Key Takeaways

• Slides fail when they try to be both teaching tools and study materials, creating confusion instead of clarity.

• AI’s real power is transforming rich, messy presentations into structured, complete study manuals.

• Using dictation and asking AI to craft the perfect prompt dramatically increases the quality of the output.

• Deep Search is essential when content must be accurate, referenced and scientifically reliable.

• Human validation turns AI’s plausible draft into a trustworthy, reusable knowledge base for all future learning materials.

Why this problem affects almost all teachers and trainers

Anyone who teaches at a university, in a company, or in a classroom eventually faces the same dilemma.
If I prepare rich slides, full of details, definitions, and explanations, they can help students study. But they become unmanageable during the lesson.

The audience ends up reading and, as we know well, those who read don’t listen (if you’re curious about the scientific evidence behind this, read my guide on Neuro Presentation Design).

You end up with slides that are impossible to present and a PowerPoint document you ask students to study from that is not optimized at all for effective learning.

Bottom line: we’ve created a PowerPoint that is “neither fish nor fowl” and doesn’t fulfill any of its purposes well.

In this guide, I’ll show you how, with a touch of AI, we can separate the two moments, making slides light for teaching and creating a complete manual for studying, starting from materials you probably already have.

AI shouldn’t remake slides. It should transform them.

AI is not a great designer yet. Its value is not in remaking slides (for now), but in guiding us through the entire creative process of building and transforming content.

In this guide, we’ll use ChatGPT and Microsoft Copilot to turn content into a complete, structured manual, organized into chapters, with summaries and references, starting from a real university lecture presentation.

AI’s real strength lies in its ability to maximize the value of the content we already have: a rich presentation, full of notes and annotations, is the ideal foundation for extracting an organic and coherent manual, divided into modules that are easy to consult and adapt to learners’ needs.

Instead of focusing on slide aesthetics, AI allows you to reorganize and enrich the content, expanding explanations where needed, adding practical examples and synthesizing key passages, until you get a document that not only accompanies the lesson but remains as a structured, updatable reference for any learning path.

From this point of view, AI is a powerhouse.

The raw material: a complex presentation works great

The initial presentation can be full of text, paragraphs, notes and even speaker notes. The more raw material you have, the better AI can transform it into an accurate manual.

In this case, we analyze the real presentation of a medical course, created with a double purpose: to support the lecturer during the lesson and to help students prepare for the exam.

However, as you can clearly see, this presentation is difficult to deliver and causes too much distraction during the lesson, while still being a suboptimal tool for independent study sessions.

For us, this is a rough diamond, because it contains all the contextual information the AI needs to create the complete exam preparation guide.

Prompting techniques

In the video, I used two very simple prompting techniques that significantly improve productivity and output quality. Let’s look at them:

Dictating the context is more effective than writing it

Dictating context to AI is much more effective than writing a prompt. You tell it what you want to achieve, who it’s for, which scientific or academic constraints it must respect.

When dictating, you don’t worry about punctuation or forming a perfect sentence, you don’t stop to edit.

While speaking to a person we’d need coherent discourse, with AI we can spill out all our thoughts, even in a disorganized way. This lets us quickly give a lot of contextual information without wasting time on prompt “cosmetics.”

A richer context means a more precise input and therefore higher-quality output.

In the video, I dictated the entire first part of the prompt.

Letting AI write the perfect prompt

If I asked you to look up from your screen right now and write a precise prompt describing everything you see, you’d quickly realize how complex this task actually is.

If you tried to recreate what you see using, say, Microsoft Copilot, you’d probably iterate several times and discover that your description omitted details you didn’t realize were important.

To get the most complete prompt possible and avoid forgetting key parts, I prefer to ask the model to ask me whatever it needs in order to write the perfect prompt itself.

So before generating the manual, we ask the model to write the best possible prompt based on the context we provided. The model defines role, objective, constraints, structure and format.

Comparing ChatGPT and Microsoft Copilot

In the video, I decided to work with both Microsoft Copilot and ChatGPT to compare their outputs and choose the best one. Outside of corporate settings, we can use multiple tools, and testing them helps deepen our AI Fluency (AI Fluency: the new language of collaboration between humans and AI). You’ll notice that each model produces a different type of manual.

Both documents are rich in information and great for independent study. Both contain more than what was shown in the presentation and integrate information from external sources. Each model tapped into different sources but both seem reliable.

I particularly appreciated that Copilot created interactive sections with a more visual, graphic approach.

The document produced by GPT is articulated, fluid and narrative. It resembles a university textbook: clearly defined chapters, introductory paragraphs, extended explanations, concept connections and summary sections. The writing flows well and the document has strong internal cohesion—it guides the learner through a logical journey, as if designed to be read from beginning to end. It’s a document that explains, not just informs.

The document generated by Copilot is more concise and schematic. It maintains a clear structure but tends to present content more directly and concisely. Definitions are there, topics are correct and well divided, but each section is more compact and less narrative. The advantage is that the document is easy to scan and suitable for those who want an organized summary or an overview. It’s a document that lists, more than it explains.

In short:

• The GPT document is better for in-depth study because it builds a real learning pathway.
• The Copilot document is better for quick consultation because it prioritizes clarity and synthesis.

Both are correct and useful, but they serve different needs: studying vs. reviewing.

