Three Simple Steps towards GPT Mastery

Here is a little gift. There are three essential skills that will massively improve your engagements and long term partnerships with Generative Pre-Trained Transformers (GPTs) which use Large Language Models.

- Personalise

- Prompting

- Custom GPTs

Personalise your Interactions

How to Set Up Your Personal Profile in ChatGPT (and Get Smarter Responses)

ChatGPT now lets you set a custom profile so it understands your context and preferences before every interaction.

Here’s how to do it:

Step 1: Go to Settings > Personalization

• In ChatGPT, click your name > Settings
• Select “Custom Instructions” (or “Personalization”)
• You’ll see two main fields:

Step 2: Fill in These Fields Thoughtfully

1. “What would you like ChatGPT to know about you to provide better responses?”
Here’s a powerful format:

“I am a [role] working in [industry]. I support [audience type] with [goal/problem area]. I prefer clear, concise responses written at a [reading level], with space for critical analysis. I’m especially interested in [key topics].”

Example (for you, James):

“I’m a consultant and board member in aviation and tech. I help founders and leaders grow resilient businesses. I prefer responses at postgraduate level with critical alternatives and a focus on 80/20 insights. I'm also interested in psychology, leadership, and practical AI applications.”

2. “How would you like ChatGPT to respond?”
Structure this by tone, format, and critical approach:

“Professional tone, strategic thinking, short first response (<1200 characters), then more depth in conversation. Avoid generic advice. Give clear structure, and offer a contrarian view when relevant.”

Step 3: Save and Test It

Once saved, all chats will follow this context automatically.
Try:

“Summarise this article for a strategy presentation to board-level stakeholders.”
Or:
“Give me 3 bold but evidence-based recommendations for scaling a consulting firm.”

You’ll notice sharper, more aligned responses.

Why This Matters

AI is powerful—but only if it knows who you are and what you’re trying to achieve.
Your profile acts like a flight plan—it helps GPT stay on course.

‍

Master Essentials of Effective Prompting

1. Define the Role, Not Just the Task
Effective prompts often start with contextual identity.

“You are a compliance coach helping SMEs prepare for audits…”
This creates boundaries and expectations for tone, relevance, and style.

2. Add Intent, Not Just Content
Explain what you want and why.

“Summarise this report so I can present it to non-technical stakeholders.”
Clarity of purpose increases precision.

3. Provide Format Expectations
State what output you want: bullets, a table, a persuasive argument, etc.

“Respond in a table comparing strengths and weaknesses.”

4. Layer Perspective
The best prompts are empathic:

“Explain this concept as if you're talking to a sceptical CFO.”
This generates responses attuned to audience psychology.

5. Make It Iterative
Break the “one-shot” mindset. Ask follow-ups, refine, and evolve the prompt.

“Now shorten it.”
“Now challenge that perspective.”
“Now give me a counterexample.”

In summary

• Give examples in the prompt (especially tone, style, or context).
• Use constraints (“no more than 100 words”, “use UK English”).
• Ask for reasoning to spot hallucinations: “Explain your logic.”
• Be audience-specific (“Write for a busy founder with no technical background”).
• Iterate interactively rather than expecting perfection in one shot.

‍

Build an Effective Custom GPT (in ChatGPT)

Custom GPTs can really speed up prompting by providing a more specific contextual design to prompt within. Like going to the right section of the library and the right book to look for your answers, rather than just asking aloud in a busy pub and accepting the first answer you get.

Step 1: Define Its Purpose Clearly
Start with the end user and use case. Is it for writing? Decision support? Customer onboarding? Technical FAQs?
Use a job-to-be-done framing:

“Help [who] do [what], so they can [benefit].”

Step 2: Use the GPT Builder Interface
Inside ChatGPT (Plus), click Explore GPTs > Create.
The wizard asks:

• What your GPT is for
• What tone/persona it should have
• What tools or data it can access
• Whether you want to upload files or links as reference material

Step 3: Upload Core Knowledge
Upload PDFs, docs, spreadsheets, or even SOPs that the GPT can cite internally.
This creates a “knowledge file vault” for retrieval during chats.

Step 4: Configure Instructions
Customise the GPT’s behaviour with:

• System Instructions: “Always cite sources,” “Act as a strategy coach,” etc.
• Custom Prompts: Add shortcuts or drop downs for users to ask consistent questions.

Step 5: Test Iteratively
Ask it edge-case or adversarial questions.
Use real-world prompts from your team or clients.
Refine based on how it handles ambiguity or misses context.

‍

Yeah, but what about Security and Protection?

You should definitely check in with your IT department and be wary about what you store and where you store it, who has access to it and all the usual stuff but in general security is built in, but double check the small print. Keep up to date with changes and ask more questions. This stuff matters but it is not an excuse to do nothing or do it badly.

Security and Data Protections in Custom GPTs

1. No Training on Conversations
Custom GPTs do not train on your prompts or data unless you explicitly enable it. By default:

• Uploaded files stay private
• GPTs don’t “learn” from user interactions

2. File Uploads Are Not Public
Your uploaded documents are stored securely within your GPT and not shared across accounts. Think of it as a sandboxed memory.

3. API and Tool Integrations (Optional)
You can add APIs, but these come with custom risk. Always use HTTPS, token authentication, and limit scope.

4. Privacy for End Users
Unless you publish your GPT publicly, only you and invited users can access it. If published, users still can’t see your internal files.

5. Audit Trail and Monitoring
Though OpenAI doesn't give full admin dashboards yet, enterprise customers can use usage logs and data loss prevention (DLP) policies to monitor activity.

Don't Trust Without Verification

GPTs do not verify facts, they retrieve and rephrase. If your file is outdated or incomplete, so is the advice. Build in human oversight for critical tasks and be critical about sources and sense-making.

‍