The Complete AI Tricks Playbook

40 AI TricksFor Business Automation

A comprehensive playbook spanning Sales, Marketing, Operations, Finance, Legal, and Product & Engineering — plus prompting, agentic workflows, and advanced techniques for Claude or ChatGPT.

Turn your AI assistant — Claude or ChatGPT — into specialized agents for any business function.

Each trick includes complete system instructions, example prompts, and best practices

Copy, paste, and start automating immediately

Free Weekly Workshop

Want to Master AI Agent Skills?

Join our live workshop and learn how to build, customize, and deploy these tricks in Claude Code. No coding required.

The Building Blocks

What is an AI Trick?

An AI Trick is a reusable instruction — a prompt configuration, a skill file, or a Custom GPT — that turns Claude or ChatGPT into a specialized agent for a specific task. Each one includes:

System Instructions

The role and rules your AI assistant follows for this specific task

Input Templates

Structured data your AI assistant expects to receive

Output Format

How results are structured and delivered

Context Files

Reference documents your AI assistant uses for context

Recommended File Structure

~/.claude/skills/
├── sales/
│   ├── deal-qualification.md
│   ├── proposal-customization.md
│   ├── battle-cards.md
│   ├── account-research.md
│   └── win-loss-analysis.md
├── marketing/
│   ├── campaign-diagnostics.md
│   ├── content-repurposing.md
│   ├── brand-voice.md
│   ├── competitive-positioning.md
│   └── event-followup.md
├── operations/
│   ├── sop-documentation.md
│   ├── vendor-evaluation.md
│   ├── meeting-summary.md
│   └── change-management.md
├── finance/
│   ├── variance-analysis.md
│   ├── expense-review.md
│   ├── modeling-assumptions.md
│   └── audit-prep.md
├── legal/
│   ├── contract-redline.md
│   ├── compliance-check.md
│   ├── nda-summary.md
│   └── risk-assessment.md
└── product/
    ├── feedback-synthesis.md
    ├── tech-docs.md
    ├── retro-analysis.md
    └── release-notes.md

Works in Claude or ChatGPT

Every trick here is written to be tool-neutral. A handful of mechanisms just go by different names in each tool — here is the one-time mapping, so you never have to think about it per trick:

Reusable instruction

Claude: a skill file at ~/.claude/skills/… or a Project · ChatGPT: a Custom GPT's instructions or a Project's custom instructions

Iterative deliverables

Claude Artifacts ↔ ChatGPT Canvas

Persistent context

Claude Projects ↔ ChatGPT Projects / Custom GPTs

Connected tools

Claude MCP ↔ ChatGPT connectors / Custom GPT Actions

Where a step genuinely differs, the trick shows a short Claude vs. ChatGPT note. A few power-user tricks are Claude-specific and carry a Claude only badge with the nearest ChatGPT equivalent.

Sales Tricks (1–5)

Deal Qualification Scoring

Analyze discovery call transcripts against your ICP criteria and output a qualification score with reasoning.

Required Inputs

ICP criteria document (firmographics, technographics, buying signals)
Discovery call transcript or notes
Scoring rubric (what earns points, what deducts)

Proposal Customization

Generate tailored proposals by matching prospect needs to relevant case studies and applying pricing logic.

Required Inputs

Case study library (indexed by industry, use case, company size)
Pricing calculator rules
Prospect discovery notes
Proposal template

Competitive Battle Cards

Maintain living competitive intelligence documents and generate contextual objection responses.

Required Inputs

Competitor profiles (features, pricing, positioning)
Win/loss data against each competitor
Product comparison matrix
Objection response library

Account Research Briefs

Compile pre-call intelligence from multiple sources into an actionable briefing document.

Required Inputs

Company name and key contacts
News API access or saved articles
10-K/financial filing summaries
LinkedIn activity (if available)
Previous interaction history from CRM

Win/Loss Analysis

Process CRM notes and call recordings to identify patterns in why deals are won or lost.

Required Inputs

CRM export of closed deals (won and lost)
Call recordings or transcripts
Deal metadata (size, sales cycle, competitor, rep)
Exit survey responses (if available)

Marketing Tricks (6–10)

Campaign Performance Diagnostics

Analyze marketing metrics to identify root causes of underperformance and recommend optimizations.

