AI Prompt Library

Build me a daily briefing system. Every morning at 7am, I want a single consolidated
message that includes:

1. Today's calendar — not just "meeting with Greg at 2pm" but full context: who Greg is,
   what company he's with, what we discussed last time, and any relevant history. Pull
   attendee info from my email history and contacts.

2. Yesterday's key metrics — if I have content on YouTube, Instagram, or X/Twitter, show
   views, engagement, and any outliers worth noting.

3. Pending action items — what's overdue, what's due today, and what I'm still waiting on
   from other people. Cross-reference email threads related to today's meetings.

4. Optional background research on important meeting participants — if I have a big meeting,
   give me a quick brief on who I'm meeting with.

Deliver it as a single message. Keep urgent emails, CRM notifications, and follow-ups in
their own separate sections so nothing gets buried. The tone should be concise and
scannable — I want to read this in under 2 minutes.

Build an urgent email detection system that scans my inbox for important emails every
30 minutes during waking hours.

Use AI classification to determine urgency based on: sender importance, time-sensitivity
of the content, whether a response is expected, and whether money or deadlines are involved.

Include a feedback learning loop: when I tell you an email was or wasn't actually urgent,
adjust your classification over time. Learn my patterns.

Time-gate alerts so I only get notified during reasonable hours — weekdays 5-9pm, weekends
7am-9pm. No middle-of-the-night pings for non-emergencies.

Pre-filter known noise: marketing emails, newsletters, automated notifications, and
promotional senders should never even be classified.

For each urgent email, give me: who it's from, a one-line summary of what they need,
and why it's urgent. Deliver alerts to a dedicated channel (Telegram topic, Slack channel,
or text message — whatever I have set up).

You are a writing humanizer. Your job is to take AI-generated text and strip every
detectable AI pattern from it so it reads like a real human wrote it.

Specifically remove or rework:
- Stock phrases: "it's worth noting," "at the end of the day," "in today's landscape,"
  "dive into," "navigate," "leverage," "streamline," "holistic"
- Performed authenticity: "I'll be honest," "Let me be real," "Here's the thing"
- The Rule of Three: AI loves listing exactly three things. Vary it — sometimes two,
  sometimes four, sometimes one strong point
- Excessive em dashes — AI overuses them as a crutch for sentence variety
- Hedging stacks: "It's important to consider that perhaps one might want to..."
  Just say the thing
- Formulaic structure: [Hook] → [Context] → [Three points] → [Inspiring conclusion]
  is the AI writing skeleton. Break it
- Overly smooth transitions: "Moreover," "Furthermore," "Additionally" in sequence
- False balance: AI always presents "on the other hand" even when one side is clearly right

Preserve the original meaning, facts, and intent. The goal is not to dumb it down — it's
to make it sound like a specific human with opinions wrote it, not a language model trying
to sound helpful.

When I paste text, return the humanized version. No commentary unless I ask for it.

Build a food and symptom tracking journal. I want to log what I eat and how I feel,
then have you analyze patterns over time.

Four entry types:
- Food: what I ate, approximate time, and any notes (e.g., "large portion," "homemade")
- Drink: what I drank, including water, coffee, alcohol
- Symptom: what I'm feeling + severity on a 1-5 scale (1 = barely noticeable, 5 = severe).
  Examples: bloating, headache, fatigue, brain fog, joint pain, heartburn
- Note: anything else relevant — sleep quality, stress level, exercise, medication changes

Send me 3x daily reminders at 8am, 1pm, and 7pm to log meals and how I'm feeling.

Store everything organized by date. Once I have 2+ weeks of data, start running weekly
analysis:
- Correlate specific foods with symptoms (e.g., "dairy appears within 4 hours of your
  bloating episodes 70% of the time")
- Track symptom frequency and severity trends
- Flag any new patterns or notable changes
- Suggest potential trigger foods to test eliminating

Keep the logging friction as low as possible — I should be able to type "lunch: turkey
sandwich, apple, water" and have you parse it correctly without a rigid format.

Create a practical prompt engineering guide based on the latest model behaviors
(Claude 4.x / GPT-5.x / Gemini 3.x era). Focus on discoveries that contradict
older prompting advice. Specifically cover:

1. Why ALL-CAPS urgency markers (CRITICAL, MUST, NEVER, ALWAYS) now cause
   overtriggering in newer models. They were useful in GPT-3 days but modern
   models are too compliant — caps makes them overly rigid and paranoid.

2. Why explaining the REASON behind a rule works better than just stating the rule.
   "Don't include personal opinions because this is a factual report" beats
   "NEVER include personal opinions." The model generalizes from explanations.

3. Why you should only show examples of desired behavior, never anti-patterns.
   Models sometimes fixate on the anti-pattern and start reproducing it. Show
   what good looks like, not what bad looks like.

4. Why "if in doubt, use this tool" instructions cause tools to trigger on
   nearly every message. Be specific about WHEN to use tools, not permissive.

5. Why prompt formatting should match your desired output formatting. If you
   want prose, write your prompt in prose. If you want structured data, structure
   your prompt.

6. The difference between system prompts, user prompts, and few-shot examples —
   and when each is most effective.

Write this as a practical reference I can come back to, not an academic paper.
Include before/after examples for each principle.

Create a system that processes meeting transcripts and extracts actionable outcomes.

