Prompt Debugging

I'll cut right to it: most people get mediocre results from AI because they don't give it enough context.

They ask a question like it's a Google search. Short. Vague. And then they're disappointed when the answer is generic.

Here's the thing. Every single AI conversation starts from zero. The AI doesn't know who you are. It doesn't know your company, your industry, your role, your goals, or what you had for breakfast. It knows nothing about your situation unless you tell it.

And yet, most people tell it almost nothing.

The copy-paste trick

The single most underrated AI technique is also the simplest one: just paste your stuff in.

Working on a strategy doc? Paste the whole thing into the chat and say "review this."

Analyzing a spreadsheet? Copy the data and paste it in. (ChatGPT and Claude can both handle tables.)

Prepping for a meeting? Paste the agenda, the attendee list, and any relevant background. Then ask for a briefing.

Drafting a response to a tricky email? Paste the email thread and say "help me respond to this. I want to push back on the timeline without burning the relationship."

You're not limited to typing a question from memory. You can give the AI the actual raw material. And when you do, the output quality jumps dramatically.

Think about it: if you hired a consultant and gave them zero context about your business, you'd get garbage advice. Same thing here.

Front-loading context

Beyond pasting documents, there's an art to setting the scene. Before you ask your actual question, spend 2-3 sentences telling the AI what it needs to know.

Bad: "Write me a follow-up email."

Good: "I had a sales call yesterday with the VP of Marketing at a mid-size ecommerce brand. The call went well. They were interested but concerned about our pricing versus a cheaper competitor. Write a follow-up email that acknowledges the pricing concern, reinforces our value, and suggests a 15-minute call next week to discuss a custom package."

That extra context took you 30 seconds to type. It saved you 10 minutes of editing a generic output that missed the mark.

The constraint trick

Here's one most people never think about: tell the AI what NOT to do.

"Write this in plain English. No buzzwords."

"Keep it under 200 words."

"Don't sugarcoat it. Be direct about the risks."

"Don't include anything about X, that's already been covered."

Constraints are context too. They narrow the output space and help the AI give you exactly what you need.

Set it once, benefit forever

Both ChatGPT and Claude now let you set persistent context that applies to every conversation.

In ChatGPT, it's called Custom Instructions. In Claude, you can create Projects with pre-loaded context.

Here's what I'd put in there:

• Your role and industry
• How you like things written (tone, length, format preferences)
• Common tasks you use AI for
• Things you never want (jargon, filler, certain formats)

You set this up once. Takes five minutes. And every conversation from that point forward starts with the AI already knowing who you are and how you work.

It's like the difference between explaining your business to a new person every day versus having an assistant who already knows the drill.

The context stack

If you've been following along, you're starting to see how these techniques combine:

• Role (Day 8): Tell it who to be
• Chain of thought (Day 9): Tell it how to think
• Context (today): Tell it what it needs to know

Stack all three and you're operating at a level that 95% of AI users never reach. Not because they can't. Because nobody showed them.

Tomorrow we'll cover the last piece: judgment. When to use AI and when to trust your own brain instead.

The #1 reason people get bad AI results: they don't give it enough context. Paste your actual stuff in. Set the scene. Be specific.