33 practical tips, 8 copy-ready prompts and a plain-English glossary: the playbook I use every day as an AI expert, Anthropic-certified in Claude Code, MCP and AI Fluency.
Say who it’s for, why it matters and what a great result looks like. Claude can only be as specific as the brief you give it.
I'm writing this for [audience]. The goal is [goal]. A great answer would [criteria]. Here's the task: ...
A single example of the format and tone you want beats a paragraph describing it. Two or three examples (few-shot) work even better.
Wrap pasted content in tags like <document> or <notes> and keep your instructions outside them. It stops Claude confusing your data with your request.
<document> [paste here] </document> Summarize the document above in 5 bullet points for an executive.
With big inputs, paste the material at the top and ask your question at the end. Asking Claude to quote the relevant passages first makes answers more grounded.
For anything ambiguous, ask Claude to ask you questions before it starts. You’ll skip three rounds of rewrites.
Before you start, ask me up to 5 questions that would most improve your answer. Wait for my replies.
For analysis, math, planning or tricky decisions, ask Claude to reason step by step first, or turn on extended thinking for hard problems.
“Write in short, plain sentences” works better than “don’t be wordy.” Positive instructions are easier to follow.
Explicitly allow Claude to say when information is missing or uncertain. It’s one of the simplest ways to cut hallucinations.
Answer only from the material I gave you. If it isn't there, say "Not in the source" instead of guessing.
Point to exactly what to change (“shorter intro, keep the table, more concrete examples”). Specific feedback converges fast.
After a draft, ask for the strongest objections a skeptical expert would raise, then a revised version that answers them.
Critique your draft as a skeptical senior [role]. List the 5 biggest weaknesses, then rewrite it fixing each one.
Put your brand guide, product docs or style rules in a Project once, and every chat inside it starts with that context.
Documents, code, diagrams and small apps live in an artifact you can refine turn by turn instead of copying out of the chat.
Split complex work into steps: research, then outline, then draft, then edit. Feed each result into the next prompt. Quality jumps.
Request a table, JSON with named fields, or a fixed template when the result feeds another tool or spreadsheet.
Return the result as JSON with keys: name, company, need, urgency (low/medium/high). No other text.
Run /init to generate a CLAUDE.md that records your stack, commands and conventions. Claude reads it every session, so you stop repeating yourself.
Use plan mode for anything bigger than a small fix: agree on the approach first, then let Claude implement it.
Give acceptance criteria and ask Claude to write and run tests. A failing test is the clearest instruction you can give.
Use /clear when you switch to unrelated work. A focused context gives sharper, cheaper answers.
MCP servers let Claude work with databases, browsers, design tools and CMSs. This website was redesigned with Claude Code driving the Elementor MCP.
On unfamiliar code, ask for a walkthrough of how it works first. You’ll catch wrong assumptions before they become wrong changes.
Package workflows you repeat (release notes, reviews, deploy checks) as custom commands or skills so they run the same way every time.
Ask for focused changes, review each one, and commit often. Easy to verify, easy to roll back.
Map the process, count how often each step happens, and start with the boring step that happens most. That’s where hours come back.
Let AI draft, sort and summarize freely, but require approval before anything is sent, paid or deleted.
Ask the model for structured output, then check it with plain logic (required fields, allowed values) before it moves downstream.
For company knowledge, retrieving the right documents at question time is usually faster, cheaper and easier to update than training a model.
Collect 10 to 20 real examples with the answers you expect. Re-run them whenever you change a prompt or model so you know quality didn’t slip.
In n8n, Make or Zapier, store inputs, outputs and errors for each run. Debugging goes from guessing to reading.
Route simple, high-volume steps to a smaller, faster model and save the most capable model for hard reasoning. Same quality, lower cost.
Ask it to explain a concept simply, then quiz you, then grade your answers and explain mistakes.
Teach me [topic] like I'm new to it. Then give me 5 quiz questions one at a time, wait for my answer, and grade each one.
Pick a tiny real project (a script, a dashboard, an automation) and ship it. Building beats watching every time.
Write real code and get it checked instantly. That’s why I built LearnWithSQL (live Postgres) and LearnWithPython (Python in the browser).
Spaced repetition locks concepts in. Five minutes of review the next day protects hours of learning.
Templates I use for business analysis, automation design and learning. They work in Claude and most other assistants.
Before you answer, ask me up to 5 questions that would most change your response. After I reply, give your answer and note any assumptions you still had to make.
Here is how a process works today: <process> [describe each step, who does it, which tool] </process> Map it as a numbered list. For each step mark: manual / automatable / needs human judgment. Then propose the 3 automations with the biggest time savings, the tools to use (n8n, Make, Zapier, Airtable), and the risks.
Turn these notes into user stories using "As a [role], I want [goal], so that [benefit]". Add 3-5 testable acceptance criteria (Given/When/Then) for each and flag anything ambiguous as an open question. <notes> [paste notes] </notes>
From the transcript below, extract: decisions made, action items (owner + due date if mentioned), open questions, and risks. Use a table for action items. Don't invent owners or dates. <transcript> [paste] </transcript>
Answer the question using only the documents below. First list the exact quotes you're relying on, then answer. If the documents don't contain the answer, reply "Not in the source." <documents> [paste] </documents> Question: [question]
Act as my SQL tutor. Given this table schema: [schema] Give me one practice question at a time, from easy to hard. When I answer with a query, check it, explain any mistakes, show a better version, then give the next question.
Here is my Python code and the error I get: <code> [paste] </code> <error> [paste] </error> Explain the cause in plain language, show the minimal fix, and tell me how to avoid this class of bug next time.
Review the work below as a demanding senior [role]. Score it 1-10 on clarity, correctness and impact, list the top 5 issues in order of importance, then provide an improved version. <work> [paste] </work>
Start with Anthropic’s free AI Fluency courses on Claude Academy: delegation, description, discernment and diligence.
Use the tips above, then go deeper with Anthropic’s prompt engineering guide.
Practice SQL on LearnWithSQL and Python on LearnWithPython, with Claude as your tutor.
Connect tools with n8n or Make, then build with Claude Code and MCP.
Join 1,000+ professionals in the AI Automation Hub, or let ScalingThis build the systems for you.