work. The best prompts for business tasks are typically three to six sentences. They specify role, task, audience, format, and constraints. That's it. Anything beyond that should be there only if it directly changes the output.

Myth 3: Prompting Is a One-Shot Activity

Most professionals treat prompting like a vending machine: insert request, receive output, accept or reject. If the first response isn't good, they either give up or start over with a completely different prompt. This is the wrong model entirely. Professional AI use is iterative. The first response is a draft. It shows you what the model understood and where it went wrong. The right move is to continue the conversation, refine, redirect, push back, add constraints, ask for alternatives. A skilled prompter treats the first output as the opening move in a dialog, not the final product.

Think of it the way you'd work with a capable but unfamiliar colleague on a first draft. They submit something. You don't throw it away, you mark it up. 'The tone is too formal for this audience,' 'Cut the third paragraph,' 'The opening needs to lead with the cost savings, not the features,' 'Give me three alternative headlines.' Each of those follow-up instructions is itself a prompt. Within four or five exchanges, you often have something genuinely strong. This iterative approach consistently outperforms the one-shot method, and it's faster than rewriting from scratch every time something misses.

Myth vs. Reality: What Most Professionals Get Wrong

Click to enlarge

What Actually Works: The Mental Model Behind Strong Prompts

Strong prompts share a consistent structure, even when they look different on the surface. They establish a role or persona for the AI, state the specific task, define the audience, specify a format, and include at least one quality constraint, a tone, a length, a perspective, or a list of things to avoid. You don't need all five elements in every prompt, but the more of them you include, the more precisely the model can target its output. This isn't a rigid formula. It's a checklist you internalize until it becomes instinct. Experienced AI users apply it automatically, the same way an experienced writer automatically thinks about audience before drafting.

The role element deserves special attention because it's the one most professionals skip. Telling ChatGPT or Claude to respond 'as an experienced HR business partner reviewing a job description for bias' produces fundamentally different output than asking it to 'review this job description.' The role activates a different cluster of the model's knowledge and perspective. It's not magic, it's giving the model a frame. A hiring manager, a skeptical CFO, an empathetic customer service lead, and a compliance officer will all review the same document differently. Specifying the role tells the AI which lens to use.

Format instructions are equally powerful and equally neglected. Left to its own defaults, most AI tools produce medium-length prose with occasional bullet points. That default format rarely matches your actual need. A busy executive wants three bullet points. A client proposal needs headers and a summary paragraph. A brainstorming session needs a numbered list of 10 options, not a flowing essay. Stating the format explicitly, '3 bullet points, each under 20 words' or 'a two-paragraph summary followed by three talking points', removes ambiguity and gets you output you can actually use without reformatting.

Click to enlarge

Monday Morning Practice: Build Your First Structured Prompt

Frequently Asked Questions

• Q: Does the five-element structure work the same way on all AI tools. ChatGPT, Claude, Gemini? A: Yes, the structure works across all major tools. The underlying principle, give the model a role, task, audience, format, and constraint, applies universally. Minor differences exist in how each tool responds to role prompts (Claude tends to stay in character more consistently), but the quality improvement holds across all platforms.
• Q: How long should my context briefing be at the start of a prompt? A: Two to three sentences is usually enough. State your role or situation, the purpose of the output, and who will use it. Example: 'I'm an HR manager at a 150-person tech company. I need to update our remote work policy for a staff all-hands email. Audience is non-technical employees who are skeptical of policy changes.' That's enough context to shift output quality significantly.
• Q: If AI doesn't remember previous conversations, do I have to re-explain my context every single time? A: For most tools, yes, but there are workarounds. ChatGPT Plus has a 'Memory' feature (under Settings) that stores preferences across sessions. Claude Pro allows you to paste a personal context block at the start of any conversation. Many professionals keep a saved 'context template' in their notes app and paste it at the start of any session where it's relevant.
• Q: What if I try iterating and the AI keeps giving me the same wrong answer? A: Change your approach rather than repeating the same instruction. If 'make this shorter' isn't working, try 'cut this to exactly 50 words' or 'remove everything except the three most important points.' Specificity almost always breaks the loop. If a model is consistently misunderstanding a task, try reframing the task entirely, sometimes a fresh angle works better than pushing harder on the same request.
• Q: Is there a risk of making prompts too specific and constraining the AI's usefulness? A: Rarely. The most common problem is under-specification, not over-specification. That said, if you specify a format so rigidly that it doesn't fit the content (e.g., 'exactly 5 bullet points' for a topic that only has 3 distinct points), you'll get padded output. The fix is to use constraints that match the task, 'up to 5 bullet points' gives structure without forcing it.
• Q: Should I use different prompting approaches for different tools, like Copilot in Word vs. ChatGPT? A: Yes, slightly. Microsoft Copilot in Word or Outlook works best when you give it direct, action-oriented instructions tied to the document in front of you ('Rewrite this paragraph to be more concise' or 'Suggest three subject lines for this email'). Standalone tools like ChatGPT and Claude handle more complex, multi-step prompts better because you have more space and conversational flexibility. The five-element structure still applies, you just adapt the scope.

