Capabilities and Conversation Design

A GPT isn’t just instructions and knowledge — it’s also what tools it can use and how it guides conversations. Enabling the right capabilities and designing smart conversation flows are what separate good GPTs from great ones.

🔄 Quick Recall: In the previous lesson, you learned to upload knowledge files and instruct your GPT to reference them. Now you’ll add capabilities (tools the GPT can use) and design how conversations flow from start to finish.

The Three Built-in Capabilities

Every GPT can toggle three built-in tools:

Web Browsing

What it does: Lets the GPT search the internet and reference current information.

Enable when: Your GPT needs current data — news, prices, recent events, live documentation.

Disable when: Your GPT should only answer from its knowledge files. Leaving web browsing on for a policy assistant means it might mix internet results with your company policies — creating confusion.

DALL-E Image Generation

What it does: Lets the GPT create images from text descriptions.

Enable when: Your GPT involves visual content — social media graphics, concept art, design mockups, or educational illustrations.

Disable when: Your GPT is purely text-based. Unnecessary capabilities add complexity and can confuse the GPT about when to generate images vs. text.

Code Interpreter & Data Analysis

What it does: Lets the GPT write and execute Python code, analyze uploaded files (CSV, Excel, PDF), and create charts.

Enable when: Your GPT works with data, calculations, file processing, or needs to generate charts and graphs.

Disable when: Your GPT is conversational only and doesn’t need to process files or run calculations.

✅ Quick Check: Why might you disable web browsing for a GPT that answers questions from a company handbook? (Answer: You want answers strictly from the uploaded handbook, not from internet results that might contradict or dilute your company’s specific policies.)

Capability Combinations by Use Case

Rule of thumb: Enable only what your GPT needs. Fewer capabilities mean more focused behavior.

Designing Conversation Flows

The best GPTs don’t just answer questions — they guide users through a structured process. This is called conversation design.

The Intake Pattern

Before generating output, have your GPT collect critical information:

CONVERSATION FLOW:
When the user first interacts, gather these inputs before writing:
1. Ask: "Who is the target audience for this content?"
2. Ask: "What tone do you prefer? (Professional / Casual / Fun)"
3. Ask: "How long should the output be? (Short / Medium / Detailed)"

Once you have all three answers, generate the content.
If the user provides all details upfront, skip questions
they've already answered.

The last line is important — don’t force users through a questionnaire if they’ve already given you everything.

The Progressive Detail Pattern

Start with a quick output, then offer to refine:

FLOW:
1. Generate a first draft based on the user's initial request
2. After presenting the draft, ask: "Would you like me to:
   (a) adjust the tone, (b) make it longer/shorter,
   (c) add specific details, or (d) this looks good?"
3. Iterate until the user is satisfied

This gives users fast results while allowing refinement — the best of both worlds.

✅ Quick Check: What should you add to an intake pattern to prevent forcing users through questions they’ve already answered? (Answer: An instruction like “If the user provides all details upfront, skip questions they’ve already answered” — so the GPT adapts to users who give detailed initial prompts.)

The Guardrail Pattern

Keep conversations on topic with redirect instructions:

SCOPE:
This GPT ONLY helps with resume writing and job applications.

If the user asks about something unrelated:
→ Respond: "I'm specialized in resume and job application help.
   I'd be happy to assist with that! What job are you targeting?"

Do NOT answer general knowledge questions, give medical/legal
advice, or engage in casual conversation.

Conversation Starters as Flow Triggers

Your conversation starters should map to your GPT’s main flows:

Each starter launches a distinct workflow, making the GPT feel like a polished product rather than a chatbot.

Practice Exercise

1. Open your GPT from previous lessons in the Configure tab
2. Review which capabilities are enabled — disable any you don’t need
3. Add a conversation flow to your instructions (choose Intake or Progressive Detail)
4. Update your conversation starters to match different flows
5. Test each flow in the Preview panel — verify the GPT follows the designed path

Key Takeaways

• Enable only the capabilities your GPT actually needs — fewer tools mean more focused behavior
• Design conversation flows that gather information before generating output
• Use the Intake Pattern for multi-input tasks and Progressive Detail for iterative refinement
• Add guardrails to keep conversations on topic with polite redirects
• Conversation starters should trigger different flows for different use cases

Up Next

In the next lesson, you’ll learn the most technical GPT feature: Actions — connecting your GPT to external APIs so it can fetch live data, send emails, or interact with other services.