Module 1: Master the AI Chat
Prompt smarter. Verify better. Unlock what AI can really do for you.
Learn how generative AI works, compare ChatGPT, Gemini, and Claude, write effective prompts using the RCTF framework, and build a fact-checking habit to verify everything AI tells you.
Learning Objectives - Understand what generative AI is and what it can and cannot do at work
- Compare ChatGPT, Gemini, and Claude and choose the right tool for each task
- Write structured prompts using the RCTF framework (Role, Context, Task, Format)
- Refine AI outputs iteratively to get professional-quality results
- Explore advanced AI chat features that most users overlook
- Implement a fact-checking process to verify AI-generated content before use
What You'll Learn - What generative AI is and why it matters for professionals in 2026
- ChatGPT vs Gemini vs Claude - strengths, differences, and when to use each
- The prompt engineering framework: Role, Context, Task, Format (RCTF)
- Iterative prompting: how to refine a vague prompt into a precise result
- Advanced AI chat features: memory, canvas, file uploads, voice mode
- AI hallucinations: why they happen and how to spot them
- Fact-checking workflow: verifying AI outputs before you use or share them
What Generative AI Is and Why It Matters
Generative AI has become one of the most practical tools available to working professionals. You may have already used it without realising - autocomplete in your email, suggested replies in messaging apps, or translation tools. But tools like ChatGPT, Gemini, and Claude go much further. They can draft emails, summarise reports, brainstorm marketing ideas, generate images, explain complex topics in plain language, and help you prepare for meetings - all from a simple conversational prompt.
Generative AI works by predicting the most useful response to your input. These models have been trained on enormous amounts of text from books, articles, websites, and other sources, so they have absorbed a vast range of knowledge and writing styles. You type something in, and the AI generates something new based on patterns it has learned. This is fundamentally different from a search engine, which finds existing web pages. Generative AI creates new content based on your request.
What AI can do well: Draft documents and emails, summarise long reports, brainstorm ideas, translate languages, explain unfamiliar topics, create images, and analyse data. Tasks that typically take 15 to 60 minutes can often be completed in seconds with a good prompt.
What AI cannot do: Access your private files unless you share them, guarantee factual accuracy, replace human judgment, or make decisions for you. AI is a capable assistant - not a decision-maker. It can also produce confident-sounding answers that are completely wrong, which is why verification matters (we cover that later in this module).
The key mindset shift is to treat AI like a smart junior colleague. You give it clear instructions, review its output, correct mistakes, and build on what it produces. The better your instructions, the better the output. That is what the rest of this module is about.
Key Insight: Generative AI does not search for existing content - it creates new content based on what you ask. Treat it like a smart junior colleague: give clear instructions, review the output, and refine.
Real-World Example: A project coordinator spends 45 minutes writing a weekly progress report every Friday. With ChatGPT, she types her bullet-point notes and asks it to write a professional summary. The first draft takes 10 seconds. She spends 5 minutes reviewing and adjusting. Time saved: 40 minutes every week - over 34 hours a year.
Q: What is the most accurate way to describe how generative AI like ChatGPT works?
Generative AI creates new content by predicting the most useful response based on patterns learned from training data. Unlike a search engine that finds existing pages, AI generates something new each time. It does not copy text directly or access private files without your permission.
Think about a task at work that takes you more than 30 minutes each week. What would you need to give AI as input to help you complete it faster?
ChatGPT vs Gemini vs Claude - Your Three AI Tools
Three AI tools dominate the professional landscape in 2026:
ChatGPT (by OpenAI),
Gemini (by Google), and
Claude (by Anthropic). All three can write, summarise, explain, analyse, and generate content. All three offer free tiers powerful enough for most professional tasks.
ChatGPT is available at
chatgpt.com. It is one of the strongest all-round AI assistants for writing, analysis, and creative work. The free tier includes the latest GPT model, web search, image uploads, and image generation. ChatGPT Plus (USD 20/month) removes usage caps and lets you create Custom GPTs and run deeper data analysis. Its
Canvas feature lets you co-edit documents side by side with the AI.
Gemini is available at
gemini.google.com. Built by Google, every response can draw on current web information by default. The free tier includes image generation via
Nano Banana 2, custom
Gems (specialised AI assistants you build),
Gemini Live voice mode,
Canvas, and limited
Deep Research. Google AI Pro (USD 20/month) unlocks the most powerful models and full Google Workspace integration.
Claude is available at
claude.ai. Made by Anthropic, Claude is known for careful, nuanced responses and strong handling of long documents. The free tier includes
Projects for organising instructions and files,
Artifacts for creating interactive documents and visual tools,
web search, and
voice mode on mobile. Claude Pro (USD 20/month) unlocks higher usage limits. Claude excels at detailed analysis and writing with specific style requirements.
ChatGPT vs Gemini vs Claude - Key Strengths (2026)
Which one should you use? For this course, any of the three works for all exercises. Many professionals use two or all three - ChatGPT for drafting, Gemini for live research and Google Workspace tasks, and Claude for detailed analysis. Start with whichever you find most comfortable.
Watch video: ChatGPT vs Gemini vs Claude - Your Three AI Tools
Key Insight: All three AI tools - ChatGPT, Gemini, and Claude - offer capable free tiers. ChatGPT excels at writing and creativity, Gemini at real-time research, and Claude at careful analysis and long documents.
Real-World Example: A marketing manager uses Gemini to research competitor pricing (it searches the web in real time and shows sources). She pastes the findings into Claude and asks it to write a detailed comparison report. She then uses ChatGPT to turn the report into a punchy slide outline. Three tools, each doing what it does best, in under 15 minutes.
Q: Which AI tool is specifically designed around real-time web search as a core feature?
Gemini was built by Google with real-time web search as a core feature - every response can draw on current information by default. ChatGPT and Claude also offer web search capabilities, but Gemini's integration with Google's search infrastructure makes it particularly strong for research that needs up-to-date information.
Do you currently use any AI chat tools? Based on what you have just read, which of the three - ChatGPT, Gemini, or Claude - would you try first for your most common work task, and why?
The RCTF Prompt Framework
The single biggest factor in getting useful AI output is the quality of your
prompt - the instruction you type into the AI. Most people start by typing short, vague requests like "write an email" or "explain this topic." The AI does its best, but the result is usually too generic to be useful. The fix is simple: give it more context using a framework.
A reliable framework for writing better prompts is
RCTF:
Role, Context, Task, Format.
Role - Tell the AI who it should act as. Example: "You are a professional business writer" or "You are an experienced HR manager." This sets the tone, expertise level, and perspective of the response.
Context - Provide relevant background information. Who is the audience? What is the situation? What should the AI include or avoid? The more specific your context, the more useful the output.
Task - State clearly what you want it to do. Be specific. "Write a follow-up email" is better than "write an email." "Summarise this report in 5 bullet points for a non-technical manager" is better than "summarise this."
Format - Specify how you want the output structured. Should it be a numbered list, a formal email, a table, a paragraph, or a step-by-step guide? If you do not specify, the AI will choose a format that may not match what you need.
RCTF Prompt Framework - Four Elements of an Effective Prompt
Not sure what to ask AI? Use the
AICoach.my Prompt Library - over 255 ready-made prompts organised by profession. Pick your role, copy a prompt, and paste it straight into ChatGPT, Gemini, or Claude.
Watch video: The RCTF Prompt Framework
Key Insight: RCTF stands for Role, Context, Task, Format. Use this framework every time you write a prompt and you will consistently get better, more usable results from any AI tool.
Real-World Example: RCTF in action - a team leader needs to announce a new policy:
Role: "You are a senior HR manager."
Context: "Staff can now work from home up to 3 days a week. The policy starts on 1 July. Keep the tone warm and positive."
Task: "Write an internal announcement about the new flexible working policy."
Format: "Format it as a short email with bullet points for the key rules."
Result: A ready-to-send email that matches the company tone and covers all the key points - in under 10 seconds.
Q: In the RCTF prompt framework, what does the "C" stand for?
In RCTF, the C stands for Context - the background details about the task, the audience, the purpose, and any constraints. Providing good context is what separates a generic AI response from one that feels tailored and immediately useful for your specific situation.
Take a work task you did recently - writing an email, preparing a summary, or making a proposal. Rewrite your request using the RCTF framework. What would your Role, Context, Task, and Format be?
