Curate Context and Streamline PRDs with the "What Does This Token Buy?" Rule

Streamline PRDs and Context with the "What Does This Token Buy?" Rule

The previous two lessons covered personas and overall prompt length. This lesson extends the curation principle to the entire context payload - from a short exchange to an 8-page PRD.

Running Example: The Product Manager of FreelanceFlow needs AI to explain a billing charge to a customer, and to draft a quick-start guide from an 8-page PRD.

1. Eliminate the 4 Context-Stuffing Traps That Dilute Model Signals

Extending beyond personas, the entire context payload must be disciplined. When engineering production prompts, PMs must actively eliminate 4 stuffing anti-patterns:

1. Repeating instructions for "emphasis": Stating a rule at the top, repeating it in the middle, and shouting "NOTE AGAIN" in ALL CAPS at the end does not increase adherence weights - it consumes token budget and drowns out other constraints.
2. Enumerating endless exception lists: Dumping 20 hypothetical edge cases causes uneven compliance (the model remembers rule #3 but misses rule #17). Provide core decision principles paired with 2–3 high-variance examples instead.
3. Dumping full schemas or unabridged background docs: Pasting a 50-page specification just to answer a refund policy question buries key signals in irrelevant tokens.
4. Ignoring token opportunity cost: Every uninformative token consumes attention budget that should belong to critical decision parameters.

Production Example: A FreelanceFlow user asks: "Why was my Pro plan charged an extra $5 this month?"

• Maximize Approach: Pasting all 30 pages of terms of service, international tax charts, and fee schedules across 10 tiers.
• Curated Approach: Extracting the exact 2-sentence cross-border currency processing rule for Pro tiers, paired with 1 boundary constraint: "Explain the fee rationale based on multi-currency terms; do not initiate a refund unless explicitly requested."

2. Streamline PRDs and Background Docs with the "What Does This Token Buy?" Rule

When using AI across the Product Development Life Cycle (PDLC) - generating user stories, drafting test scenarios, or summarizing specs for stakeholders - the PM's core skill is extracting high-signal behavioral drivers from source docs.

Before injecting any text block into a prompt, apply the counter-intuitive test: "If I delete this section, will the AI output degrade or deviate?" If the answer is no, that section must be cut.

Exercise 1.3: You have an 8-page PRD for FreelanceFlow's "Automated Income Tax Withholding by Invoice" feature. You need AI to draft a 4-step Quick Start Guide for end users. Determine:

1. Which PRD sections are mandatory context?
2. Which PRD sections must be cut to prevent signal dilution?
3. Draft the complete prompt following Curated Context standards.