Key Takeaways from Part 1

1. AI has no automatic context. Every prompt starts cold, you must brief it the way you'd brief a capable new hire who knows nothing about your work.
2. Longer prompts aren't better prompts. Relevant specificity beats volume. Front-load your most important instruction and keep prompts to 3-6 focused sentences.
3. Prompting is a conversation, not a vending machine transaction. Treat the first response as a draft and use follow-up messages to refine, redirect, and improve.
4. The five-element structure, role, task, audience, format, constraint, is a practical checklist that improves first-draft quality on almost any work task.
5. Format instructions are one of the most neglected and highest-impact elements. Telling AI exactly how to structure the output saves significant editing time.

Three Things Most Professionals Get Wrong About Prompting

Most professionals believe that getting better results from AI is mostly about finding the right magic words, a secret phrase that unlocks better output. They also tend to believe that longer, more detailed prompts are always better than short ones. And a third very common assumption: that if an AI gives you a bad answer, you should start over from scratch with a completely new prompt. All three of these beliefs lead to real frustration and wasted time. Each one is either wrong or incomplete enough to be actively unhelpful. Here is what the evidence and practical experience actually show.

Myth 1: Better Prompting Is About Finding Magic Words

The idea that prompting is about discovering special trigger phrases, 'say please,' 'tell it you're an expert,' 'use the word jailbreak', treats AI like a vending machine with a secret code. This belief gets reinforced by viral social media posts listing 'power prompts' or '10 words that make ChatGPT smarter.' It feels intuitive because we're used to search engines, where specific keywords dramatically change results. But large language models don't work that way. They respond to meaning, context, and structure, not keyword combinations. A prompt that worked brilliantly for someone else's use case may produce mediocre results for yours, because the underlying task and context are different.

What actually drives output quality is the clarity of your request, the context you provide, and the constraints you set. Think of it like briefing a highly capable but brand-new contractor. If you hand them a vague sticky note, you'll get vague work, no matter what words are on that note. But if you explain the project background, the audience, the format you need, and what 'good' looks like, you get something useful on the first attempt. The AI isn't responding to magic; it's responding to information. Professionals who understand this stop hunting for secret phrases and start thinking about what information the AI actually needs to do the job well.

The practical shift here is from 'what words should I use?' to 'what does this AI need to know to do this well?' That reframe changes everything. You start thinking about role, context, format, and constraints as the real levers, because they are. A marketing manager who tells ChatGPT 'write a subject line' will get something generic. The same manager who says 'write 5 subject lines for a re-engagement email to lapsed B2B customers in the HR software space, targeting HR directors, with a tone that's direct but not pushy' will get something they can actually use. The difference isn't a magic word. It's specificity.

Myth 2: Longer Prompts Are Always Better

Once professionals learn that context matters, many overcorrect. They start writing prompts that are 400 words long, include every possible piece of background information, list 15 requirements, and end with three paragraphs of caveats. The logic seems sound: more information means better output. But this creates a different problem. When a prompt contains too many competing instructions or buries the core task under layers of context, the AI has trouble identifying what you actually want most. Output quality can actually decrease when prompts become unfocused, because the model is trying to satisfy too many constraints simultaneously.

The right length for a prompt is the length that gives the AI what it needs, no more, no less. For a simple, well-defined task like 'summarize this meeting transcript in five bullet points,' a short prompt is perfect. For a complex task like drafting a client proposal with specific tone, structure, and audience considerations, more detail is warranted. The test isn't word count, it's completeness. Ask yourself: does this prompt answer the four key questions? What is the task? Who is it for? What format do I need? What constraints apply? If you've answered those four questions clearly, your prompt is long enough.

