Prompting and Using AI Chatbots Effectively - resource file

A chatbot is a software application or web interface designed to converse through text or speech.

Generative AI is so called because it can generate different types of content. ChatGPT, for instance, was originally designed to generate text. Its name breaks down as follows: Chat (conversational interface), G for Generative (it generates text), P for Pre-trained (trained in advance on large datasets), and T for Transformer — the algorithm that enables it to understand the context and structure of human language and to predict the next word in a sentence.

Today, these tools are described as multimodal because they can generate not only text but also images, video, audio, and more.

Getting the best results from a chatbot is an art known as prompt engineering. A well-crafted prompt significantly improves the quality and accuracy of the AI's output.

Two key principles to keep in mind:

• Most popular chatbots maintain the context of a conversation. You can return to a chat at any time and pick up where you left off.
• Don't settle for the first response. Ask for clarification or revisions, request a different angle, or provide additional details until you get the result you're looking for.

Prompt Engineering Best Practices

1. Structure your request. Organize your prompt clearly. Use bullet points or numbered steps to guide the chatbot's response.
2. Assign a role. This steers the AI toward a specific tone or expertise level. For example:

• "You are a cybersecurity expert who regularly trains beginners. Write a summary note explaining how to conduct cybersecurity monitoring."

3. Be specific. The more precise and detailed the prompt, the better the output. Avoid vague or open-ended instructions.
4. Provide context. If your request depends on specific background information, include it explicitly.
5. Give examples. If you want a specific format or style, show the chatbot a concrete example of what you expect.
6. Use text rewriting features. You can ask the chatbot to:

• Summarize or expand a text or paragraph
• Correct spelling and grammar
• Translate text — or summarize in English an article written in another language (you can even provide the web link to the article)
• Convert formats: from email to tweet, from written text to spoken script, from email to formal letter
• Reformat content: as a bulleted list, as a table
• Add a theme or structure
• Adjust tone: more formal, warmer, more humorous, or more approachable
• Adapt writing style: paste a sample of your own writing, and the chatbot can match it — it can even write in prose!
• Change the output format: by default it uses Markdown, but you can request JSON, XML, HTML, and more

Don't hesitate to ask for a new version of the answer multiple times until it meets your expectations.

 

Ideas for Using Chatbots

Here are some practical and creative use cases:

• Brainstorming: Ask the chatbot to explore ideas, generate approaches for a project, or propose team activities.
• Anticipate objections: Ask it to identify likely objections to a project you plan to present, then build counterarguments.
• Onboarding: Ask it to draft an onboarding process for new employees.

 

Beyond Text: What Chatbots Can Do

Chatbots are far more than text generators. Here is what they can handle:

Images

• Read and describe images
• Identify plant species from a photo and provide care instructions
• Solve geometry problems from an image
• Translate road signs in foreign languages
• Generate HTML code from a picture of a website

Files

• Summarize PDFs
• Analyze spreadsheets and generate trend charts
• Create annotated geographic maps
• Compress images

Calculations

• Perform complex calculations
• Explain their reasoning step by step when asked — this actually improves their accuracy on complex problems (see Chain-of-Thought below)

Content Generation

• Generate images from text descriptions
• Convert images into videos
• Transcribe or summarize videos

 

Good to know: Even though chatbots have a knowledge cutoff date, they can perform a web search if you explicitly ask for recent data or current events. Gemini excels at this, thanks to its native integration with Google Search.

 

Chain-of-Thought Prompting

Research has shown that asking a model to explain its reasoning step by step — rather than jumping straight to the answer — significantly improves its ability to solve complex problems.

Instead of: "What is the answer to X?"

Try: "Explain step by step how you would approach X."

This technique is especially effective for logical reasoning, math, and multi-step analysis.

 

Using Chatbots for Coding

Chatbots can assist throughout the entire development workflow:

• Write code in the language of your choice (Claude and Gemini are generally strongest here)
• Test code
• Refactor code (improve structure and readability without changing behavior)
• Debug and correct errors
• Explain error messages in plain language

 

Image Generation

Several tools exist for generating images from text prompts:

• DALL·E 3 — integrated directly into ChatGPT. Use the "reload" button to generate variations; click on an image and then the "i" button to view the auto-enhanced prompt.
• Midjourney — high-quality artistic outputs
• Adobe Firefly — integrated with Adobe tools
• Stable Diffusion — open source, more complex to master
• Bing Image Creator — free, also based on DALL·E
• DreamStudio — online interface for Stability AI

 

Custom Agents

Each major platform allows you to create pre-configured AI agents:

• ChatGPT → GPTs: A GPT is a pre-configured version of GPT-4 with custom instructions prepended to every conversation. Example: a "Logo Generator" GPT that automatically asks for shape and color preferences before generating a logo. Access via Explore GPTs, or create your own.
• Gemini → Gems
• Claude → Projects

 

Custom Instructions

Custom instructions let you personalize the chatbot's behavior by providing context about yourself and your preferences — so you don't have to repeat it every time.

