From Theory to Practice

Homework Review: Context Extraction

Caiti completed last week’s homework and reported it as a “game changer.” She created two ChatGPT Projects — one personal and one for her bookstore business — and used the conversation technique to extract context into reusable files. The immediate benefit: she can now go straight to specific questions without repeating background information every time.

Otakar introduced the OHIO principle (Only Handle It Once): extract your context once, save it somewhere accessible, and reuse it across any AI tool or conversation. Keeping context files on your computer means you’re not locked into one platform.

Custom Instructions vs. Projects

Caiti mentioned she hadn’t set up general Custom Instructions yet because she wasn’t sure what to put there. Otakar clarified the difference:

• Custom Instructions – Communication preferences that apply everywhere: response style, language, verbosity, and general behavior (e.g., “use British English,” “be concise,” “give me a one-sentence answer first, then explain”)
• Projects – Domain-specific context for a particular area of your life or work (e.g., bookstore business details, personal situation)

Caiti also noticed that ChatGPT is overly complimentary — always saying “great idea!” regardless of quality. Otakar confirmed this is its default personality and can be adjusted through Custom Instructions.

Live Prompting Exercise: Bookstore Cost Calculator

Caiti chose a real task: figuring out all costs involved in purchasing books (price, delivery, VAT) and calculating profit margins. She wrote a prompt asking ChatGPT to create a Google Sheets document and to ask her what it needed.

ChatGPT responded with roughly 15 clarifying questions instead of producing a spreadsheet. Caiti’s initial reaction was that she hadn’t done a good job. But this was actually the framework in action:

• Delegation phase – She was still figuring out what she actually needed. The clarifying questions were helping her map out the system, not just the calculation
• Description–Discernment loop – She described her goal, evaluated the response, and realized she needed to narrow the scope before asking for a file

“AI is now helping you to figure out what the system is. Once you figure out how this system works — and I don’t mean software, I mean literally: where your books come from, where you store them, how much you sell them — then you can build the spreadsheet.”

Voice Dictation

With 15 questions to answer, typing would take a long time. Otakar demonstrated two voice features in ChatGPT:

• Microphone button – Dictates your message as text. You can review and edit before sending. Otakar’s preferred method for everyday use
• Voice conversation mode – Full back-and-forth spoken conversation with AI. Useful for brainstorming, but less control over what gets sent

Tip: always review dictated text before submitting, especially for brand names or uncommon terms that speech recognition might misinterpret.

“As much as I can, I talk to the AI tools. I don’t type when I don’t have to. It’s just much more natural to express myself this way.”

Choosing the Right Tool

Caiti asked if ChatGPT can create Google Sheets directly. The answer: not quite. ChatGPT can generate CSV files you can import, but it can’t create native Sheets. For working directly inside Google Sheets, Gemini (Google’s AI) is built in — though Otakar noted its capabilities are still basic.

The broader lesson: don’t expect one tool to do everything. Use ChatGPT for planning and thinking through problems, Gemini where it’s integrated into Google tools, and specialized tools like Canva for specific tasks like image editing.