Dos and Don'ts
Dos and Don’ts
These are the patterns that separate people who get great results from AI from people who get generic output.
Do: Give context before asking
Bad: “Write an email to a customer.”
Good: “Write an email to a Dubai customer who asked about the Specialized Allez Sprint. He came into the showroom last week and rode a test bike. He has not replied to my follow-up. Tone should be warm but not pushy. Two short paragraphs max.”
Claude works with what you give it. The more relevant context you include upfront, the less iteration you need.
Do: Specify the format you want
Bad: “Summarize this.”
Good: “Summarize this in 5 bullet points. Each point max one sentence. Focus on actionable items.”
Format instructions are free. Use them.
Do: Iterate
Claude responds well to follow-up instructions. After a first draft:
- “Make the first sentence more direct.”
- “Remove the part about pricing — we handle that separately.”
- “Make this 30% shorter.”
- “The tone is too formal. We talk to cyclists, not corporations.”
You do not need to start over. You can shape the output through conversation.
Do: Use the context file
The Adam Bike context file is the most valuable thing in your AI setup. It contains your brand voice, customer profile, product focus, and team information. Every conversation in the Claude Cowork project starts with Claude already knowing your business.
If you are working on something outside of Cowork, paste the context file content at the start of your conversation.
Do: Tell Claude what role it is playing
“You are Adam Bike’s social media manager. You know our brand is premium but approachable. Our customer is a serious Dubai cyclist aged 28-45.”
This sets the frame for everything that follows. Claude adapts its language, tone, and assumptions to match the role you give it.
Don’t: Accept the first draft as final
The first draft is a starting point. The best results come after one or two rounds of feedback. If the output feels generic, it usually means the prompt lacked specificity, or you just need one round of “make this more [X].”
Don’t: Ask for everything at once
Bad: “Write me a content strategy, five Instagram captions, a product description for the Trek Domane, and an email newsletter for this month.”
Good: Start with one thing. Get it right. Move to the next.
Long, multi-part prompts produce long, mediocre outputs. Break complex work into steps.
Don’t: Publish AI-generated facts without checking
Claude can write convincing product descriptions, but it does not know your actual specs, your current stock, or your prices. Any factual claim in customer-facing content should be checked before it goes live.
Write with Claude, verify with your own sources.
Don’t: Treat a failed output as a failed tool
If Claude gives you something bad, it is almost always a prompt issue. Add more context, be more specific about format, or adjust the role. Do not abandon the task.
Don’t: Re-explain your business every time
That is what the context file is for. If you find yourself typing the same background information repeatedly, add it to your context file. Fifteen minutes of setup saves hours over time.
The single most useful habit
Before you send a prompt, read it back and ask: “Does Claude have everything it needs to do this well?”
If the answer is no, add what is missing. That one habit, applied consistently, will double the quality of your AI output.