Iterative Prompting - From Vague to Precise
Getting a useful result from AI is rarely a one-shot process. The most effective approach is to treat your first prompt as a starting point, not a finished instruction. Review what the AI gives you, identify what is missing or not quite right, and then give it a follow-up instruction to refine the output. This approach is called iterative prompting, and it is how experienced AI users consistently get high-quality results.
There are several practical techniques for refining prompts:
Be more specific. If the result is too general, add details. Instead of "make it shorter," say "reduce this to 3 sentences, keeping the most important point about the deadline."
Change the tone or style. If the output sounds too formal or too casual, ask: "Rewrite this in a friendlier, conversational tone" or "Make this more concise and professional."
Ask for alternatives. If you do not like the first version, say: "Give me 3 alternative versions of this headline" or "Rewrite this with a different opening paragraph." Having options is often more useful than a single output.
Build on what works. If part of the response is good, tell the AI which part to keep and what to change. "Keep the first paragraph but rewrite the second to focus on the deadline, not the budget."
Ask it to explain its reasoning. If you are unsure why the AI made certain choices, ask: "Why did you structure it this way?" Understanding its reasoning helps you guide it more effectively on the next attempt.
The key insight is that AI is patient. You can iterate as many times as you need. Each follow-up instruction costs you only a few seconds. Compare this to revising a document yourself, where each round of edits might take 10 to 20 minutes. Use that asymmetry to your advantage - keep refining until the output genuinely meets your needs.
A common mistake is starting a brand-new conversation every time the output is not right. Instead, stay in the same conversation thread. The AI remembers everything you have discussed and builds on it. Each follow-up prompt benefits from all the context you have already provided.
Key Insight: Your first prompt is a starting point, not a finished instruction. Iterating with specific follow-up prompts - in the same conversation thread - is how you move from an average AI response to an excellent one.
Real-World Example: A manager asks ChatGPT: "Write talking points for a team meeting."
First output: Generic bullet points about team performance.
Refinement 1: "Make these specific to a software team. We are behind on the Q3 release by 2 weeks."
Refinement 2: "Add a question at the end of each point to encourage discussion."
Refinement 3: "Shorten each point to one sentence."
The final output is specific, actionable, and ready to use - achieved in under 2 minutes through three short iterations.
Q: What is the most effective approach when your first AI prompt produces a result that is not quite right?
The most effective approach is to give a specific follow-up instruction in the same conversation thread. The AI remembers everything you have discussed and builds on it. Telling it exactly what to change - such as "make it shorter and focus on the deadline" - is far more effective than starting over or repeating the same prompt.
Think of a time when a first draft - from yourself or anyone else - was not quite right. What specific instruction would you give to improve it? Practice writing that as a follow-up prompt.
Advanced AI Features You Should Know About
Most people use AI for simple question-and-answer conversations. But ChatGPT, Gemini, and Claude each have powerful features that go far beyond basic chat. Knowing about them can transform how you work.
File uploads and analysis. All three tools accept file uploads - PDFs, Word documents, spreadsheets, images, and more. Upload a 50-page report and ask: "Summarise the key findings in 5 bullet points." Upload a spreadsheet and ask: "What are the top 3 trends in this data?" This is one of the most underused features. Instead of reading a long document yourself, let AI extract what matters.
Memory and personalisation. ChatGPT can remember details about you across conversations if you enable memory in settings. Tell it your role, your writing style preferences, or your industry, and it will apply this context automatically in future conversations. Gemini integrates with your Google account, so it can reference your Gmail, Calendar, and Drive when you ask questions. Claude remembers context within a conversation, and Projects (free on all plans) let you save persistent instructions and reference files across sessions.
Canvas and Artifacts. ChatGPT's Canvas, Gemini's Canvas, and Claude's Artifacts all let you work on documents, code, or visual content side by side with the AI. Instead of receiving a wall of text in the chat, you get a live workspace where you can edit directly while the AI helps. Canvas (on ChatGPT and Gemini) is ideal for co-writing documents, while Artifacts can create interactive tools like calculators, timers, and visual diagrams.
Voice mode. All three tools support voice conversations. ChatGPT and Gemini have advanced voice modes on mobile and web. Claude offers voice mode on its mobile apps. You can talk to the AI naturally and it responds with a human-like voice - useful for brainstorming, practising presentations, or getting quick answers hands-free.
Image generation. ChatGPT generates images from text descriptions (free and paid plans). Gemini generates images using Nano Banana 2. Both create professional visuals for presentations, social media, and marketing materials. Claude does not generate images but can analyse uploaded images and create SVG graphics through Artifacts.
Web search. All three tools can search the web for current information on their free plans. Gemini does this by default. ChatGPT and Claude offer web search as a feature you invoke when needed.
Key Insight: File uploads, memory, Canvas/Artifacts, voice mode, web search, and image generation are features most users overlook. These are not gimmicks - they save hours of work every week when you learn to use them.
Q: Which feature allows you to upload a 50-page PDF and ask the AI to summarise it?
File upload and analysis is available on ChatGPT, Gemini, and Claude. You can upload PDFs, Word documents, spreadsheets, and images, then ask the AI to summarise, analyse, or extract specific information. It is one of the most underused AI features - instead of reading a long document yourself, let AI extract what matters.
Action step: Pick one advanced feature you have never tried - file upload, memory, or Canvas/Artifacts. Try it this week with a real work document and see how it changes your workflow.
AI Hallucinations and Your Fact-Checking Workflow
AI tools can produce confident, well-structured answers that are completely wrong. This phenomenon is called an
AI hallucination - when the model generates plausible-sounding content that has no basis in fact. Hallucinations happen because AI predicts the most likely next word based on patterns, not because it understands truth or verifies facts. It is not lying deliberately - it simply does not know the difference between true and false.
Common types include
invented statistics (citing a percentage or study that does not exist),
fabricated sources (inventing an author or journal),
outdated information (presenting old facts as current), and
plausible but wrong explanations (sound logic, incorrect conclusion).
The risk is highest for
factual claims, legal or medical information, financial data, or anything with real-world consequences. A casual internal email is low-risk. Quoting a regulation in a client proposal is high-risk.
A practical fact-checking workflow for professionals:
Step 1 - Flag claims. As you read AI output, mentally flag any specific facts: numbers, dates, names, statistics, or references to studies or regulations.
Step 2 - Check primary sources. For each flagged claim, search for the original source. Does the study actually exist? Is the statistic from a real report? Does the regulation say what the AI claims? Use Google, official government websites, or industry databases.
Step 3 - Cross-check with the AI. Ask the AI itself: "Can you provide the source for that statistic?" or "Are you confident about that date?" If the AI cannot provide a verifiable source, treat the claim as unverified.
Step 4 - Use or discard. If a claim checks out, use it with confidence. If it does not, rewrite that section using verified information.
Four-Step Fact-Checking Workflow for AI-Generated Content
The golden rule:
the higher the stakes, the more you verify. For a casual internal brainstorm, a quick scan is fine. For a client proposal or public document, check every factual claim against a primary source.
Watch video: AI Hallucinations and Your Fact-Checking Workflow
Key Insight: AI hallucinations are not bugs - they are a fundamental feature of how language models work. The solution is not to stop using AI, but to build a consistent fact-checking habit: flag claims, check sources, cross-check, then use or discard.
Real-World Example: A consultant asks ChatGPT to draft a market analysis report. The AI writes: "According to a 2025 McKinsey study, 78% of Malaysian SMEs have adopted at least one AI tool." The consultant searches for this study on McKinsey's website - it does not exist. The AI invented it. The consultant asks ChatGPT: "Can you provide the exact source for that statistic?" ChatGPT admits it cannot verify the source. The consultant replaces the claim with a real statistic from the SME Corp annual report.
Q: Why do AI hallucinations occur?
AI hallucinations occur because language models predict the most likely next words based on patterns learned during training. They do not understand truth or verify facts - they generate what sounds plausible. This is why confident-sounding answers can be completely wrong, and why a fact-checking workflow is essential for any high-stakes content.
Have you ever shared information from AI without verifying it first? What could go wrong in your specific work context if a key fact turned out to be wrong? How would you apply the four-step workflow to catch it?
Module 2: From Research to Presentation
Research deeply. Listen to your notes. Present brilliantly - all with AI.
Use ChatGPT Deep Research to investigate complex topics, organise sources with Google NotebookLM, generate audio briefings, and create professional presentations with Gamma.