Required Inputs

Campaign metrics (impressions, clicks, conversions, spend)
Historical benchmarks
Audience and targeting parameters
Creative assets used
Landing page analytics

Content Repurposing Engine

Transform long-form content into multiple formats optimized for different channels.

Required Inputs

Source content (webinar transcript, blog post, whitepaper)
Target channels and their requirements
Brand voice guidelines
Content calendar constraints

Brand Voice Enforcement

Audit content against brand guidelines before publication to ensure consistency.

Required Inputs

Brand voice guidelines
Terminology do's and don'ts
Tone descriptors with examples
Style guide (formatting, punctuation)
Draft content to review

Competitive Positioning Analysis

Track competitor movements and identify positioning whitespace opportunities.

Required Inputs

Competitor website/messaging archives
Recent competitor announcements
Market research reports
Customer feedback on competitors
Your current positioning

Event Follow-Up Sequences

Generate personalized post-event outreach based on interaction context.

Required Inputs

Event attendee list with interaction notes
Sessions attended (if tracked)
Booth conversation summaries
Lead scoring criteria
Follow-up templates by segment

Operations Tricks (11–14)

SOP Documentation

Transform process walkthroughs into formatted, version-controlled playbooks.

Required Inputs

Process recording or step-by-step narration
Screenshots or system references
Exception handling requirements
Roles and permissions involved

Vendor Evaluation Matrices

Score vendor/RFP responses consistently against predefined criteria.

Required Inputs

Evaluation criteria with weights
RFP responses from vendors
Technical requirements
Business requirements
Reference check notes

Meeting Summarization

Transform meeting recordings into structured action items with owners and deadlines.

Required Inputs

Meeting recording transcript
Attendee list with roles
Meeting agenda (if available)
Previous meeting notes (for context)

Change Management Communications

Generate stakeholder-appropriate messaging for organizational changes.

Required Inputs

Change description and rationale
Stakeholder groups affected
Timeline and milestones
FAQ from leadership
Talking points for managers

Finance Tricks (15–18)

Variance Analysis Narratives

Transform financial variances into board-ready explanations with business context.

Required Inputs

Budget vs. actuals data
Prior period comparisons
Business context (initiatives, market changes)
Materiality thresholds
Board presentation template

Expense Policy Review

Audit expense submissions against company policy and flag exceptions with reasoning.

Required Inputs

Expense report data
Company expense policy
Approval matrix
Historical exception patterns
Employee context (role, project)

Financial Modeling Assumptions

Document, validate, and stress-test the assumptions underlying financial projections.

Required Inputs

Financial model or projections
Historical data for benchmarking
Industry comparables
Market research
Strategic plan inputs

Audit Preparation

Compile supporting documentation for audit requests from general ledger and policy documents.

Required Inputs

Auditor's PBC (Prepared by Client) list
General ledger export
Policy documents
Contract repository
Prior year workpapers

Legal Tricks (19–22)

Contract Redlining

Compare incoming contracts against your playbook and suggest markup with fallback positions.

Required Inputs

Incoming contract
Contract playbook (standard positions)
Fallback position matrix
Approved exceptions
Risk tolerance by term

Policy Compliance Checks

Review documents against regulatory requirements and cite specific provisions.

Required Inputs

Document to review
Applicable regulations/policies
Compliance checklist
Prior findings (if any)
Regulatory guidance documents

NDA/MSA Summarization

Extract key terms from agreements into a standardized tracker format.

Required Inputs

Contract document (NDA, MSA, etc.)
Term extraction template
Obligation tracking fields
Alert thresholds (expiration, renewal)

Risk Assessment Memos

Generate structured risk analysis for business decisions requiring legal review.

Required Inputs

Proposed action/decision
Relevant legal framework
Business context and objectives
Risk tolerance parameters
Prior precedents (if any)

Product & Engineering Tricks (23–26)

User Feedback Synthesis

Process support tickets and reviews into prioritized feature themes.

