The PROMPT Anatomy Framework: Prompt Engineering for Goal Setting

There’s a structural reason most AI goal-setting conversations produce mediocre output: the user asks for an outcome without providing the inputs the AI needs to reason carefully about it.

Large language models are trained on vast amounts of human-generated text, but that doesn’t mean they can infer your specific situation from a brief request. When Wei et al. (2022) demonstrated that chain-of-thought prompting—structuring a prompt to walk through intermediate reasoning steps—dramatically improved performance on complex tasks, the finding was generalizable: structure in the prompt produces structure in the thinking.

PROMPT Anatomy is an application of that principle to goal setting. It is a six-component framework that turns a vague request into a structured brief. The AI doesn’t become smarter; it receives better inputs. That’s the mechanism.

Why Frameworks Beat Intuition for Prompt Design

Most people write prompts the way they write text messages: optimized for speed, minimal context, trusting the other party to fill in the gaps. That works in human-to-human communication because the other party can ask clarifying questions and already shares large amounts of context with you.

AI models do ask clarifying questions if you tell them to—but by default, they fill gaps with the statistical average of how humans usually respond to similar prompts. For goal setting, that average produces suggestions applicable to anyone, which means optimal for no one.

The OpenAI prompt engineering guide and Anthropic’s own documentation converge on a set of recurring recommendations: be specific about the task, provide relevant context, specify the output format, and use clear delimiters to separate components. PROMPT Anatomy operationalizes these recommendations into a memorable structure you can apply consistently.

The Six Components, Explained in Depth

P — Persona: Situate Yourself Specifically

Persona is not your job title. It is the set of facts about your current situation that are relevant to the goal at hand.

Compare these two Persona statements:

Weak: “I’m a marketing manager.”

Strong: “I’m a marketing manager at a 45-person B2B software company. I manage a team of three. My biggest challenge is that we generate good MQL volume but lose deals late in the sales cycle—I’m trying to figure out whether that’s a marketing attribution problem or a handoff problem.”

The second statement tells the AI something it can reason about. It narrows the domain of relevant advice and eliminates thousands of generic suggestions that simply don’t apply.

The Persona component should answer: What is your role? What is your current state in the relevant domain? What is the core tension you’re trying to resolve?

R — Resources: Constraints Are Data

Resources is the component most consistently omitted, and its absence is what generates goals that sound good on paper and collapse in practice.

Resources means: what you have available to work with. This includes time (the non-negotiable constraint), money, skills, tools, team or support, and relationships. It also includes what you do not have.

[Resources] I have 6 hours per week to dedicate to this goal. My budget is $0 for tools or advertising for the next 90 days. I'm a strong writer but have no technical development skills. I'm working alone—no team or VA.

When you provide constraints, the AI can distinguish between goals that are genuinely achievable in your situation and goals that look reasonable without context. Without this information, the AI cannot make that distinction.

O — Objective: Name the Actual Outcome

The Objective component specifies not just what you want to achieve, but what kind of output you want from this particular conversation with the AI.

These are different:

• “Help me set better goals” is a vague task.
• “Generate three 90-day goals for growing my freelance income, each with a measurable outcome and a leading indicator I can track weekly” is a specific output specification.

The second tells the AI what success looks like for this conversation. It specifies the number of goals, the time horizon, the structure of each goal, and the type of metric required. The AI has no ambiguity about what to produce.

Write your Objective in terms of: how many outputs, what structure they should take, what time horizon they should cover, and how you’ll use them.

M — Mode: Override the Default

Every AI model has a default conversational mode: cautious, balanced, hedged. That default is useful for many tasks. For goal setting, it often produces output that validates your existing thinking instead of challenging it.

Mode is where you override the default.

Challenge mode:

[Mode] Challenge my framing. If the goals I seem to be aiming for are lower-leverage than alternatives I haven't mentioned, point that out. I'd rather hear an uncomfortable truth now than discover it in three months.

Question-first mode:

[Mode] Before generating any goals, ask me 3-5 questions to make sure you have what you need. Prioritize questions that would change which goals you suggest.

Structured-output mode:

[Mode] Produce your output as a table with columns: Goal | Measurable Outcome | Weekly Action | Failure Risk. No prose summary.

Specifying mode prevents the AI from interpreting ambiguity charitably—which, in goal setting, usually means suggesting goals that are too safe.

P — Parameters: Set the Boundaries

Parameters define the scope constraints on the output itself. These are different from Resources, which describes your situation. Parameters describe the shape of the response you want.

Useful parameters for goal setting:

• Number of goals (“limit to three”)
• Time horizon (“for the next 90 days only, not longer-term”)
• Domains to include or exclude (“professional goals only, not health or personal”)
• Level of ambition (“each goal should be achievable but require consistent effort—not trivial, not heroic”)
• Output length (“one paragraph per goal, no more”)

[Parameters] Generate exactly 3 goals. Time horizon: 90 days. Focus only on revenue-generating activities—exclude brand-building and content creation for now. Each goal should be achievable without working more than my current hours.