Learning Objectives - Use ChatGPT Deep Research to investigate complex topics quickly and thoroughly
- Organise and query source documents using Google NotebookLM
- Generate Audio Overviews from research sources using NotebookLM
- Turn research findings into a professional presentation using Gamma
- Review and refine AI-generated presentations before sharing
What You'll Learn - The research problem: information overload and how AI solves it
- ChatGPT Deep Research: when to use it and how to interpret results
- Google NotebookLM: uploading sources and asking questions
- NotebookLM Audio Overviews: turning dense documents into podcast-style briefings
- From research to outline: using AI to structure key findings
- Gamma: creating professional presentations from a prompt or outline
- Reviewing AI-generated presentations: what to check before you share
The Research Problem AI Solves
Every professional faces the same research challenge: there is too much information and too little time. Preparing a market analysis, writing a proposal, or getting up to speed on a new topic can take hours of reading, filtering, and organising. You open dozens of browser tabs, skim through reports, copy key quotes into a document, and try to piece together a coherent picture. By the time you start writing, you have spent most of your energy just finding the information.
AI does not eliminate the need for research. But it dramatically changes how you do it. Instead of manually searching, reading, and extracting information from each source one at a time, you can use AI tools to search broadly, synthesise findings, organise sources, and even generate audio summaries - all in a fraction of the time.
This module covers a three-tool research workflow that takes you from a question to a polished presentation:
Step 1 - Deep Research with ChatGPT. Ask ChatGPT to investigate a complex topic. It reads multiple sources, synthesises findings, and delivers a structured report with citations - in minutes rather than hours.
Step 2 - Organise and query with NotebookLM. Upload your research sources (PDFs, articles, documents) into Google NotebookLM. Then ask questions about your own sources - "What are the three key arguments across all these papers?" - and get answers grounded in your specific documents.
Step 3 - Present with Gamma. Take your research findings and turn them into a professional presentation using Gamma, an AI-powered presentation tool. Give it your outline or key points, and it generates slides with layout, visuals, and structure.
This workflow replaces hours of manual work with a structured pipeline that any professional can learn. You do not need special research skills or expensive software. Each tool has a free tier that is more than enough for professional use.
Key Insight: The AI research workflow has three steps: (1) Deep Research with ChatGPT to investigate and synthesise, (2) NotebookLM to organise and query your sources, (3) Gamma to create a professional presentation from your findings.
Real-World Example: A consultant is asked to prepare a 20-minute presentation on green energy trends in Southeast Asia. Previously, this would take a full day of research and half a day building slides. With the AI workflow: ChatGPT Deep Research delivers a structured report in 15 minutes. She uploads the report plus 3 industry PDFs to NotebookLM and asks targeted questions. She pastes the key findings into Gamma and gets a 15-slide deck in 5 minutes. Total time: under 2 hours, including review and refinement.
Q: What is the main problem that the AI research workflow described in this module solves?
The main problem solved is the time and effort spent on information overload. Professionals spend hours searching, reading, filtering, and organising research. AI tools dramatically reduce this time by searching broadly, synthesising findings, and organising sources - while human judgment remains essential for evaluation and decision-making.
Think about a recent work task that required research. How long did you spend finding and organising information versus actually using it? Where in that process could AI have saved you the most time?
ChatGPT Deep Research
ChatGPT's Deep Research feature goes far beyond a normal AI conversation. When you activate it, ChatGPT spends several minutes actively browsing the internet, reading multiple sources, and synthesising what it finds into a structured research report - complete with citations and source links. Think of it as hiring a research assistant who reads 20 to 50 articles on your behalf and delivers a summary.
How it works: In ChatGPT, select the Deep Research mode (available on all plans, including the free tier with a limited number of monthly queries). Type your research question - the more specific, the better. ChatGPT then browses the web, reads relevant pages, and compiles its findings into a detailed report. The process typically takes 2 to 10 minutes, depending on the complexity of the topic. The output includes a structured summary with sections, key findings, and links to the sources it used.
When to use Deep Research: Use it when you need to understand a topic broadly - market trends, competitor analysis, regulatory changes, technology comparisons, or any question that requires synthesising information from multiple sources. It is particularly useful when you do not know where to start or when the topic spans many different sources.
When NOT to use it: Do not use Deep Research for simple factual questions ("What is the capital of France?"), tasks that need your specific internal data (use file uploads instead), or creative writing. These are better handled by standard ChatGPT prompts.
Getting the best results: Write your Deep Research prompt like a research brief. Include the specific topic, the angle you care about, who the research is for, and any constraints. For example: "Research current trends in AI adoption by small businesses in Southeast Asia. Focus on tools being used, barriers to adoption, and government support programmes. The research is for a client presentation to Malaysian SME owners."
Important limitation: Deep Research reports still need fact-checking, just like any AI output. The sources it cites are real (you can click the links), but its interpretation and synthesis may contain errors. Always review the cited sources for critical claims.
Watch video: ChatGPT Deep Research
Key Insight: Deep Research is not a simple Google search - it actively browses, reads, and synthesises multiple web sources into a structured report with citations. Use it when you need to understand a topic broadly rather than answer a single question.
Real-World Example: A training manager needs to compare three AI certification programmes for her company. She types into ChatGPT Deep Research: "Compare the top AI certification programmes available in 2026 for non-technical business professionals. Include pricing, duration, curriculum scope, and employer recognition. Focus on programmes that do not require coding skills." ChatGPT browses the web for 5 minutes and returns a structured comparison table with 8 programmes, covering exactly the criteria she requested, with links to each programme's website.
Q: What makes ChatGPT Deep Research different from a regular ChatGPT prompt?
Deep Research goes beyond a standard chat response by actively browsing the internet, reading multiple web pages, and synthesising what it finds into a detailed, structured report with citations and source links. It typically takes 2-10 minutes and is designed for complex research questions that require information from many sources.
What is a research topic you need to investigate for work this month? Write a Deep Research prompt for it using the research brief format: topic, angle, audience, and constraints.
Google NotebookLM - Your Smart Research Notebook
Google NotebookLM is a free AI research tool that lets you upload your own sources - PDFs, Google Docs, web links, YouTube videos, even audio files - and then ask questions about them. Unlike ChatGPT or Gemini, which answer from their general training data, NotebookLM answers
only from the documents you provide. This makes it ideal for working with specific reports, research papers, company documents, or course materials.
How to use it: Go to notebooklm.google.com and sign in with your Google account. Create a new notebook and upload your sources (up to 50 sources per notebook, with individual files up to 500,000 words). Once your sources are loaded, you can ask questions in the chat panel. NotebookLM reads across all your uploaded sources and gives you answers grounded in your specific documents, with inline citations showing exactly which source each piece of information came from.
What makes it powerful: Traditional research requires you to read each document, highlight key passages, and mentally connect ideas across sources. NotebookLM does this for you. Ask "What are the common themes across all five reports?" and it synthesises information from all your uploaded documents. Ask "Does any source disagree with the claim that remote work improves productivity?" and it finds contradicting evidence in your own materials.
Practical uses for professionals: Upload meeting minutes from the past quarter and ask for a summary of key decisions. Upload competitor annual reports and ask for a comparative analysis. Upload a textbook chapter and ask for a study guide. Upload a contract and ask "What are my obligations under Section 4?" NotebookLM turns static documents into a queryable knowledge base.
Key limitation: NotebookLM only knows what you give it. If a piece of information is not in your uploaded sources, it will tell you it cannot find an answer rather than making one up. This is actually an advantage - it means you can trust that its answers come from your documents, not from general AI knowledge that might be inaccurate.
Watch video: Google NotebookLM - Your Smart Research Notebook
Key Insight: NotebookLM answers only from your uploaded documents - not from general AI knowledge. This means its answers are grounded in your specific sources, with inline citations showing exactly where each piece of information comes from.
Real-World Example: A policy analyst uploads 8 government reports on renewable energy to NotebookLM. She asks: "What targets has Malaysia set for renewable energy by 2030, and which reports mention progress towards those targets?" NotebookLM scans all 8 reports and returns a structured answer with specific targets, progress data, and inline citations showing which report each fact comes from. Work that would take 3 hours of reading takes 5 minutes.
Q: How does Google NotebookLM differ from ChatGPT or Gemini in how it answers questions?
NotebookLM answers only from the documents you upload to it - not from general AI knowledge. This means its answers are grounded in your specific sources with inline citations. ChatGPT and Gemini answer from their broad training data plus web search, which is useful for general questions but less reliable for specific document-based research.
