The 30-Second Pitch
Most people “try AI” and get nice drafts. Pros run AI and get results with receipts: faster delivery, fewer errors, and outputs clients can audit. The difference isn’t secret models—it’s how you prompt: clear specs, the right context, and simple verifications. This article gives you punchy, real-life playbooks, copy-paste prompts, and pricing so you can turn prompting into profit this week.
The Engine (Simple, but Unfair)
Great prompts aren’t poetry; they’re systems. Use CLEAR+GSCP to make your results look superhuman and reproducible.
C onstraints: define the contract (schema, format, rules).
L ogging: keep traces of prompts, sources, and costs.
E vidence: require citations or calculations for claims.
A utomation: add verification + repair loops (don’t re-prompt manually).
R eview: uncertainty gates → escalate when confidence is low.
+ GSCP rails: task decomposition, tool policies, programmatic checks, self-improver cycles.
Copy-paste “spec-sandwich” you’ll use everywhere
ROLE & GOAL: You are a [role]. Produce a [artifact] that meets the contract.
CONTRACT (immutable):
- Output format: [JSON/markdown exact fields]
- Rules: [math must balance], [dates in ISO], [banned terms/style], [policy clauses]
- Refusal: If coverage or confidence is low, return "needs_review" with missing items.
CONTEXT: [paste docs/notes/numbers/links]
EVIDENCE: Cite sources or show calculations for each claim.
RETURN: Output only in the specified format.
Playbook 1. The Freelancer’s “Offer Factory” (48 hours → $
$)
Turn chaos into clean deliverables that clients can see and sign. Package as fixed-scope products with a short turnaround.
A) Pitch-Ready One-Pager (plus 6-slide deck)
Prompt
Goal: Create a one-page company brief + 6-slide outline that obeys the contract.
Contract:
- Sections: Problem (≤100w), Solution (≤120w), Proof (3 bullets), Pricing (3 tiers), CTA, Visual idea
- Tone: confident, concrete, no hype adjectives; avoid clichés
- Evidence: use only client notes; label assumptions explicitly
Return: Markdown with H2/H3; put slide outline at the end
Client notes: [paste]
If info is missing, list questions first, then best-effort draft.
Price it: $750–$2,000 package, 1 revision, 72-hour SLA.
Upsell: add branded PDF export +$200.
B) Website-in-a-Week Copy Pack
Sitemap, hero lines, FAQs, CTA variants, image prompts.
Quote $1,500–$3,500 fixed; show tokens/time saved in your handoff report.
Playbook 2. Sales & Ops Copilot (Daily Use = Durable ROI)
Drowning in notes and emails? Convert everything into tasks, owners, and due dates, with quotes as evidence.
Meeting → Tasks (JSON you can auto-post)
Prompt
Goal: Convert notes into actionable tasks.
Contract (JSON array of objects):
{title, owner, due_date, priority, blockers, source_quote}
Rules: ISO dates; owner must exist in roster; include exact quote from notes.
Roster: [names/emails]
Notes: [transcript/bullets]
If owner missing: owner:"UNASSIGNED", reason:"not found"
Return: JSON only.
Auto-verify: check date format + owner names before pushing to your task tool.
Manager digest prompt: “Wins, risks (with quoted reason), slipped deals (cause), next actions by owner.”
Playbook 3. Service Business Ops (Quotes → Invoices → Follow-ups)
Admin kills margins. Force math to balance and policy to stick— before anything reaches accounting.
Invoice Extractor (email/PDF → clean JSON)
Prompt
Goal: Extract invoice data with evidence.
Contract (JSON):
invoice_id, vendor, date, items[{desc, qty, unit_price, line_total}],
subtotal, tax, total, currency, needs_review
Rules:
- total = subtotal + tax
- line_total = qty * unit_price
- qty > 0; currency = ISO-4217
Evidence: Include source_quote for each number (page:line or email line).
Refusal: If any uncertainty, set needs_review:true with reasons.
Input: [paste text]
Return: JSON only.
Repair loop: If totals fail, re-prompt: “R3 failed: totals mismatch—recompute only line_total/subtotal/tax/total .”
Package: Setup $3,000 + $600/mo (≤300 invoices). KPI: posting time −70%, error rate ↓.
Playbook 4. Real Estate / Local Services (Listings & Replies)
Ground answers in approved info; perfect tone and length; never guess policies.
