Five-part prompt architecture

Nearly everyone in your job market has access to the same AI tools now. So why do some job seekers pull gold out of them while everyone else gets mush? Here's the answer, and it's the whole lesson: the edge isn't the model. The edge is how you engineer the conversation. Most people use AI like a search box — vague question in, vague answer out. And today I'm going to teach you to use it like a collaborator with a spec, because that difference shows up in every single application you send. Now, one piece of context from the hiring side of the desk before we build anything. Employers' screening software reads for meaning these days, and the humans behind it have gotten very good at smelling generic, machine-generated sameness. So the goal of everything in this lesson isn't to have AI do your work for you. It's to have AI sharpen work that's genuinely yours. Keep that straight and these tools become your superpower. Get it backwards and you're just adding your voice to the noise. So here's the architecture, and I want you to memorize it: five parts. Context — your level, your industry, your target. Role — who the AI should play, like a VP interviewing you or a skeptical recruiter. Task — exactly what to produce. Constraints — the tone, the length, and what it must never invent. And output — the format you want back, a table, a score, a list. Now, want to see the difference? Bad prompt: help me prepare for my interview. Good prompt: you're a VP of engineering at a growing fintech company, you're interviewing a senior backend developer, you value ownership — ask me five behavioral questions, and after each answer, rate me one to ten and tell me what was missing. See it? One's a wish. The other's a spec. Now, two workhorse prompts from your library. First, the recorded-screen simulator, because so many first interviews now happen on camera where there's nobody on the other end, and those reward pace, structure, and calm. So paste in the job description, add what you know about your likely interviewers, and run your three rounds — screening, technical, leadership. After every answer, demand your score, the gap, and a stronger example answer. Then repeat until your scores level off, usually eight to ten runs. The simulator is for reps, not truth — it won't know the company's secret rubric, but it will absolutely break your um habit and force structure under time pressure. And here's my bright line, same as always: the machine is your training partner before the interview, never a whisper in your ear during one. Employers are actively watching for that now, and getting flagged costs you everything. Second, the semantic gap analyzer. Job postings don't spell everything out — they carry implied requirements, like strategic partner or comfort with ambiguity, that never show up as keywords, and the software reading your resume infers meaning the same way. So have the AI compare your resume against the posting and return the explicit gaps, the tonal mismatches, and the seniority signals you're missing — then rewrite one bullet to show scope, action, number. And protect yourself while you do it: placeholders for company names and metrics, never anything confidential. Does that make sense? Now picture two candidates chasing the same role. One types help me prep, skims whatever comes back, and that's the entire strategy. The other builds the spec, runs eight simulator rounds, closes her top three gaps, and walks in already having answered the hard questions eight times. Same tools — so how does one of them walk in that much readier? The spec. That's the gap you're closing today. And listen, the honest part: the model is a mirror, not an oracle. And cross-check anything about salary or the market with two human sources you trust. Never let it invent an employer, a title, or a number you didn't give it. And every real decision — stay or go, apply or pass, accept or decline — stays yours, because the machine reflects. It doesn't decide. Using AI without a spec is like hiring a world-class chef and telling him, make food. You'll get something. You just won't get what you needed. So here's your assignment, and it takes thirty minutes. Pick one live target role with a posted description. Run the gap analyzer and save your top three gaps. Fix one resume bullet — scope, action, number. Then run round one of the simulator and write down your score, with the date on it. That log entry's your proof that you actually practiced, not just read about practicing. Go run that first round.