This post is for my friends on Facebook (yes, I am on Facebook, I’m old) who asked me how I got ChatGPT to cooperate with me when previously I had issues.
To offer some background: a few days ago I started working with ChatGPT to create a PowerPoint deck on cybersecurity. I uploaded a number of my own decks and articles along with some ethics opinions. For whatever reason, ChatGPT and I had trouble getting this done properly, until we set some guidelines and broke the process down into steps.
If you’ve ever felt like AI is “brilliant but chaotic,” this is basically the antidote: you don’t ask it for everything at once. You give it constraints, inputs, and steps—like you would a smart intern who needs supervision.
I Already Set Working Guidelines with ChatGPT
The guidelines we set involved ChatGPT promising it would tell me when it couldn’t do something, and also that it would not try beyond 30 minutes before it would tell me that it was having problems. Sometimes ChatGPT isn’t quite sure what it can and cannot do, and this can lead to wasted time and frustration.
We also dealt with the issue of hallucinations (though I still have to be careful to check). In other words: I assume I’m responsible for verification, just like I would be with any other tool.
And this is also where the “lawyer ethics” part matters. The ABA has now issued formal guidance on lawyers using generative AI tools (competence, confidentiality, communication, supervision, fees, etc.). If you’re a lawyer reading this and thinking “this feels like a professional responsibility issue,” you’re right.
Why a Structured Workflow Matters with ChatGPT
A structured workflow:
- Prevents hallucinations — AI inventing details
- Keeps timing and pacing under control
- Helps you extract value from PDFs/PPTs reliably
Also, it’s important to understand that AI has certain capabilities. ChatGPT can generate structured content, outlines, scripts, and even PowerPoint files, but it does not replace the human judgment required to design a polished, presentation-ready deck with appropriate visual layout, branding, and accessibility considerations.
There are tools that draft slides directly inside PowerPoint or automatically generate slide layouts, but that’s a separate category from using ChatGPT as a thinking and writing partner.
ChatGPT Shows Me How to Prompt It
Now I am moving on to creating a new deck for an artificial intelligence talk. Again, I want to use my own prior decks, articles, and bar association ethics opinions. Here is the conversation I had with ChatGPT to get started.
Me: I want to create an AI lecture based on my PowerPoints, ethics opinions, and some articles I wrote. Then I want to do a script. It needs to be 70 minutes long. Given we worked out how to do this with cybersecurity this should be possible. Do you have suggestions for how I should go about giving you data and what prompts I should use?
Everything Below Is Exactly What ChatGPT Gave Me in Response to My Request
ChatGPT’s response: Yes — this is absolutely possible, and you’re right: it should follow the same workflow we refined for your cybersecurity CLE. The key is how you give me the source material and how you sequence the prompts so we don’t lose fidelity, ethics accuracy, or timing control.
Next, based on our history working on the cybersecurity deck, along with its own knowledge and history of working with others, no doubt, this is what ChatGPT told me to do.
Below is a battle-tested, CLE-specific workflow, plus copy-paste prompt templates you can reuse.
Recommended Workflow for a 70-Minute AI / Ethics CLE
Phase 0 — Ground Rules (Do This Once)
Start the project with a control prompt so I don’t hallucinate, compress, or editorialize.
Prompt 0 (Project Control Prompt):
We are creating a 70-minute prerecorded CLE on AI and legal ethics.
Constraints:
No hallucinated law, opinions, or cases
If something is missing or ambiguous, flag it instead of guessing
Use my voice and teaching cadence (measured, practical, ethics-focused)
Assume Pennsylvania + ABA Model Rules unless otherwise specified
Output must be CLE-ready (speaker script, not an outline)
Timing matters — we must hit ~70 minutes
Acknowledge and wait for materials.
✅ This sets your “hard stops” and avoids optimistic filler.
Phase 1 — Source Intake (DO NOT MIX TYPES)
Critical rule:
👉 Never upload or paste mixed source types in the same turn.
This is the #1 reason long CLE projects go sideways.
A. PowerPoints (Structure + Timing)
Upload PPTs first, one deck at a time.
Prompt 1A (for each deck):
This is a PowerPoint deck for an AI ethics CLE.
Task:
Extract slide topics and flow
Estimate speaking time per slide (CLE pacing, not TED talk)
Do NOT write script yet
Identify where ethics rules or opinions are referenced
Return:
Slide-by-slide table: topic | estimated minutes | notes
➡️ This lets us lock timing before words are written.
B. Ethics Opinions (Authority Lock)
Upload opinions separately (PDF or pasted text).
Prompt 1B:
These are ethics opinions relevant to AI use by lawyers.
Task:
Identify the exact rules implicated
Summarize holdings accurately
Note jurisdictional limits
Flag any conflicts or unresolved questions
Do NOT paraphrase loosely or generalize.
Return bullet summaries suitable for CLE citation.
➡️ This prevents misstatements that could torpedo CLE approval.
C. Articles (Voice + Policy)
Upload your articles separately from third-party articles.