What documents do you work with regularly that you wish you could search more easily - reports, contracts, meeting minutes, research papers? Pick one and try uploading it to NotebookLM this week.
NotebookLM Audio Overviews
One of NotebookLM's most distinctive features is Audio Overviews - it can turn your uploaded documents into a podcast-style audio briefing. Two AI-generated voices discuss your content in a natural, conversational tone, highlighting key points, explaining complex ideas, and making connections between different parts of your sources. The result sounds like a podcast episode created specifically about your documents.
How to generate an Audio Overview: After uploading your sources to a NotebookLM notebook, look for the "Audio Overview" option (sometimes called "Generate audio"). Click it, and NotebookLM will process your sources and create an audio file. Generation typically takes 2 to 5 minutes. Once ready, you can listen directly in the browser or download the audio file.
Why this matters for professionals: Audio is a different way of processing information. Some people understand and retain content better when they hear it discussed rather than read it on screen. Audio Overviews let you "read" a stack of reports while commuting, exercising, or doing other tasks. They also surface insights you might miss when skimming text - the conversational format often highlights connections between ideas that are not obvious when reading.
Customising your Audio Overview: You can guide the focus of the audio by selecting which sources to include and by writing a brief prompt. For example, upload five industry reports and tell NotebookLM: "Focus the audio discussion on market size projections and competitive threats." The generated audio will emphasise those themes rather than trying to cover everything equally.
Practical workflow: Upload your research materials (Deep Research output, PDFs, articles) to a NotebookLM notebook. Generate an Audio Overview. Listen during your commute or while reviewing your notes. Use the key points you hear to structure your outline for the next step - building a presentation.
Limitations: Audio Overviews are a summary, not a replacement for reading the original documents. They work best for getting a high-level understanding and identifying which sections to read in detail. The AI voices are natural-sounding but may occasionally mispronounce technical terms or names.
Key Insight: Audio Overviews turn your uploaded documents into a podcast-style discussion between two AI voices. Listen while commuting or exercising - it is a different way to process and understand your research that often surfaces insights you would miss when skimming text.
Real-World Example: A manager is preparing for a board meeting. She uploads the quarterly financial report, two competitor analyses, and a strategy document to NotebookLM. She generates an Audio Overview focused on "key risks and growth opportunities." During her 30-minute commute, she listens to the AI hosts discuss her documents, highlighting three risks she had not noticed in the financial report. She arrives at the office with a clear mental outline for the board discussion.
Q: What is the primary benefit of NotebookLM's Audio Overviews for busy professionals?
Audio Overviews let you process research content in a different way - by listening rather than reading. This is especially valuable for busy professionals who can listen during commutes or other activities. The conversational format often highlights connections and insights that are easy to miss when skimming text. However, they are summaries, not replacements for reading the originals.
Do you prefer learning by reading or listening? How could Audio Overviews fit into your daily routine - during your commute, lunch break, or exercise?
Creating Presentations with Gamma
Gamma is an AI-powered presentation tool that creates professional slide decks from text input. Instead of starting with a blank slide and manually choosing layouts, fonts, and images, you describe what you want - or paste in your research outline - and Gamma generates a complete presentation with structure, visuals, and formatting.
How to use Gamma: Go to gamma.app and create a free account. Click "Create new" and choose "Presentation." You have two main options: (1) type a prompt describing what you want ("Create a 12-slide presentation on AI adoption trends for SMEs in Southeast Asia"), or (2) paste in your own outline or key points and let Gamma turn them into slides. Gamma generates a full deck in about 30 seconds.
What Gamma does well: Layout and design are handled automatically. Each slide gets a professional layout with appropriate headings, bullet points, images, and spacing. You do not need design skills. Gamma also offers multiple themes so you can match your company's branding or choose a style that fits your audience. The output looks polished enough for client presentations or internal meetings.
The research-to-presentation workflow: This is where the full pipeline comes together. Take your ChatGPT Deep Research output, refine it in NotebookLM, identify your key points, and paste them into Gamma as a structured outline. Gamma turns your outline into a complete presentation. The entire process - from research question to finished slide deck - can take less than an hour.
Editing and refining: Gamma's first draft is rarely perfect. Review each slide and look for: accuracy of content (does each slide say what you mean?), logical flow (do the slides tell a coherent story?), visual consistency (do images and layouts work together?), and audience fit (is the level of detail right for your audience?). You can edit text directly on any slide, regenerate specific slides, swap images, or add and remove slides. Spend 15 to 20 minutes reviewing and refining - this is where your professional judgment adds the most value.
Free tier limitations: Gamma's free plan gives you 400 AI credits (enough for approximately 10 presentations) and includes Gamma branding on slides. Paid plans (starting at USD 8/month) give unlimited AI credits, remove branding, add custom fonts, and unlock export to PowerPoint. For trying out the workflow, the free tier is a great starting point.
Watch video: Creating Presentations with Gamma
Key Insight: Gamma turns text or an outline into a complete, professionally designed presentation in about 30 seconds. The real value is in combining it with your research - paste your Deep Research findings or NotebookLM insights as an outline, and Gamma handles the design.
Real-World Example: After using ChatGPT Deep Research and NotebookLM, a trainer has a clear outline with 8 key points about cybersecurity best practices for SMEs. She pastes the outline into Gamma with the prompt: "Create a professional presentation for Malaysian business owners. Use a clean, corporate theme. Include relevant images." Gamma generates a 12-slide deck in 30 seconds. She spends 15 minutes editing two slides for accuracy and swapping one image. The deck is ready for her workshop the next morning.
Q: What is the most effective way to use Gamma in the research-to-presentation workflow?
The most effective approach is to paste your research outline or key findings into Gamma as structured input. Gamma excels at turning organised content into professional slide layouts with appropriate visuals. The more specific your outline, the better the output. Always review and refine the generated slides before presenting.
Think about the last presentation you created. How long did you spend on design and layout versus content? How would using Gamma change that balance?
Reviewing AI-Generated Presentations
AI can generate a presentation in seconds, but presenting it without review is risky. The review step is where your professional judgment, subject expertise, and audience awareness make the difference between a forgettable slide deck and one that genuinely persuades.
Content accuracy. This is the most critical check. Read every slide and verify that the content is factually correct. AI may rephrase your research in ways that subtly change the meaning. A statistic might be rounded incorrectly. A comparison might oversimplify a complex point. If you used the fact-checking workflow from Module 1, apply the same rigour here - especially for data, quotes, and specific claims.
Logical flow. Do the slides tell a coherent story from beginning to end? A common AI weakness is generating slides that each make sense individually but do not connect well as a sequence. Look for a clear narrative arc: context (why does this matter?), key findings (what did you learn?), implications (what should the audience do?), and next steps (what happens now?). Rearrange slides if the flow feels disjointed.
Audience fit. Is the level of detail appropriate for your audience? A board presentation needs high-level insights with clear takeaways. A technical team briefing needs more detail and data. AI tends to generate content at a middle level of detail - you may need to simplify for executives or add depth for specialists. Remove jargon if your audience is non-technical. Add context if your audience needs background.
Visual quality. Check that images are relevant and professional. AI-selected stock images are sometimes generic or mismatched. Check that text is readable (not too small, not too crowded). Check that charts or diagrams are accurate. Replace any visual that does not add value.
The "so what?" test. For each slide, ask: "If my audience sees only this slide, will they understand why it matters?" If a slide does not clearly answer "so what?", it needs a stronger headline, a clearer takeaway, or it should be removed entirely.
Research-to-Presentation Workflow and Review Checklist
Key Insight: AI generates slides in seconds, but the review step is where your professional judgment adds the most value. Apply five checks: content accuracy, logical flow, audience fit, visual quality, and the "so what?" test.
Real-World Example: A sales director uses Gamma to create a quarterly review presentation. Gamma produces 14 slides. During review, she catches three issues: Slide 4 rounds a revenue figure incorrectly (content accuracy), Slide 7 presents a competitor comparison before explaining the criteria (logical flow), and Slide 11 uses a generic stock photo of a handshake that adds no value (visual quality). She fixes all three in 10 minutes. The final deck is accurate, coherent, and polished.
Q: What does the "so what?" test involve when reviewing a presentation?
The "so what?" test means asking for each slide: "If my audience sees only this slide, will they understand why it matters?" If a slide does not clearly answer that question, it needs a stronger headline, a clearer takeaway, or it should be removed entirely. This test ensures every slide earns its place in the deck.