There's also a structural point worth making: when you do have a lot of requirements, organize them. Use a numbered list of constraints rather than burying them in paragraphs. Put the most important instruction first. Separate the context (background information) from the task (what you want done) and the format (how you want it delivered). A well-organized 150-word prompt will outperform a rambling 400-word one almost every time. Think of it like a good project brief, structure makes requirements readable, and readable requirements get executed correctly.

Myth 3: A Bad Answer Means You Should Start Over

When AI produces something that misses the mark, the instinct for many professionals is to abandon that conversation and open a fresh one with a different prompt. This is almost always the wrong move. The conversation history in tools like ChatGPT, Claude, and Copilot is an asset. The AI already has your context loaded. A bad first response is feedback, not failure, it tells you what information was missing or what instruction was ambiguous. The most effective users of AI tools iterate within a conversation, using the bad response as a diagnostic to write a sharper follow-up.

Iteration within a conversation is one of the most underused techniques in professional AI use. Instead of starting over, try one of these targeted follow-ups: 'That's too formal, rewrite it in a more conversational tone.' Or: 'The second paragraph is good. Rewrite only the first paragraph to be more concise.' Or: 'You missed the key constraint, the audience is non-technical. Revise with that in mind.' Each of these corrections builds on what the AI already has, rather than asking it to start from zero. Three targeted iterations almost always produce better results than three completely different first prompts.

Click to enlarge

Myth vs. Reality: The Full Picture

Click to enlarge

What Actually Works: The Techniques That Consistently Deliver

Three techniques consistently produce better results across tools, tasks, and professional contexts. The first is role assignment. When you tell an AI to act as a specific type of professional, 'You are an experienced HR business partner' or 'You are a financial analyzt writing for a non-finance audience', you're not using a magic word. You're activating a specific register of language, a set of professional conventions, and a particular perspective. The AI draws on the enormous volume of text it was trained on from that professional domain. The result is output that sounds like it was written by someone who knows that world, because in a meaningful sense, it was.

The second technique is output formatting as instruction. Most professionals type what they want the AI to say, but forget to specify how they want it formatted. This is a significant missed opportunity. Telling the AI 'respond in a table with three columns: action, owner, deadline' or 'write this as a 5-bullet executive summary, each bullet under 20 words' doesn't just make the output look better, it forces the AI to organize its thinking. Structured output requirements produce more disciplined, scannable, and usable content. This is especially powerful for meeting summaries, project plans, comparison analyzes, and status updates.

The third technique is constraint-setting. Constraints feel counterintuitive, why limit the AI? But constraints dramatically improve relevance and usability. 'Write a performance review comment' produces generic output. 'Write a performance review comment for a mid-level project manager who exceeded delivery targets but needs to improve stakeholder communication, max 80 words, no jargon, suitable for a formal HR system' produces something you can paste directly into your HR platform. Constraints eliminate the guesswork about what you need. They also prevent the AI from padding output with unnecessary caveats, disclaimers, or tangential information that you'll have to edit out anyway.

Click to enlarge

Frequently Asked Questions

• Q: Does it matter which AI tool I use for these techniques? A: The four-part prompt structure works across ChatGPT, Claude, Copilot, and Gemini. Each tool has different strengths. Claude tends to handle long documents and nuanced writing instructions particularly well, while Copilot integrates directly into Microsoft 365 workflows, but the core prompting principles apply to all of them.
• Q: How do I know when my prompt is specific enough? A: Ask yourself: if I handed this brief to a smart new employee who knew nothing about my situation, would they produce what I need? If the answer is no, add more context. The 'smart new employee' test is more reliable than any word count.
• Q: What if I don't know how to describe the format I want? A: Give the AI an example instead. Say 'Format the output like this example: [paste a real example you like].' AI tools respond extremely well to format examples, it's often faster and more accurate than trying to describe structure in words.
• Q: Is it bad to ask the AI to play a role it might not know well? A: For common professional roles. HR manager, marketing director, financial analyzt, project manager, role assignment works reliably because the AI has been trained on vast amounts of professional content from those fields. For very niche or highly specialized roles, verify outputs carefully and treat them as first drafts.
• Q: Why does the same prompt sometimes give different results? A: AI tools have a setting called 'temperature' that introduces variability, meaning they're designed to not always give the identical answer. This is intentional, because rigid repetition would make them less useful for creative and analytical tasks. If you need consistency, specify it: 'Use a consistent formal tone throughout' or 'Follow this structure exactly each time.'
• Q: Can I save prompts I've built so I don't have to rewrite them every time? A: Yes, and you should. Keep a simple document, a Word file, Notion page, or even a Notes app, with your best prompts organized by task type. ChatGPT Plus also has a 'Custom Instructions' feature where you can set persistent context that applies to every conversation, so you don't have to re-explain your role, industry, or preferences each time.