Prompt 1C (Your articles):
These are articles I wrote.
Task:
Extract key arguments
Preserve my tone and phrasing where possible
Identify sections suitable for spoken delivery
Return talking-point bullets in my voice.
Prompt 1D (Others’ articles):
These are third-party articles.
Task:
Extract defensible, consensus positions
Note where authors disagree
Avoid adopting strong claims unless widely supported
Return neutral summaries with attribution notes.
Phase 2 — Build the Master Timing Map (Non-Negotiable Step)
Once all sources are processed:
Prompt 2 (Timing Architecture):
Using the slide timing estimates and source materials,
create a 70-minute CLE timing map with:
Section titles
Minutes per section
Learning objective per section
Where ethics rules are discussed
Where hypotheticals or examples appear
Do not write script yet.
Confirm total time = ~70 minutes.
➡️ This is where we prevent the “oops, this is 95 minutes” problem.
Phase 3 — Script Writing (Chunked, Controlled)
Never ask for a 70-minute script in one go.
We write 10–15 minute blocks, exactly like cybersecurity.
Example: First Script Chunk
Prompt 3A:
Write the speaker script for Section 1 (minutes 0–12).
Requirements:
CLE tone, not marketing
Spoken cadence (complete sentences, no bullets)
Cite rules and opinions accurately
Include brief pauses and emphasis cues
No placeholders
End with: [SECTION COMPLETE — approx. X minutes]
Phase 4 — Cadence + Timing Validation (Important for You)
After each chunk:
Prompt 4:
Estimate spoken runtime at CLE pace.
Flag any sections that run long or short.
Suggest trims ONLY if necessary.
This is especially useful since you:
Tend to over-record (your words, not mine 🙂)
Want approval-safe pacing
Phase 5 — Final Assembly Outputs
Once all sections are approved:
You can ask for:
✅ Full teleprompter-ready Word doc (PromptSmart-friendly)
✅ Large-print / accessibility version
✅ Slide-sync notes
✅ Ethics citation appendix
What to Upload vs. Paste (Practical Rules)
Content Type Best Method
PowerPoint Upload file
Ethics opinions Upload PDF
Your articles Paste or upload
Short excerpts Paste
Timing instructions Paste
Style constraints Paste
Red Flags I Will Warn You About (By Design)
Per your standing agreement, I will explicitly tell you if:
A source conflicts with another
A rule is jurisdiction-specific
A section cannot reasonably fit its time slot
CLE approval risk exists
No “we’ll figure it out later.”
If You Want, Next Step
If you’re ready, the next best move is:
👉 Upload one PowerPoint deck and say:
“Phase 1A — analyze this deck only.”
We’ll lock timing first — everything else flows cleanly from there.
Conclusion – Work With the AI, Not Against It
First, it is important to understand that the more you work with an AI like ChatGPT, the better it understands you. Second, it is easy to confuse an AI, and when you do, a mess can result. Given this, sometimes it is best to actually ask the AI how to ask it to do something.
As you can see, at least in my case, it gave me very clear instructions, based on our prior history of creating a similar slide deck and presentation.
Working with the AI and not fighting against it tends to get better results. Third: you don’t just ask ChatGPT to “make a deck.” You guide it through a series of controlled steps, each with clarity about inputs and expectations.
Additional Resources
I asked ChatGPT to review my blog post and make certain it was a fair assessment of what happened and where we are going. It recommended I provide a list of useful resources to people reading the post, so I am including those recommendations from ChatGPT here.
- OpenAI — Prompt engineering best practices for ChatGPT
- Official guidance from OpenAI explaining how prompt structure, constraints, and sequencing affect output quality. This provides general context for why a structured, step-by-step approach works better than open-ended requests.
- ABA — ABA issues first ethics guidance on a lawyer’s use of AI tools (news summary of Formal Opinion 512)
- A plain-language overview of Formal Opinion 512, summarizing the key ethical considerations for lawyers using generative AI tools without requiring a full read of the opinion.
- ABA Formal Opinion 512 (PDF)
- The ABA’s formal ethics opinion addressing lawyers’ use of generative AI tools, including issues of competence, confidentiality, supervision, communication, and fees.
- Microsoft Support — Create a new presentation with Copilot in PowerPoint
- An overview of Microsoft’s built-in AI tools for PowerPoint, useful for understanding how AI-assisted slide creation differs from using ChatGPT as a content-generation and workflow tool.
- 24Slides — How to Create a PowerPoint with ChatGPT for Free
- A practical walkthrough of one common approach to using ChatGPT in a PowerPoint workflow, illustrating general methods and limitations that contrast with the more controlled CLE-specific process described above.
Common Questions About Using ChatGPT for PowerPoints
- ChatGPT can generate PowerPoint files and slide content, but it does not replace the human judgment required to design, refine, and approve a CLE-ready presentation.
- Is this better than built-in AI tools? ChatGPT gives control over structure and ethics content, which many built-in tools don’t handle deeply.
- How does this workflow differ from generic methods? It’s tailored for substantive CLE content with timing and source integrity.