Think about the last presentation you sat through that felt unfocused or hard to follow. Which of the five review checks would have improved it the most - content accuracy, logical flow, audience fit, visual quality, or the "so what?" test?
Module 3: Your Digital Sidekick
Organise your work. Customise your AI. Build assistants for what you actually do.
Set up ChatGPT Projects to organise your work, build a Custom GPT for a specific business use case, and create a free Gemini Gem as your personalised AI specialist.
Learning Objectives - Set up ChatGPT Projects with custom instructions and reference files
- Build a Custom GPT tailored to a specific business use case
- Write effective system instructions to control assistant behaviour
- Upload knowledge files to give an assistant domain expertise
- Build a Gemini Gem for a specific professional role or task
- Choose between Projects, Custom GPTs, and Gems for different workflows
What You'll Learn - What ChatGPT Projects are and how they differ from regular conversations
- Setting up a Project: instructions, reference files, and a consistent persona
- Custom GPTs: what they are, what they can do, and real business use cases
- Designing your Custom GPT: persona, instructions, and knowledge files
- Gemini Gems: building a free specialised assistant in Google's ecosystem
- Choosing the right tool: Projects vs Custom GPT vs Gem
What ChatGPT Projects Are
If you use ChatGPT regularly, you have probably noticed that every new conversation starts from scratch. The AI forgets everything from your previous chats - your role, your preferences, the documents you shared, the writing style you asked for. You end up repeating the same instructions over and over.
ChatGPT Projects solve this problem. A Project is a workspace inside ChatGPT where you group related conversations, upload reference files, and set custom instructions that apply to every chat within that workspace. Think of it as giving ChatGPT a permanent briefing for a specific area of your work.
How Projects differ from regular conversations:
Regular chat: Each conversation starts with zero context. You must re-explain who you are, what you need, and how you want the output every time. Files you uploaded in a previous chat are not accessible.
Project: Every conversation within a Project automatically inherits the custom instructions and has access to the uploaded reference files. The AI knows your role, your audience, your preferences, and your reference materials from the moment you open a new chat.
What you can add to a Project:
- Custom instructions - Tell the AI your role, audience, tone, preferred output format, and any rules it should follow. Example: "You are a training consultant. Write in simple, jargon-free English. Use bullet points. Keep responses under 300 words."
- Reference files - Upload documents (PDFs, Word files, spreadsheets) that the AI can reference when answering your questions. Free users can upload up to 5 files per Project, Plus users up to 25, and Pro users up to 40.
Projects are free on all ChatGPT plans. The only difference between free and paid is the number of files you can upload. The custom instructions, workspace grouping, and conversation features all work the same way.
Practical examples: A training manager creates a "Workshop Development" Project with instructions to write in a facilitation-friendly tone, and uploads her company's training guidelines PDF. A freelance writer creates a "Client: TechCorp" Project with the client's brand voice guide and product documentation. Every new chat in these Projects starts with full context.
Key Insight: ChatGPT Projects are workspaces where your custom instructions and reference files apply to every conversation. They are free on all plans - the only difference is the file upload limit (5 free, 25 Plus, 40 Pro).
Real-World Example: A consultant creates a Project called "Monthly Reports" with these instructions: "You are a business analyst. Write in formal British English. Use data-driven language. Format all outputs as executive summaries with bullet points." She uploads her company's report template and last quarter's data. Now every time she opens a new chat in this Project, the AI already knows her style, format, and reference data.
Q: What is the main advantage of using a ChatGPT Project over a regular conversation?
The main advantage of Projects is that your custom instructions and reference files persist across every conversation in the workspace. You do not need to repeat your preferences, role, or upload the same documents each time. Projects are free on all ChatGPT plans - the only difference between tiers is the file upload limit.
Think about an area of your work where you keep repeating the same instructions to AI. What custom instructions would you write for a Project to handle this automatically?
Setting Up Your First Project
Setting up a ChatGPT Project takes about five minutes and immediately improves the quality and consistency of every AI conversation you have within that workspace. Here is a step-by-step guide.
Step 1 - Create the Project. In ChatGPT, look for the "Projects" option in the left sidebar. Click "New Project" and give it a clear name that describes its purpose. Good names: "Client Proposals," "Training Materials," "Weekly Reports," "Social Media Content." Avoid vague names like "Work stuff" or "Misc."
Step 2 - Write custom instructions. This is the most important step. Your custom instructions act as a permanent briefing that the AI follows in every conversation within this Project. A strong set of instructions covers:
- Your role: "I am a corporate trainer specialising in leadership development."
- Your audience: "My participants are mid-level managers in Malaysian companies."
- Tone and style: "Write in simple, conversational English. Avoid academic jargon."
- Output preferences: "Use bullet points. Keep responses under 500 words unless I ask for more detail."
- Rules and constraints: "Never make up statistics. Always ask for clarification if my request is unclear."
Step 3 - Upload reference files. Add documents that the AI should use as reference material. These could be your company's brand guidelines, a product catalogue, a training manual, past reports, or any document that provides context. The AI can read and reference these files in any conversation within the Project.
Step 4 - Test with a real conversation. Open a new chat inside the Project and ask a question related to your work. Check whether the AI follows your instructions: Does it use the right tone? Does it reference your uploaded files when relevant? Does it respect your formatting preferences? If something is off, refine your custom instructions.
Pro tip - separate Projects for separate contexts. Do not put everything in one Project. Create separate Projects for different clients, different types of work, or different audiences. A "Client A" Project should have different instructions from a "Training Workshop" Project. This keeps the AI's responses focused and relevant.
Key Insight: A strong Project setup has three parts: a clear name, detailed custom instructions (covering your role, audience, tone, and rules), and relevant reference files. Test the setup with a real conversation and refine the instructions until the AI consistently produces the output you need.
Real-World Example: A HR manager sets up a Project called "Employee Handbook Q&A" with these instructions:
Role: "You are an HR advisor for a Malaysian company with 200 employees."
Tone: "Professional but approachable. Use simple language."
Rules: "Always reference the uploaded Employee Handbook when answering policy questions. If the handbook does not cover a topic, say so clearly."
She uploads the 45-page Employee Handbook PDF. Now when any manager asks a policy question in this Project, the AI answers based on the actual handbook - not general knowledge.
Q: What is the most important step when setting up a ChatGPT Project?
Writing detailed custom instructions is the most important step because these act as a permanent briefing that the AI follows in every conversation. Good instructions cover your role, audience, tone, output preferences, and any rules or constraints. The quality of your instructions directly determines the quality of every response within the Project.
Draft custom instructions for a Project you would use at work. Include your role, your typical audience, your preferred writing style, and one important rule the AI should always follow.
Custom GPTs: Build Your Own AI Expert
A Custom GPT is a personalised version of ChatGPT that you design for a specific purpose. While Projects organise your own conversations, Custom GPTs create a standalone assistant that anyone can use - your colleagues, your clients, or the public. Think of it as building a specialist employee who knows exactly what to do, every time.
What makes a Custom GPT different from a regular ChatGPT conversation?
- It has a fixed persona - it always behaves the same way (friendly customer support agent, strict document reviewer, creative brainstorming partner)
- It has built-in knowledge - you upload documents that give it domain expertise
- It has consistent behaviour - every user gets the same quality experience
- It can be shared via link or published to the public GPT Store
Real business use cases:
- Client onboarding assistant: Upload your onboarding guide. The GPT walks new clients through the process step by step, answering their questions along the way.
- Product FAQ bot: Upload your product catalogue and FAQ document. The GPT answers product questions accurately, referencing your actual materials.
- Writing coach: Configure the GPT with your company's brand voice guide. It reviews drafts and suggests improvements that match your brand.
- Training quiz generator: Upload your course content. The GPT generates quiz questions with correct answers based on the actual material.
- Meeting minute summariser: The GPT takes meeting transcripts and produces structured summaries in your preferred format.
Important access note: Creating Custom GPTs requires a ChatGPT Plus subscription (USD 20/month). However, anyone can use a Custom GPT that has been shared via link - even free ChatGPT users. So you can build a GPT and share it with your entire team or client base without requiring them to pay.
The GPT Store: OpenAI runs a public marketplace of Custom GPTs built by other users. Browse it for inspiration or to find GPTs that solve problems similar to yours. Free users can explore and use any GPT from the Store.