Required Inputs

Support ticket export
App store reviews
NPS/survey responses
Feature request log
Product roadmap (for context)

Technical Documentation

Transform code comments and architecture decisions into readable documentation.

Required Inputs

Source code files
Architecture decision records (ADRs)
API endpoints
Database schemas
Existing docs (to update vs. create)

Sprint Retrospective Analysis

Identify patterns across retrospectives to surface systemic issues.

Required Inputs

Retrospective notes (multiple sprints)
Sprint metrics (velocity, completion rate)
Team composition changes
Project/initiative context

Release Notes Generation

Transform commit messages and tickets into customer-facing announcements.

Required Inputs

Git commit log
Merged PRs with descriptions
Linked tickets/issues
Feature flags changed
Previous release notes (for format consistency)

Prompting & Reasoning Tricks (27–31)

Adaptive Reasoning Protocol

Install one custom instruction that makes your AI assistant size up each request and switch reasoning methods automatically — decomposing problems to solve them, teaching plainly to explain them.

Required Inputs

An AI assistant that supports custom instructions (Claude Projects or account Settings; ChatGPT custom instructions, a Project, or a Custom GPT)
The Adaptive Reasoning Protocol text (provided below)
One test task to solve or debug, and one to explain, so you can watch it switch modes

High-Context Prompt Framing

Turn vague replies into usable ones by structuring every prompt around three things: a plain ask, the background your AI assistant cannot see, and an example of the output you want.

Required Inputs

A one-line statement of the task (the plain ask)
The context your AI assistant lacks (who, what, constraints, tone)
An example of the result you want, if you have one

Projects with Custom Instructions

Create a reusable Project that loads your reference files and standing custom instructions into every chat, so you stop re-explaining who you are and how you want answers.

Claude vs. ChatGPT

Persistent context ports across tools: a Claude Project maps to a ChatGPT Project or a Custom GPT. Upload the same reference files and paste the same custom instructions there.

Required Inputs

A tool with Projects — Claude Pro, or ChatGPT (Projects / Custom GPTs)
A few reference files (a style guide, past examples, key facts)
Your custom instructions: role, audience, and response preferences

Artifacts for Iterative Deliverables

Have your AI assistant build documents, tables, checklists, or small apps in a side panel you can read, copy, and revise in place — instead of watching results scroll away in the chat.

Claude vs. ChatGPT

Claude's Artifacts side panel maps to ChatGPT's Canvas. Ask for one concrete deliverable and refine it in place in either tool.

Required Inputs

A concrete deliverable to produce (document, table, checklist, or small app)
The specific edits you want as you iterate

Deliberate Memory & Privacy Controls

Set your AI assistant's history, memory, and training controls on purpose — and make a personal rule about what is safe to paste — before you trust it with real work.

Claude vs. ChatGPT

Both Claude and ChatGPT have memory plus conversation-history and data-training controls in Settings. Set them deliberately in whichever you use.

Required Inputs

Access to Settings (Privacy / data controls and Memory)
A decision on the model-training toggle for your kind of work
A personal rule for what not to paste

Agentic Workflows with Cowork (32–36)

Claude only

Enable Connections (MCP)

Connect Claude Desktop to your real tools — Gmail, Calendar, Drive, Linear, Stripe — through MCP so it can read and act in them instead of only talking about them.

Claude vs. ChatGPT

MCP is Anthropic's open connection protocol. Nearest ChatGPT equivalent: connectors and Custom GPT Actions, which wire ChatGPT to outside tools in a similar spirit.

Required Inputs

Claude Desktop app, signed in on Pro
A tool you actually use (e.g. Google Calendar) and permission to connect it
A verification question only that connection can answer

Six-Part Workflow Instruction Framework

Turn a vague wish into a repeatable routine by writing workflow instructions that cover all six parts: purpose, trigger, steps, output format, quality checks, and error handling.

Required Inputs

A Project to hold the instructions
The workflow's purpose, trigger phrase, and ordered steps
The exact output format, quality checks, and an error plan

Map Each Workflow Need to a Connection

Before running a workflow, list everything it must do and match each need to a live connection, so the run does not stall halfway because a tool was not wired up.