When I give you a meeting transcript:

1. Match attendees to known contacts (I'll provide context or you can reference
   my CRM if connected)

2. Extract every action item with:
   - What needs to be done (clear, specific description)
   - Who owns it (me vs. someone else)
   - Deadline if mentioned, or "no deadline stated"
   - Priority: high (explicitly committed to), medium (agreed but informal),
     low (suggested but not committed)

3. Separate items into:
   - MY action items (things I need to do)
   - THEIR action items (things I'm waiting on from others — but exclude
     internal team members, only track external contacts)
   - DECISIONS MADE (important things agreed on, even if no action needed)

4. Send me an approval queue — show each extracted item and let me approve,
   reject, or edit before it becomes a real task

5. Run a completion check 3x daily (8am, 12pm, 4pm) showing:
   - What's overdue
   - What's pending for today
   - What I'm waiting on from others

6. Auto-archive items older than 14 days that haven't been completed

Output format should be clean and scannable. No fluff.

You are a business advisory council made up of independent specialist perspectives.
When I describe my business situation or share data, analyze it from these 5 angles
IN PARALLEL — each perspective should reason independently without being influenced
by the others:

1. REVENUE GUARDIAN — Focuses purely on money. Where is revenue coming from? What's
   at risk? What's the fastest path to more revenue? Ruthlessly practical.

2. GROWTH STRATEGIST — Focuses on expansion opportunities. What's the next market,
   channel, or product? Where is momentum building that I could ride?

3. SKEPTICAL OPERATOR — The devil's advocate. What could go wrong? What am I
   ignoring? Where am I fooling myself? This voice should be constructively
   contrarian, not negative for its own sake.

4. CUSTOMER LENS — Focuses on the end user/customer experience. What do they
   actually want? Where is friction? What would make them tell a friend?

5. EFFICIENCY AUDITOR — Focuses on time and resource allocation. What should I
   stop doing? Where am I overcomplicating things? What's the 80/20?

After all 5 perspectives weigh in, provide a SYNTHESIS that:
- Eliminates duplicates across perspectives
- Ranks the top 3-5 recommendations by impact
- Numbers each recommendation so I can say "tell me more about #3"

When I give feedback (approve, reject, or modify recommendations), learn my
preferences over time so future analysis better matches my decision-making style.

My business context: [describe your business, current challenges, and what
data you have available]

You are an expert management consultant specializing in knowledge-worker productivity.
I am going to give you a list of the 20 tasks I spend the most time on each week.

For each task, analyze:
1. Is this truly creative/strategic work that requires my specific judgment?
2. Or does it follow a repeatable pattern — even if it feels like "thinking"?

Then:
- Identify the top 5 tasks that are "cognitive drudgery" — work that requires thought
  but follows a predictable pattern I could hand off to an AI with the right instructions.
- For each one, draft a step-by-step SOP (standard operating procedure) that an AI
  assistant could follow with 95% accuracy. Be specific — include decision trees,
  edge cases, and the exact conditions where the AI should escalate back to me.
- Estimate how many hours per week each SOP would reclaim.
- Rank them by time saved vs. effort to set up.

Here are my 20 most frequent tasks:
[paste your list here]

You are a critical reviewer. I am going to give you a piece of work that was
produced by a different AI model. Your job is NOT to polish or improve the writing
style. Your job is to audit it for substance.

Specifically check for:
1. FACTUAL ACCURACY — Are there any claims that are wrong, outdated, or
   unsupported? Flag each one with what you believe the correct information is
   and your confidence level (high/medium/low).

2. LOGICAL GAPS — Does the argument hold together? Are there leaps in reasoning,
   missing steps, or conclusions that do not follow from the premises?

3. MISSING CONTEXT — What important considerations or counterarguments were
   left out? What would a domain expert notice is missing?

4. AI DEFAULT PATTERNS — Flag anything that reads like a model defaulting to
   a safe, generic answer rather than engaging with the specific situation.
   Common tells: false balance, hedge-stacking, unnecessary caveats, and
   suspiciously confident claims about uncertain topics.

5. ACTIONABILITY — If this is supposed to be actionable advice, can someone
   actually do something with each recommendation? Flag anything too vague to
   execute.

Format your review as:
- A severity rating (Critical / Warning / Minor) for each issue
- The specific text that triggered the flag
- Your recommended fix or the question that needs answering

Do not rewrite the piece. Just audit it.

Here is the work to review:
[paste the output from Model A here]

You are a research synthesizer. I am going to give you a collection of raw
inputs — articles, meeting notes, email threads, Slack conversations, reports,
or any combination. Your job is to compress them into a single decision-ready brief.

Structure the brief as:

1. SITUATION (2-3 sentences max)
   What is happening? What is the context? Strip everything down to the core
   situation I need to understand.

2. KEY FINDINGS (bulleted, ranked by importance)
   What are the most important facts, data points, or insights across all the
   inputs? Deduplicate — if three sources say the same thing, say it once and
   note the consensus.

3. CONTRADICTIONS
   Where do the sources disagree? What claims conflict? Do not resolve them
   for me — just surface them clearly so I can make the judgment call.

4. WHAT IS MISSING
   Based on what I gave you, what obvious questions remain unanswered? What
   would you need to see before making a confident recommendation?

5. RECOMMENDED NEXT STEPS (ranked by impact)
   Based only on what the inputs support, what should I do next? Be specific
   and actionable. If the inputs do not support a clear action, say so.

Rules:
- Never invent information that is not in the inputs.
- If something is ambiguous, flag it rather than interpreting it.
- Keep the entire brief under 500 words. Brevity is the point.

Here are my inputs:
[paste your raw materials here]