Watch video: Custom GPTs: Build Your Own AI Expert
Key Insight: Custom GPTs are standalone AI assistants with a fixed persona and built-in knowledge. Creating them requires ChatGPT Plus (USD 20/month), but anyone can use a shared Custom GPT for free. Think of them as specialist employees you can deploy to your team or clients.
Real-World Example: A training company builds a Custom GPT called "Course Advisor." It is configured with the persona of a friendly education consultant and uploaded with their full course catalogue (12 courses, pricing, schedules, prerequisites). When a potential client chats with the GPT, it recommends courses based on the client's goals and experience level. The company shares the GPT link on their website. Visitors get instant, personalised course recommendations without needing to talk to a salesperson.
Q: Who can use a Custom GPT that has been shared via link?
Anyone with a ChatGPT account - free or paid - can use a Custom GPT that has been shared via link or published to the GPT Store. Creating a Custom GPT requires a Plus subscription (USD 20/month), but sharing and using them is free for all users.
Think of one task at work that you or your team repeat frequently. Could a Custom GPT handle it? What persona, knowledge files, and instructions would it need?
Designing Your Custom GPT
Building a Custom GPT is a conversation-based process - you describe what you want, and ChatGPT helps you create it. No coding is required. Here is the step-by-step design process.
Step 1 - Open the GPT Builder. In ChatGPT (requires Plus), click "Explore GPTs" in the sidebar, then "Create." You will see two tabs: "Create" (conversational) and "Configure" (manual). Start with "Create" to describe your GPT in plain language.
Step 2 - Define the persona. Tell the builder who the GPT should be. Example: "This GPT is a friendly HR policy advisor. It answers employee questions about leave, benefits, and workplace rules based on the uploaded handbook." Be specific.
Step 3 - Write system instructions. Switch to "Configure" to fine-tune. System instructions are the rules the GPT follows. Include:
-
What it should do: "Answer HR policy questions based only on the uploaded Employee Handbook."
-
What it should NOT do: "Never give legal advice. Never make up policies that are not in the handbook."
-
Tone: "Be friendly and reassuring. Use simple language."
-
Format: "Always cite the relevant handbook section number in your answer."
-
Fallback: "If the handbook does not cover the question, say: 'This is not covered in our current handbook. Please contact HR directly.'"
Custom GPT Architecture: System Prompt + Knowledge Files + User Input = Tailored Response
Step 4 - Upload knowledge files. Upload up to 20 documents (handbooks, product guides, pricing sheets) that give the GPT domain expertise. Responses are then based on your actual content rather than general AI knowledge.
Step 5 - Add conversation starters. Write 3-4 example questions that appear when someone opens the GPT. Examples: "What is the leave policy for new employees?" or "Recommend a course for beginners."
Step 6 - Test and share. Use the Preview panel to test with real questions. Check it follows instructions and references files. Then choose sharing: "Only me" (private), "Anyone with the link" (shareable), or "Public" (GPT Store).
Key Insight: The GPT Builder is a conversation-based tool - describe what you want and ChatGPT helps you create it. The key ingredients are: a clear persona, detailed system instructions (including what NOT to do), knowledge files for domain expertise, and thorough testing.
Real-World Example: A real estate agency builds a Custom GPT called "Property Advisor." Persona: friendly property consultant specialising in Kuala Lumpur. Knowledge files: 15 property listings with prices, floor plans, and neighbourhood details. System instructions: "Answer based on uploaded listings only. Never invent properties. Always include the price range and key features. If the listing does not match what the client wants, say so honestly." Conversation starters: "What 3-bedroom condos are available under RM 800K?" The GPT handles initial client enquiries 24/7.
Q: What is the purpose of the "fallback" instruction in a Custom GPT's system prompt?
A fallback instruction tells the GPT what to do when it receives a question it cannot answer or that falls outside its scope. For example: "If the handbook does not cover this question, say: Please contact HR directly." Without a fallback, the GPT might make up answers, which defeats the purpose of building a knowledge-grounded assistant.
Write system instructions for a Custom GPT that could help your business. Include: what it should do, what it should NOT do, the tone it should use, and its fallback behaviour.
Gemini Gems: Your Free AI Specialist
Google's Gemini Gems are customised AI assistants that you build inside Gemini - and they are completely free to create and use. This makes Gems one of the most accessible ways to build a personalised AI assistant without spending anything.
What is a Gem? A Gem is a Gemini assistant with custom instructions that define its behaviour, expertise, and tone. When you open a Gem, every conversation follows the rules you set. It is similar in concept to a Custom GPT, but free to create and tightly integrated with Google Workspace.
How to create a Gem:
1. Go to gemini.google.com
2. Click "Explore Gems" in the left sidebar
3. Click "New Gem"
4. Give your Gem a name and write detailed instructions
5. Upload up to 10 reference files or link Google Drive documents
6. Test it in the preview panel
7. Save and use it anytime from your Gems library
What makes Gems powerful:
- Free for everyone: Unlike Custom GPTs (which require ChatGPT Plus to create), Gems are free to build on any Google account.
- Google Drive integration: Link Google Drive documents directly to your Gem. When you update the Drive document, the Gem automatically sees the latest version - no need to re-upload.
- Real-time web search: Gems inherit Gemini's built-in web search, so they can answer questions using both your uploaded files and current web information.
- Pre-built Gems gallery: Over 10,000 public Gems are available in Google's gallery, organised by category (productivity, creativity, education, business).
Practical Gem examples:
- Email drafter: Instructions: "Write professional emails in British English. Keep them under 150 words. Always include a clear subject line and call to action." Now every email draft follows your preferred style.
- Meeting prep assistant: Link your meeting agenda Google Doc. Ask the Gem to prepare talking points, anticipate questions, and draft a post-meeting summary template.
- Content reviewer: Instructions: "Review marketing content for clarity, grammar, and brand consistency. Flag jargon. Suggest simpler alternatives." Upload your brand guide as reference.
Free tier limits: Free users can create unlimited Gems with up to 10 files each. The daily prompt limit is approximately 5 prompts across all Gemini conversations (including Gems). Paid plans (Google AI Pro, USD 20/month) give higher prompt limits and access to more powerful Gemini models.
Watch video: Gemini Gems: Your Free AI Specialist
Key Insight: Gemini Gems are free to create on any Google account - no paid subscription required. This makes them the most accessible way to build a personalised AI assistant. Add up to 10 files per Gem, link Google Drive docs, and leverage Gemini's built-in web search.
Real-World Example: A freelance consultant creates a Gem called "Proposal Writer" with these instructions: "You are a business consultant. Write proposals in professional but warm British English. Always structure proposals with: Executive Summary, Objectives, Approach, Timeline, Investment, and Next Steps. Reference the uploaded rate card for pricing." She links her Google Drive rate card (which auto-updates when she changes prices). Now she asks the Gem: "Write a proposal for a 3-day leadership workshop for 25 managers at a manufacturing company." The Gem produces a structured proposal with correct pricing in under 30 seconds.
Q: What is a key advantage of Gemini Gems over Custom GPTs?
The key advantage is that Gemini Gems are completely free to create on any Google account. Creating Custom GPTs requires ChatGPT Plus (USD 20/month). Both tools let you build personalised AI assistants with custom instructions and knowledge files, but Gems remove the cost barrier to building them.
Create a Gem for a task you do at least once a week. What instructions would you give it? What files or Google Drive documents would you link?
Choosing the Right Tool
You now know three ways to customise AI: ChatGPT Projects, Custom GPTs, and Gemini Gems. Each serves a different purpose, and choosing the right one depends on
who uses it and
what it needs to do.
Projects vs Custom GPTs vs Gems - When to Use Each Tool
Use ChatGPT Projects when:
- The assistant is for
your own work only
- You want to organise conversations by topic or client
- You need consistent custom instructions across multiple chats
- You do not need to share the assistant with others
Use Custom GPTs when:
- You want to build an assistant that
others can use (team, clients, public)
- You need a standalone tool with a fixed persona and knowledge base
- You want to publish to the GPT Store or share via link
- You are willing to pay for ChatGPT Plus (USD 20/month) to create it
Use Gemini Gems when:
- You want a free personalised assistant with zero cost
- Your workflow is integrated with
Google Workspace (Drive, Docs, Gmail)
- You want reference files that auto-update via Google Drive links
- You prefer Gemini's real-time web search capabilities
Can you use more than one? Absolutely. Many professionals use Projects to organise their daily work in ChatGPT, build Custom GPTs for their team or clients, and keep Gems for Google Workspace-heavy tasks. The tools complement each other - choose based on who needs the assistant and which ecosystem you work in.