Listing → Q&A
Prompt
Goal: Create a 10-pair Q&A from listing docs.
Contract:
- Each pair: {question, answer(2–3 sentences), source_section}
- Tone: friendly, specific, zero clichés
Rules: If parking or pet policy is unknown, say "Contact agent" (no guessing).
Sources: [paste HOA/listing/map notes]
Return: JSON array only.
Inbound reply helper: 120–160 words, answer → propose viewing time → list two alternatives → single clear CTA.
Playbook 5. Support RAG that Refuses When Unsure
Deflect tickets and earn trust with citations and honest refusals.
Prompt
Goal: Answer user's question using only approved docs.
Contract:
- Provide answer, 2–4 citations, and a "confidence" score 0–1
- If coverage < 2 independent sources OR confidence < 0.7 → output "needs_review:true"
Sources: [doc chunks]
Return: JSON: {answer, citations[], confidence, needs_review}
Monthly report to client: deflection rate ↑, TSR (task success rate), escalation reasons, $ saved.
The “Make It Pop” Wrapper (Zero Design Needed)
Outputs that look pro win deals. Ask for styled markdown; render to PDF.
Prompt
Take this content and apply house style:
- H1/H2, short paragraphs, checklists
- Use tables for data; highlight decisions
- 2-page cap; keep facts unchanged
Return: Markdown only.
Metrics That Close Renewals (and Justify Raises)
Stop selling vibes. Show numbers.
TSR (Task Success Rate): % that pass contract without human fix.
Violation rate: schema/math/policy fails per 100 runs.
Escalation rate & minutes: what actually needed people.
Cost per successful task: tokens+tools / passes.
Coverage: % with sufficient sources (separate retriever vs. generator issues).
Use a one-page scoreboard in renewals: “TSR +12 pts, errors −40%, 14.6 hours saved—new scope attached.”
Pricing That Protects Margin
Pilot (2 weeks, 1 workflow): $1,500–$3,500 with SLA + weekly report.
Standard (monthly): ≤3 workflows, eval pack, support channel — $900–$2,500/mo.
Plus: multilingual + analytics — +$300–$600/mo.
Guardrail: keep COGS (inference + vector DB + storage) ≤30% of MRR.
Your Reusable Library (Moat You Own)
Build a private folder of contracts (schemas + rules) , prompt skeletons, and verifier snippets. Each client = remix, not reinvent.
Verifier snippet (JSON Schema + math)
Checks:
- required fields present/types correct
- sum(items.line_total) == subtotal
- subtotal + tax == total
- dates in ISO; currency in [ISO-4217]
On fail: emit rule_id + message; trigger targeted repair.
Uncertainty gate
If confidence < 0.7 OR citations < 2 unique sources → needs_review:true
7-Day Blitz (Start → Pilot → Invoice)
Day 1: Pick one workflow you touch weekly; write the one-page contract.
Day 2: Build the spec-sandwich prompt; test on 5 samples.
Day 3: Add JSON output + tiny verifier (schema + math).
Day 4: Add a repair prompt for the top failure case.
Day 5: Wrap with style prompt; render a clean PDF demo.
Day 6: Measure TSR/time/error on 10 runs; screenshot dashboard.
Day 7: Send 20 targeted emails offering a 2-week fixed-price pilot (include demo & metrics).
Outbound that gets replies
Subject: Cut [team] time on [workflow] by ~[X]%
We deploy a spec-verified AI workflow that delivers [outcome] with citations & audits.
2-week fixed-price pilot. 3-min demo + 1-page metrics?
—[You], [site/LinkedIn]
Common Pitfalls → Fast Fixes
Pretty but wrong: enforce schema + math; never accept free-form for structured data.
Hallucinations: require citations; refuse when coverage is weak.
Spec drift: version the contract; pin prompts to version tags.
Cost creep: log token/tool spend; add cheap pre-checks before expensive retrieval.
Prompt injection: strip/escape untrusted text; whitelist tool functions and arguments.
Conclusion — Make It Boringly Reliable, Then Make It Big
“Brain-blowing” results come from boring discipline: a contract, grounded context, simple verifiers, and a tiny scoreboard. Do it once, and your outputs look unfairly good; clone the pattern, and it turns into a business. Prompt engineering pays when it’s engineered.
Join the conversation! Your thoughts help the community grow.