Parameters prevent the AI from scope-creeping into adjacent areas or generating more output than you can process.

T — Tests: The Self-Evaluation Step

Tests is the component that adds the most value for the least effort—and is most consistently skipped.

The principle: ask the AI to evaluate its own output against explicit criteria before presenting it to you. This adds a built-in quality check that catches the most common goal-quality problems.

[Tests] Before presenting the goals, verify:
1. Each goal is specific enough that a neutral observer could determine in 30 seconds whether I achieved it.
2. Each goal is achievable within the constraints I described (especially the time constraint).
3. Each goal measures an outcome, not just an activity.
4. The three goals don't compete for the same limited time or energy.

If any goal fails one of these tests, revise it before showing me. Show me the original and the revised version side-by-side.

The Tests component works because it forces explicit evaluation rather than relying on the model’s implicit quality control. It also surfaces the reasoning behind any revisions, which helps you understand the thinking rather than just receiving the output.

Three Complete Worked Examples

Example 1: Career Transition

[Persona] I'm a software engineer, 6 years in, currently individual contributor at a mid-size startup. I want to move into an engineering manager role. I've been told informally that I'm "not quite ready" but without specifics.

[Resources] I have 3 hours per week I can commit to deliberate development toward this transition. No formal management training. I have one informal mentorship conversation per month with a senior EM at another company.

[Objective] Generate 3 specific goals for the next 6 months that, if achieved, would build the skills and evidence most relevant to an EM promotion case. Each goal needs a measurable outcome and a monthly checkpoint.

[Mode] Be direct. If the areas I'm implicitly targeting (technical leadership) are less important for EM readiness than other areas I haven't mentioned, say so.

[Parameters] 3 goals maximum. 6-month horizon. Professional domain only. The goals should build evidence visible to my manager and skip, not just internal development.

[Tests] Verify each goal is outcome-based (not activity-based), is achievable in 3 hours/week, and would plausibly be mentioned in a promotion review. Revise any that fail.

Example 2: Health and Fitness

[Persona] I'm a 38-year-old knowledge worker. Sedentary job, 50-hour weeks. I've tried and abandoned running programs twice in the past two years. My goal is sustainable fitness, not performance.

[Resources] I have 30 minutes per day available on weekday mornings before 7am. No gym membership, but I own a pull-up bar and some dumbbells. My main constraint is inconsistency when work gets busy.

[Objective] Give me 2 health goals for the next 60 days. Each should have a measurable outcome and be explicitly designed around the constraint of inconsistency—I need a goal structure that doesn't collapse the first time I miss a week.

[Mode] Be realistic. Don't optimize for maximum progress—optimize for what I'll actually maintain for 60 days given my track record.

[Parameters] 2 goals only. 60-day horizon. No outcome goals tied to weight or aesthetics—focus on process metrics I control.

[Tests] Before presenting, check: Could I maintain these goals during a stressful work week? Would missing one week make the goals unachievable? If yes to either, revise.

Example 3: Creative Project

[Persona] I'm a researcher who wants to write a book. I have a clear thesis and about 15,000 words of notes. I've never written a book before. My main fear is losing momentum after the initial excitement fades.

[Resources] 45 minutes per weekday morning, 2 hours on Saturday mornings. No deadline, no publisher—this is for myself with the aspiration of publishing eventually.

[Objective] Give me a writing goal structure for the next 6 months that produces a complete first draft. Break it into phases. Each phase should have a completion criterion.

[Mode] Focus on sustainability over speed. I don't care if the draft takes 9 months instead of 6—I care that I don't abandon it.

[Parameters] 6-month plan. Defined phases with completion criteria. Account for a realistic assumption that I'll miss 20% of planned sessions.

[Tests] Check whether the word count targets are achievable in my available time, accounting for the 20% miss rate. If not, revise downward.

The Minimum Viable Version

When you need a quick session rather than a deep planning conversation, P+R+O is the minimum viable version of the framework.

[P] I'm a [role/situation in 2-3 sentences].
[R] I have [time/budget/constraints].
[O] Give me [number] goals for [timeframe], each with [specific output format].

This three-component version removes the most common failure modes—missing context and vague output specs—without requiring a full five-minute prompt.

Add Mode when you want the AI to behave differently from its cautious default. Add Parameters when you need strict scope control. Add Tests when you want self-evaluated output before you see it.

Beyond Time’s built-in prompt templates are structured around PROMPT Anatomy components, which means you can run structured goal-setting sessions without writing prompts from scratch—a practical starting point at beyondtime.ai.

Your action for today: Take your most important current goal, write a full PROMPT Anatomy prompt for it using all six components, and run it. Compare what the AI produces to how you originally wrote that goal.

Related:

• The Complete Guide to AI Prompts for Goal Setting
• How to Write AI Prompts for Goal Setting
• 5 Prompt Styles for Goal Setting Compared
• What Makes an AI Prompt Effective for Goals

Tags: prompt engineering framework, PROMPT anatomy, AI goal setting, structured prompts, goal setting with AI