Claude vs. ChatGPT

Connections differ by name: Claude uses MCP; ChatGPT uses connectors and Custom GPT Actions. Map each workflow need to whichever your tool provides before you run it.

Required Inputs

The list of actions your workflow performs
The connection each action requires
Confirmation each connection is enabled and reachable from the Project

Run & Steer a Workflow Live

Trigger a workflow, watch each step, correct drift in plain language while it runs, and save good outputs as reference examples for better future runs.

Required Inputs

A Project with workflow instructions and its connections in place
The trigger phrase from your instructions
A place to save good outputs as examples

Workflow Scorecard & Iteration

Score each workflow run against a few metrics, fix the lowest-scoring part, and once it is reliable, capture the refined instructions as a reusable skill or saved prompt.

Claude vs. ChatGPT

Save a proven workflow where your tool keeps reusable work: a Claude skill (.md) file, or a ChatGPT Custom GPT or saved prompt.

Required Inputs

A completed workflow run to evaluate
The scorecard metrics and their targets
A place to save the refined instructions as a new Skill (.md) file

Claude Code & Meta-Prompting Tricks (37–40)

Choose the Right Surface

Match the task to the right surface — plain chat, a browser agent, a connected-tools workspace, or a local-files coding app — and know each one's name in both Claude and ChatGPT.

Claude vs. ChatGPT

Every surface has a counterpart in both tools — the table pairs each Claude surface with its ChatGPT equivalent (Chat/ChatGPT, Claude for Chrome/Operator, Cowork+MCP/connectors, Claude Code/Code Interpreter).

Required Inputs

The task you want to do
Where the work lives (a conversation, a website, an online tool, or your computer's files)

Claude only

Non-Technical Claude Code Tasks

Use Claude Code without writing any code — build a one-page site, tidy and rename a messy folder, or draft and revise documents, all in plain English.

Claude vs. ChatGPT

This is the Claude Code desktop app. Nearest ChatGPT equivalent: Code Interpreter for data and file tasks, or Operator — a different model, and neither is a drop-in match.

Required Inputs

The Claude Code app, signed in
A practice folder opened in the app (so it works only on those files)
A plain-language description of the goal

Claude only

Meta-Prompting Starter Pack

Add two slash commands and two skills to Claude Code that help it write, save, reuse, and plan prompts — starting with /create-prompt.

Claude vs. ChatGPT

These are Claude Code slash commands and skills in your local .claude folder. No direct ChatGPT equivalent.

Required Inputs

Claude Code installed and opened at least once (so the hidden .claude folder exists)
The free meta-prompting starter pack download (about 166 KB)
A task to turn into a saved prompt, phrased as "how might we…"

Claude only

Browser-Use Safety & Prompt-Injection Awareness

Use Claude for Chrome safely: grant access one site at a time, review every action before you approve it, and recognize prompt injection — hidden instructions on a page trying to redirect Claude.

Claude vs. ChatGPT

Built around Claude for Chrome. ChatGPT's browser agents (Operator / Atlas) carry the same prompt-injection risks, so these safety habits apply there too.

Required Inputs

Claude for Chrome installed on a paid plan
A safe, low-stakes site to practice on
Awareness of which sites to keep off-limits while you learn

Implementation Checklist

For Each Trick, Ensure You Have:

Clear system instructions saved as a file
Required context files identified and accessible
Input templates documented
Output format defined
Sample prompts tested
Edge cases considered

Integration Tips:

Chain tricks for complex workflows
Schedule regular runs (weekly diagnostics)
Build reusable output templates
Rate outputs to improve prompts over time

Keys to Success

Invest in context documents

Quality of your ICP criteria, brand guides, and playbooks determines output quality

Iterate on prompts

First attempts rarely perfect; refine based on results

Standardize outputs

Consistent formats make skills more useful over time

Measure impact

Track time saved and quality improvements

Take the Next Step

Ready to Build Your Own AI Agent Skills?

These 40 AI Tricks are just the beginning. In our free weekly workshop, you'll learn how to:

Create custom skills from scratch (no coding required)
Use the skill-creator to auto-generate new skills
Chain skills together for complex workflows
Deploy skills across your team and projects

Live sessions every week with Q&A and hands-on demos