Key Insight: Projects are for organising your own work (free). Custom GPTs are for building tools others can use (requires Plus). Gems are for free personalised assistants with Google Drive integration. Use all three for different purposes.
Real-World Example: A training company uses all three:
- ChatGPT Project "Course Development" - the instructional designer uses it daily with custom instructions and uploaded curriculum standards
- Custom GPT "Course Recommender" - shared via link on the company website for potential clients to explore courses
- Gemini Gem "Weekly Report Writer" - linked to the team's shared Google Drive folder, pulls data from the latest spreadsheet to draft reports
Each tool does what it does best. No single tool replaces the others.
Q: When should you choose a Custom GPT over a ChatGPT Project?
Custom GPTs are the right choice when you need to build an assistant that other people can use. Projects are personal workspaces (only you use them), while Custom GPTs can be shared via link or published to the GPT Store. If you just need to organise your own work, a Project is simpler and free. If others need the tool, build a Custom GPT.
Look at three tasks you do regularly. For each one, which tool would you choose - a Project, a Custom GPT, or a Gem? Why?
Module 4: Wow with Chatbot
Build a chatbot that knows your business. No code. Live on your website.
Plan, build, train, test, and publish a no-code AI chatbot using Jotform AI Agent. Your chatbot answers customer questions from your own knowledge base and goes live on your website.
Learning Objectives - Understand what AI chatbots can do for a business and how they differ from Custom GPTs
- Plan a chatbot's purpose, persona, scope, and fallback behaviour before building
- Build a working chatbot using Jotform AI Agent with no coding required
- Train the chatbot with your own content: documents, URLs, and Q&A pairs
- Test the chatbot by simulating real user questions and edge cases
- Embed or link the chatbot on a website for live visitor use
What You'll Learn - What AI chatbots are and how they differ from Custom GPTs and regular AI chat
- Chatbot use cases: support, lead qualification, product FAQs, appointment booking
- Planning your chatbot: purpose, persona, scope, and fallback behaviour
- Jotform AI Agent: building your chatbot with the no-code visual builder
- Training your chatbot: documents, URLs, Q&A pairs, and the knowledge base
- Testing and publishing: simulating questions, embedding on a website
What AI Chatbots Can Do for Your Business
An AI chatbot is a virtual assistant that lives on your website (or WhatsApp, Messenger, or other channels) and answers questions from visitors automatically, 24 hours a day, 7 days a week. Unlike a static FAQ page where visitors must scroll through a list of questions, a chatbot lets visitors ask questions in their own words and get immediate, conversational answers.
How is a chatbot different from a Custom GPT?
In Module 3, you built Custom GPTs inside ChatGPT. Those are powerful, but they require the user to have a ChatGPT account and go to chatgpt.com to use them. A chatbot, on the other hand, is embedded directly on your website. Visitors do not need any AI account - they just type a question into the chat widget that appears on your page.
What can a business chatbot do?
- Answer customer questions: "What are your prices?" "Do you offer online training?" "Where is your office?" The chatbot answers instantly from your knowledge base.
- Qualify leads: Ask visitors a few questions to understand their needs, then recommend the right product or service - or collect their contact details for follow-up.
- Handle support: Answer common support questions like "How do I reset my password?" or "What is your refund policy?" without a human agent.
- Book appointments: Guide visitors through selecting a date and time, collecting their details, and confirming a booking.
- Provide product recommendations: Based on the visitor's answers, suggest the most suitable product or service from your catalogue.
Why chatbots matter for trainers and coaches: If you run a training business, a chatbot can answer questions about your courses, recommend the right programme based on the visitor's experience level, collect enquiries while you sleep, and handle the repetitive questions that take up your time. It is like having a knowledgeable receptionist who never takes a day off.
The key difference from general AI chat: A chatbot trained on your knowledge base answers only from your content. It does not make up answers from general internet knowledge. If it does not know the answer, it tells the visitor to contact you directly. This is critical for trust - you control what the chatbot says.
Watch video: What AI Chatbots Can Do for Your Business
Key Insight: An AI chatbot lives on your website and answers visitor questions 24/7 from your own knowledge base. Unlike Custom GPTs (which require a ChatGPT account), chatbots are accessible to any website visitor. Unlike static FAQ pages, chatbots let visitors ask questions in their own words.
Real-World Example: A training company has 5 courses. Every week, they receive 30+ WhatsApp messages asking the same questions: "What courses do you offer?" "How much does it cost?" "When is the next intake?" "Do you have online options?" They build an AI chatbot, train it with their course catalogue, and embed it on their website. The chatbot handles 80% of these enquiries instantly. The team now only responds to complex questions that need a personal touch.
Q: What is the key difference between a Custom GPT and a website chatbot?
The key difference is accessibility. A chatbot is embedded directly on your website - any visitor can use it without creating an account. A Custom GPT lives inside ChatGPT and requires the user to have a ChatGPT account. For customer-facing use cases, a chatbot reaches a much wider audience.
What are the top 5 questions your website visitors or potential clients ask most often? Could a chatbot handle these automatically?
Planning Your Chatbot
Building a chatbot without a plan is like writing a report without knowing the audience. Before you touch any tool, spend 15 minutes answering four planning questions. This prevents the most common chatbot mistakes: chatbots that try to do everything, chatbots that give wrong answers, and chatbots that frustrate visitors instead of helping them.
Question 1 - Purpose: What is the chatbot's job?
Be specific. "Help visitors" is too vague. "Answer questions about our 5 training courses and collect enquiry details for courses the visitor is interested in" is specific enough to build. Write one sentence that describes the chatbot's primary job.
Question 2 - Persona: How should the chatbot behave?
Give your chatbot a personality that matches your brand. A law firm's chatbot should be professional and precise. A yoga studio's chatbot should be warm and encouraging. Define:
- Name (optional but helps)
- Tone (professional / friendly / casual)
- Language style (formal / conversational)
- Response length (short and snappy / detailed and thorough)
Question 3 - Scope: What should it answer, and what should it NOT answer?
This is the most important planning step. List the topics the chatbot
should handle (your courses, your prices, your contact details). Then list the topics it should
refuse to answer (competitor comparisons, medical advice, legal opinions, anything outside your expertise). A chatbot that tries to answer everything will eventually say something wrong.
Question 4 - Fallback: What happens when it does not know the answer?
Every chatbot will encounter questions it cannot answer. Plan what happens next:
- "I'm sorry, I don't have that information. Would you like me to connect you with our team? You can email us at hello@example.com or WhatsApp us at +60 12-345 6789."
- Never let the chatbot guess or make up an answer. A clear fallback builds trust.
Chatbot Conversation Flow: Knowledge Base Check with Fallback
Key Insight: Plan before you build. Answer four questions: What is the chatbot's job? How should it behave? What topics are in scope (and out of scope)? What happens when it does not know the answer? This 15-minute planning step prevents the most common chatbot mistakes.
Real-World Example: A freelance coach plans her chatbot:
Purpose: "Answer questions about my 3 coaching packages and collect enquiry details."
Persona: "Friendly, warm, professional. Name: Coach Ally. Conversational tone."
Scope: IN - coaching packages, pricing, schedule, testimonials, booking process. OUT - personal advice, competitor comparisons, medical or legal topics.
Fallback: "Great question! I don't have that specific info. You can reach Coach Sarah directly at +60 12-345 6789 or sarah@example.com."
Q: Why is defining the chatbot's scope the most important planning step?
Defining scope prevents the chatbot from giving wrong or unreliable answers. A chatbot trained on your course catalogue should answer course questions confidently but refuse to give medical advice or legal opinions. Without a clear scope, the chatbot will attempt to answer everything and eventually produce something inaccurate, damaging your credibility.
Write a one-sentence purpose statement for a chatbot you could use in your business. Then list 3 topics it SHOULD answer and 3 topics it should REFUSE to answer.
Building with Jotform AI Agent
Jotform AI Agent is a no-code chatbot builder that lets you create, train, and deploy AI chatbots without writing a single line of code. It is part of the Jotform platform (known for its form builder) and is designed for non-technical users who want a professional chatbot for their business.
Why Jotform AI Agent?
-
Free tier: 5 AI agents, 100 conversations per month, 10 million-character knowledge base. No credit card required.
