Day 6: Give AI better context, and it gets much more useful

What you'll see today:

• Why blank-chat AI keeps making you repeat yourself
• How Projects and workspaces reduce drift
• When NotebookLM is better than a normal chatbot
• Why image upload is one of the most underused beginner moves
• Your action: turn messy input into clean structured output

---

A lot of people use AI like this:
open a new chat, paste the same context again, repeat the same instructions again, and hope the model remembers what matters.

That works for quick one-off tasks.

It gets frustrating fast for anything bigger:

• an ongoing work project
• a research topic
• a pile of documents
• a repeated weekly workflow
• a conversation that depends on earlier decisions

The core issue is not always model quality.
A lot of the time, the issue is context management.

Today is about a simple upgrade:
stop treating every task like a fresh blank chat.
Start giving AI a better workspace.

---

1) Projects and workspaces: stop starting from zero every time

If you are doing ongoing work, the best move is usually not “write a better one-shot prompt.”
It is “create a better place for the work to live.”

Most major AI tools now offer some version of:

• Projects
• Workspaces
• Knowledge folders
• Persistent instructions
• Uploaded files tied to a specific task or thread of work

The point is simple:
instead of re-explaining the same project every time, you create one place where the AI can keep using the same core context.

This helps with three beginner problems:

• less repetition
• less drift
• better continuity across sessions

A practical example:
If you are job hunting, a project workspace might include your resume, target role description, a few past cover letters, and a note saying “keep this practical and concise.”

---

2) NotebookLM: use it when the job is “understand these sources”

Normal chatbots are often used for general conversation.
NotebookLM is especially useful when the job is:
“Here are the materials. Help me understand them.”

That makes it useful for:

• class readings
• meeting notes
• research reports
• policy documents
• interview prep packets
• long internal docs nobody wants to read line by line

A simple rule:

• use a regular chatbot when you want broad help
• use a source-grounded notebook when you want answers based on a specific pile of material

That distinction matters because a lot of beginner frustration comes from asking a general model to act like it has already read your documents carefully when it has not.

---

3) Vision and structured output: turn messy input into usable work

One of the most underused beginner moves is simply uploading an image.

Modern tools can often help with:

• screenshots
• whiteboards
• handwritten notes
• charts
• menus
• forms
• receipts
• interface errors

Useful beginner examples:

• upload a screenshot of an error message and ask what it means
• upload a photo of a messy whiteboard and ask for action items
• upload a chart and ask for the biggest takeaway
• upload a long form and ask what needs attention first

Then make one more upgrade:
ask for structure.

Instead of:
“Summarize this.”

Try:
“Extract the key dates, people, and action items into a table.”

Or:
“Turn this into a checklist.”

Or:
“Pull the vendor name, invoice date, and amount into CSV format.”

That tiny change often makes the output far more useful.

---

Your action for today

Use one real piece of messy input.
Choose one of these:

• a long email thread
• a PDF
• a screenshot
• a whiteboard photo
• a meeting note dump
• a page of notes

Then ask AI to turn it into a structured output.

You can use a prompt like:

“Read this material and turn it into a clean Markdown table with these columns: item, status, owner, deadline, and notes. If something is missing, leave it blank instead of guessing.”

Reply with:

• what type of input you used
• the exact prompt you used
• the structured output you got back

I’ll tell you:

• whether the output format was the right choice
• what the prompt did well
• the one change that would make it more reliable next time