 

What does it need to know about me to give you better answers?

• Which country or town you live in
• Your professional interests and role
• What you do in your free time
• Your professional goals
• Whether you work in a cross-cultural context

 

How you want it to respond

• Preferred tone (professional, casual, etc.)
• Level of detail (concise vs. thorough)
• Format preferences (prose, bullet points, tables)
• Domains where you'd like references or resources

 

In ChatGPT, go to Customize ChatGPT to set these. Note: If a custom GPT is used, the "custom instructions" are read first.

Custom instructions are also available in Gemini (under Settings and Help > Instructions for Gemini). Note: not available in SIL and Wycliffe instances.

 

Comparison of Major Chatbots

ChatGPT — The Swiss Army Knife

• Strengths: Leading logical reasoning (via Thinking models), seamless multimodality (voice, image, native video), largest ecosystem of custom GPTs.
• Weaknesses: Can slow down with very long instructions, tends to be verbose, and sometimes overly agreeable. Less natively integrated into professional ecosystems than Copilot or Google. Smaller context window than Gemini or Claude.
• Best for: General-purpose use

 

Gemini — The Data Powerhouse

• Strengths: Massive context window  — analyze hundreds of files or hours of video at once. Unbeatable for long document analysis. Seamless Google Workspace integration (Drive, Docs, Gmail). Excellent multimodal capabilities. Up-to-date information via Google Search integration.
• Weaknesses: Responses can be verbose. Weaker than ChatGPT in pure reasoning and creative tasks.
• Best for: Productivity in a Google environment

 

Claude — The Analyst's Choice

• Strengths: Most natural and nuanced writing style. Deep analytical reasoning. Very broad context window. Precise, instruction-following behavior. Excellent at coding (Claude Code). Very low hallucination rate.
• Weaknesses: No native image generation. Web search is less capable than Gemini. Occasionally over-cautious (may refuse edge-case queries).
• Best for: Complex analysis, writing, and technical tasks

 

Copilot — The Microsoft Productivity Assistant

• Strengths: Deep integration with Office 365. Can summarize Teams meetings, create PowerPoint from a Word doc, or analyze Excel files. Enterprise-grade data security.
• Weaknesses: Interface can feel cluttered. Less creative for open-ended brainstorming — constrained by business workflows.
• Best for: Productivity in a Microsoft environment

 

Use Case Recommendations

 

By task

• IT / technical tasks: ChatGPT, then Gemini
• Coding: Gemini or Claude
• Long document analysis: Gemini, Claude, ChatGPT
• Complex reasoning and writing: Claude, ChatGPT, Gemini
• Enterprise productivity: Gemini (Google) or Copilot (Microsoft)
• Creative tasks: ChatGPT, Claude, Gemini
• Web search / current events: Gemini
• Multimodal (image, voice, video): ChatGPT, Gemini
• Reliability and safety: Claude, ChatGPT, Gemini

 

Practical recommendation: use 2–3 complementary tools

• Claude — for in-depth analysis and writing
• ChatGPT — for image generation and creative tasks
• Gemini — for large document databases and real-time web search

 

Other AI Tools to Know

 

Conversational AI

• Copilot — integrated with Bing and Microsoft 365
• Llama — open source, by Meta
• Mistral AI — French, independent from GAFAM
• Perplexity — search-focused AI

 

Audio Generation

• ElevenLabs — realistic AI voice generation
• OpenAI Jukebox — AI-generated music
• Suno at https://suno.com/

 

Video Generation

• Runway — AI video editing and generation
• Synthesia — AI avatars for training videos
• D-ID — talking portrait generation
• Google AI Studio — Google's video and multimodal AI platform

 

Slide Generation

• Gamma — AI-powered presentation creation
• Tome — narrative-driven AI slides
• Beautiful.ai — smart slide design

 

Website Generation

• Mixo — generate a website from a single prompt