Key Takeaways from Part 2

1. Prompting is not about magic words, it's about giving the AI the context, role, format, and constraints it needs to do your specific task well.
2. Prompt length should match task complexity. The goal is completeness and structure, not word count.
3. When AI gives a bad answer, iterate within the conversation using targeted corrections rather than starting over. The conversation history is an asset.
4. The four-part prompt formula. Role, Task, Context, Format/Constraints, is the most reliable general-purpose structure for professional use.
5. Constraint-setting is a power move, not a limitation. Specific constraints eliminate guesswork and produce output you can actually use without heavy editing.
6. Save your best prompts as reusable templates. The time you invest in building one good prompt pays back every time you use it.

What You Believe About Prompting Is Probably Wrong

Most professionals approach AI prompting with three confident assumptions: that shorter prompts are cleaner and more professional, that AI tools understand context the same way a colleague would, and that once you find a prompt that works, you should never change it. All three beliefs are understandable. All three will quietly sabotage your results. The good news is that fixing these misconceptions takes about ten minutes, and the payoff shows up immediately in the quality of your outputs.

Myth 1: Short Prompts Are Better Prompts

The instinct makes sense. You've been trained to communicate efficiently. In email, in meetings, in Slack, brevity signals confidence. So professionals type 'write a summary of this report' and get a bland, generic output, then conclude AI isn't that useful. The problem isn't the AI. It's that a three-word instruction gives the model almost nothing to work with. It doesn't know your audience, your tone, your format preferences, or what 'summary' even means to you, one paragraph or ten bullet points?

Research from Stanford's Human-Centered AI group consistently shows that specificity in prompts correlates directly with output quality. When you add role context ('you are a senior marketing strategist'), format requirements ('give me five bullet points under 20 words each'), and audience framing ('for a CFO who hasn't read the report'), outputs become dramatically more usable. This isn't complexity for its own sake, it's the same briefing you'd give a junior employee before asking them to draft something.

Think of it this way: if you walked up to a new contractor and said 'write the proposal,' they'd have questions. Lots of them. AI models don't ask follow-up questions unless you tell them to, they just make assumptions and fill the gaps. Your job as the prompter is to eliminate the gaps before they get filled with something generic. More context almost always means better output.

Click to enlarge

Myth 2: AI Remembers What You Told It Earlier

This one catches almost everyone. You spend five minutes at the start of a ChatGPT session explaining your company, your role, your communication style, and your project goals. Then you have a productive exchange. You come back the next day, open a new chat, and get completely generic responses again. The context is gone. Most AI tools, including the free tiers of ChatGPT and Gemini, don't carry memory between separate conversations. Each new chat is a blank slate.

Even within a single conversation, context degrades. Long chat sessions can cause earlier instructions to receive less 'attention' from the model. If you set up detailed instructions at the start of a 40-message conversation, the AI may drift from them by the end. This is a real architectural constraint, not a bug someone forgot to fix. Understanding it changes how you work: you stop assuming the AI 'knows you' and start building reusable context blocks you can paste in quickly.

The practical fix is a personal context block, a short paragraph you keep in a notes app that describes your role, your company, your audience, and your tone preferences. Paste it at the top of any important AI session. ChatGPT Plus users can also use the 'Custom Instructions' feature to set persistent context. Claude Pro has a similar capability. Two minutes of setup at the start of a session saves you from ten minutes of mediocre outputs.