-
No-code builder: Everything is done through a visual interface - no programming or technical skills needed.
-
Multi-channel: Deploy to your website, WhatsApp, Messenger, SMS, or even phone (voice calls).
-
7,000+ templates: Start from a ready-made template and customise it for your business.
Building your first chatbot - step by step:
Step 1 - Sign up. Go to jotform.com/ai/chatbot and create a free account. You can also sign in with your Google account.
Step 2 - Create a new AI Agent. Click "Create Agent" and choose either "Start from scratch" or browse the template gallery. Templates cover common use cases like customer support, lead generation, product recommendations, and appointment booking.
Step 3 - Set the persona. Give your chatbot a name, set its tone (professional, friendly, or casual), and write a description of its role. This is where your planning from the previous section comes in. Example: "You are CourseBot, a friendly assistant for ABC Training. You answer questions about our courses and help visitors find the right programme."
Step 4 - Configure the behaviour. Set the default language, response style, and any rules. You can tell the chatbot to always greet visitors, ask follow-up questions, or collect contact details at the end of the conversation.
Step 5 - Customise the appearance. Choose colours, add your logo, and match the chatbot's look to your brand. The widget will appear on your website as a small chat icon in the corner of the screen.
Watch video: Building with Jotform AI Agent
Key Insight: Jotform AI Agent offers a genuinely free tier: 5 chatbots, 100 conversations per month, and a 10 million-character knowledge base. No credit card required. Build, train, and deploy a professional chatbot entirely through a visual interface - no coding needed.
Real-World Example: A training company wants a chatbot on their website. They sign up for Jotform AI Agent (free), choose the "Course Advisor" template, rename it to "LearnBot," set the tone to "friendly and professional," and write: "You help visitors learn about our training programmes. Answer questions about course content, pricing, schedules, and how to enrol. If you do not know something, direct them to email us." The basic chatbot is running in under 10 minutes.
Q: What does the free tier of Jotform AI Agent include?
Jotform AI Agent's free tier includes 5 AI agents, 100 monthly conversations, and a 10 million-character knowledge base. No credit card is required. This is enough to prototype and run a small-scale chatbot for a business website.
Go to jotform.com/ai/chatbot and browse the template gallery. Which template is closest to what you would need for your business?
Training Your Chatbot
A chatbot is only as good as the knowledge you give it. Without training, it is just a generic AI that does not know anything about your business. Training is the step that transforms a blank chatbot into one that answers questions accurately from
your actual content.
What is a knowledge base?
The knowledge base is the collection of information your chatbot draws from when answering questions. When a visitor asks "What courses do you offer?", the chatbot searches its knowledge base for the answer. If the answer is there, it responds accurately. If not, it triggers the fallback.
Three Training Sources Feed Your Chatbot's Knowledge Base
Three ways to train your chatbot:
1. Upload documents. Upload PDFs, Word documents, or spreadsheets containing your business information. Examples: course catalogue, pricing sheet, company profile, product descriptions, policies.
2. Add website URLs. Paste links to your website pages. The chatbot reads the page content and adds it to its knowledge base. This is particularly useful for existing FAQ pages, service descriptions, and blog posts.
3. Write Q&A pairs. Manually type common questions and their ideal answers. This is the most precise method because you control exactly how the chatbot responds. Example:
- Q: "What are your office hours?"
- A: "We are open Monday to Friday, 9am to 6pm (Malaysian time). We are closed on weekends and public holidays."
Best practice - combine all three: Upload your main documents for comprehensive coverage, add your website URLs for consistent messaging, and write Q&A pairs for the most common questions where you want a specific answer format. The combination creates a robust knowledge base that handles most visitor questions accurately.
Key Insight: Train your chatbot from three sources: uploaded documents, website URLs, and manual Q&A pairs. The combination creates a knowledge base that lets the chatbot answer from your actual content, not from general AI knowledge. The free tier supports up to 10 million characters of training data.
Real-World Example: A training company trains its chatbot with:
1. Documents: Course catalogue PDF (12 courses with descriptions, prices, schedules)
2. URLs: Their "About Us" page, FAQ page, and "How to Register" page
3. Q&A pairs: 15 hand-written answers to their most common questions (pricing, refund policy, online vs in-person options, certificate details)
Total training time: 20 minutes. The chatbot now accurately answers questions about any of their 12 courses, directs visitors to the registration page, and gives consistent answers about policies.
Q: What is the most precise method for training a chatbot to give a specific answer to a common question?
Manual Q&A pairs are the most precise training method because you control both the question and the exact answer. For your most common and important questions (pricing, hours, policies), writing specific Q&A pairs ensures the chatbot gives exactly the answer you want, in exactly the format you want.
List 5 documents or web pages you already have that could immediately be used to train a chatbot. What knowledge gaps would you fill with manual Q&A pairs?
Testing Your Chatbot
A chatbot that has not been tested is a chatbot that will embarrass you. Testing is where you catch wrong answers, awkward responses, and missing information before real visitors encounter them. Spend at least 20 to 30 minutes testing before you publish.
Test with real questions. Think like a visitor, not like the person who built the chatbot. Ask the questions your actual customers ask:
- "How much does your workshop cost?"
- "Do you offer online training?"
- "What is included in the premium package?"
- "Can I get a certificate?"
Test edge cases. Ask questions that push the boundaries of the chatbot's knowledge:
- "Can you compare your courses with [competitor]?" (should refuse)
- "Give me a refund" (should direct to human support)
- "Tell me a joke" (should stay professional or redirect)
- "What is the weather today?" (should explain it only handles business questions)
Test different phrasings. Visitors will not always ask questions the way you expect. Test variations:
- "How much?" vs "What's the price?" vs "Pricing please" vs "I want to know the cost"
- "When is the next class?" vs "Schedule" vs "When do you run workshops?"
Test the fallback. Ask something the chatbot definitely does not know and verify that the fallback response is helpful, professional, and provides a way to reach a human.
Common issues to look for:
- Wrong facts: The chatbot states an incorrect price, wrong date, or outdated information. Fix by updating the knowledge base.
- Too much information: The chatbot dumps everything it knows instead of answering the specific question. Fix by refining instructions ("Keep answers concise - under 100 words unless the visitor asks for more detail").
- Off-topic answers: The chatbot answers questions outside its scope. Fix by adding scope restrictions in the instructions.
- Generic responses: The chatbot gives vague answers that could apply to any business. Fix by adding more specific content to the knowledge base.
Testing is iterative. After each round of testing, update the knowledge base, refine the instructions, add missing Q&A pairs, and test again. Two to three rounds of testing typically catch most issues.
Key Insight: Test with real questions, edge cases, different phrasings, and off-topic requests. Look for wrong facts, overly long responses, off-topic answers, and generic responses. Testing should take at least 20-30 minutes and 2-3 rounds before publishing.
Real-World Example: A trainer tests her chatbot with these questions:
"How much is the leadership workshop?" ✓ Correct price
"Do you do online?" ✓ Accurate response about virtual options
"What makes you better than XYZ Training?" ✗ The chatbot compared competitors - FIX: add "Never compare with competitors" to instructions
"I need a refund" ✗ The chatbot processed a fake refund - FIX: add fallback "For refunds, please contact us directly at..."
"Tell me a joke" ✗ The chatbot told a joke instead of staying professional - FIX: add "Stay focused on business topics only"
After 3 rounds of testing and fixing, the chatbot handles 95% of questions correctly.
Q: Why should you test your chatbot with different phrasings of the same question?
Visitors will ask questions in their own words - "How much?", "Pricing?", "What does it cost?", "Tell me the price" all mean the same thing but are phrased differently. Testing with variations ensures the chatbot handles natural language diversity and does not only respond to the exact phrasings you used in your Q&A pairs.
Think of 3 "edge case" questions a mischievous visitor might ask your chatbot. How should your chatbot respond to each one?
Publishing Your Chatbot
Your chatbot is built, trained, and tested. Now it is time to make it live. Jotform AI Agent offers several ways to deploy your chatbot so visitors can interact with it.
Option 1 - Website embed (most common). Jotform provides an embed code - a small piece of HTML that you paste into your website. The chatbot appears as a small icon in the bottom corner of your page. When visitors click it, the chat window opens.
How to embed: In Jotform AI Agent, click "Publish" on your agent, select "Embed," and copy the code snippet. Paste it into your website's HTML just before the closing