What Deep Search is and when to use it

Deep Search is not for “generating text”, it’s for reconstructing knowledge.

It works differently from standard generative models: instead of producing content based only on model reasoning, it analyzes documents, scientific articles, and reliable sources, and integrates them into the final content. It’s like asking AI not only to write, but also to research.

When is it needed?
Deep Search becomes essential when:

• Content is scientific, technical, or regulated, no room for ambiguity
• You need verifiable data, citations, official definitions, reference values, and recognized classifications
• You want bibliographies, comparative tables, guidelines, or clinical charts
• The document must be defensible in university, healthcare, or corporate settings

In this project, turning a university medical presentation into a study manual, Deep Search is perfect because it:

1. Interprets and enriches: understands the content and expands it using solid scientific sources.
2. Cites and organizes: allows adding references, official guidelines, recognized definitions.
3. Builds rigor: makes the manual not only readable, but reliable and compliant with educational standards.

Deep Search is not necessary for writing a narrative text or summary; it is necessary when the text must be accurate, referenced, and verifiable.

In other words, LLMs elaborate, Deep Search documents. In GPT, Deep Search is an option available in every chat. In Microsoft Copilot, you must use one of its official agents: Researcher or Analyst.

Validation and revision of AI-generated documents

This is where the real difference emerges between using AI and mastering it. In the AI Fluency framework, I say it clearly: AI output is always unreliable by definition. Not because it’s “wrong,” but because it’s just a starting point a statistically generated compromise, not validated, not reasoned, not contextualized. It’s “good enough” to inspire, but not safe enough to approve, especially in medicine.

This is where the expert comes in.

The professional reads, verifies, corrects, improves, adapts to context, and only then approves. This process is not a waste of time; it is the very thing that lets us leverage AI’s power without falling into its traps.

It means turning an unreliable but useful output into an accurate, safe, educationally effective and professionally coherent document.

An AI-generated document, on its own, is like an unedited draft. In expert hands, it becomes validated, reliable, compliant and responsible content.

This is the real difference between those who “prompt and copy” and those who use AI as a professional value lever.

It is not the machine that creates quality.
It is human expertise that guides it, filters it and brings it to destination.

The final result is not just a simple document, but a validated knowledge base

From this review, we didn’t redesign the slides; we did something more important: we built a structured, readable, and reliable version of the content, ready to be taught, studied and above all, updated over time.

The document becomes the original core from which you can efficiently generate:

• simplified versions for quick study
• outlines, concept maps, exam questions
• handouts, summaries, and materials for lectures
• future slide decks, with validated, consolidated content

The chat that generated and optimized this content is not “archived.”
It becomes a continuous educational asset, reusable to update, expand, or deepen any section without starting from scratch.

We turned a draft into a dynamic, reliable, regenerable document that grows over time along with the knowledge.

Conclusion

This work demonstrates that building a structured and validated knowledge base creates far more value than simply generating documents or slides.

The proposed approach transforms a draft into a dynamic and reliable system, ready to be taught, studied and continuously updated. The document becomes not an endpoint, but the starting point for developing teaching materials, summaries, concept maps and other learning tools, ensuring clarity, reusability and ongoing updates.

The integration of AI as a support optimizes content production while keeping quality high thanks to professional review.

This methodology is effective not only in universities, but also in corporate and healthcare contexts, becoming a valuable, continuously regenerable learning asset.

FAQ

Why must AI output always be reviewed, even when it looks correct?
Because AI doesn’t reason: it predicts. It generates “probable” text, not necessarily “reliable” text. Especially in medicine, it’s not enough for content to seem correct, it must be correct. Human review turns statistically plausible content into scientifically valid content. That’s where true professional value is created.

Can I create slides directly with AI, without starting from a document?
You can, but it’s a classic mistake. Slides are the final form, not the source. You first need a base content: structured, validated, in-depth. Only then should you generate clear and reliable slides. Skipping this step leads to errors, superficiality and materials that are hard to reuse later.

Can I upload a PDF or a complex book and have AI process it?
Yes, but you must first convert it into a format AI can read properly (Word or PowerPoint). PDFs are only partially readable and often lose titles, structure and logical flow. AI works best with organized content: sections, titles, lists, diagrams.

Can AI invent guidelines, dosages or clinical protocols?
If not instructed correctly, yes. To avoid this, specify clearly in the prompt that it must stick to provided sources or official guidelines (e.g., AIFA, AHA, WHO) and must not fabricate data. A professional must then validate every piece of information before use.

Can I use this methodology to create exam questions or clinical cases?
Absolutely, especially when using Deep Search. AI can generate open-ended questions, quizzes, clinical cases, reasoning schemes, discussions and even evaluation rubrics. But clinical review is always necessary.

Is this approach useful only in universities, or also in corporate and healthcare settings?
A: It works anywhere reliable, updatable and reusable content is needed. It’s perfect for:

• medical education and CME
• corporate training and onboarding
• internal documentation
• procedure manuals and protocols
• blended learning or LMS content

Once the document is created, is the job done?
No, it’s just beginning.
The AI chat becomes a living asset you can reopen to update guidelines, create new versions, produce annual revisions, generate handouts, concept maps or quizzes. You didn’t create a file, you created a knowledge base.