Myth 3: A Good Prompt Works the Same Way Every Time

Professionals who discover a prompt that works well often treat it like a template, copy, paste, done. The problem is that AI outputs have inherent variability. The same prompt on different days, on different platforms, or even in different sessions can produce meaningfully different results. Additionally, AI tools update their models regularly. A prompt optimized for GPT-4 in early 2024 may behave differently on GPT-4o or Claude 3.5 Sonnet. Treating prompts as permanent fixtures leads to stale, inconsistent results.

The better mental model is a prompt as a living document. Review your most-used prompts every few months. Test them when a tool announces a major update. Keep a simple log of what works, a notes doc or a Notion page is enough. Professional teams at larger companies are now building internal 'prompt libraries' for exactly this reason. You don't need a system that elaborate, but the habit of revisiting and refining is what separates consistent performers from occasional users.

What Actually Works

Effective prompting rests on three consistent habits. First, always specify a role, a format, and an audience, even for simple tasks. 'Write a follow-up email' becomes 'Write a follow-up email from a sales consultant to a mid-level procurement manager who attended a product demo yesterday. Tone: professional but warm. Length: under 150 words. End with a clear next step.' That takes 20 extra seconds to write and produces something you can actually send.

Second, treat your first output as a draft, not a final product. The most efficient AI users don't get one response and stop, they iterate. 'Make this more concise.' 'Change the tone to be more assertive.' 'Add a section on pricing objections.' Each instruction refines the output. Iteration isn't a sign that your prompt failed; it's how professional-grade outputs get built. A McKinsey study on generative AI adoption found that users who iterated on outputs reported significantly higher satisfaction with AI tools than those who used single-shot prompts.

Third, build a personal prompt library. Start with five to ten prompts you use regularly, weekly reports, meeting summaries, client emails, performance feedback drafts. Store them somewhere you can access in 10 seconds: Notion, Apple Notes, Google Docs. Each prompt should include your role, your context block, and the specific format you want. Over time, this library becomes one of the most valuable productivity assets you own, faster than templates, more flexible than macros, and completely personalized to your work.

Frequently Asked Questions

• Q: Does prompt length actually matter, or is this overthinking it? A: Length matters only insofar as it carries useful information. A 200-word prompt packed with role, context, format, and examples will almost always outperform a 10-word prompt. But a 200-word prompt full of repetition or vague adjectives won't help. Useful specificity is the goal, not word count.
• Q: Should I use the same prompts on ChatGPT and Claude? A: You can, and most prompts transfer well. But Claude tends to respond well to explicit reasoning instructions ('think through this step by step'), while ChatGPT Plus often handles structured formatting requests very cleanly. Test your most important prompts on both and note any differences.
• Q: How do I know if my prompt is actually good? A: Ask yourself: could a brand-new employee produce a usable first draft from this instruction alone? If the answer is no, if there are too many gaps, your prompt needs more context. If you'd have to explain it to a person, explain it to the AI.
• Q: Is it safe to paste real work documents into AI tools? A: Check your organization's policy first. Many companies restrict pasting confidential client data or internal financials into consumer AI tools. When in doubt, anonymize or paraphrase. Microsoft Copilot (integrated into Microsoft 365) and enterprise-tier ChatGPT are designed with data privacy protections for business use.
• Q: My outputs are inconsistent, same prompt, different results. Why? A: AI models have a temperature setting that introduces variability by design. This makes outputs feel more natural rather than robotic. For high-consistency tasks (like filling a structured template), ask the AI to 'follow this format exactly' and provide a clear example. Consistency improves with tighter format instructions.
• Q: How often should I update my prompts? A: Review them whenever a tool announces a major model update (e.g., GPT-4o, Claude 3.5), or when you notice outputs drifting in quality. A quarterly review of your top ten prompts takes about 30 minutes and keeps your library current.

Key Takeaways

• Short prompts force AI to guess, specificity in role, format, and audience is what drives usable outputs.
• AI tools do not retain memory between sessions. Build a reusable context block and paste it at the start of important conversations.
• Prompts are living documents. Review and refine them when AI tools update, and when your outputs start feeling generic.
• The Role + Context + Format structure works across every tool. ChatGPT, Claude, Gemini, Copilot, and Notion AI.
• Iteration is the professional workflow. Treat the first output as a draft and refine it with follow-up instructions.
• A personal prompt library is a concrete productivity asset. Five to ten strong, reusable prompts will save you hours each week.