ZenkenAI
Published: Updated:

ChatGPT Slide Making Techniques — Templates, Marp, and Auto-Generation

ChatGPT presentation slides Marp python-pptx design
Table of contents 22 items
Zenken is featured as an OpenAI customer story

Most ChatGPT slide guides cover the basic flow — write a prompt, get a draft, export. That gets you to a working deck, but it doesn’t get you to a deck that looks polished, scales across a team, or ships in editable formats your colleagues can actually modify.

This article skips the introductory tour and goes straight to the techniques that compound: reusable prompt templates, design rules ChatGPT reliably follows, Marp theming for branded output, and the python-pptx route for native PowerPoint files.

If you’re new to making slides with ChatGPT, start with How to Make Slides with ChatGPT — From Outline to Deck for the end-to-end workflow. The techniques below assume you’ve built at least one deck and want to make the next one faster, better, and more on-brand.

A reusable prompt template that beats one-shot prompts

The single biggest quality jump comes from prompting in structured sections instead of writing freeform paragraphs. ChatGPT follows structured input far more reliably, and you can copy-paste the template across decks.

The “context block” template

Save this somewhere you can reach quickly — a notes app, a snippet manager, or a custom GPT system prompt:

✓# Context
- Goal: <one sentence — what do you want the audience to do after seeing this?>
- Audience: <role, seniority, prior knowledge>
- Setting: <board meeting / sales pitch / training session / conference talk>
- Time: <minutes you have to present>
- Tone: <formal | conversational | technical | persuasive>

# Constraints
- Slide count: <number>
- Per-slide limit: <e.g. 1 idea, max 3 bullets, max 12 words per bullet>
- Visuals: <charts | photos | diagrams | none>
- Brand: <color palette, font preferences, logo placement>

# Source material
"""
<paste reference docs, data, prior decks, or notes>
"""

# Task
<exactly what you want this turn to produce — outline / one slide / full deck / speaker notes>

Three things make this template work:

  1. Goal and audience first. ChatGPT calibrates word choice, depth, and examples around these. Skipping them is the most common reason drafts feel generic.
  2. Constraints are non-negotiable. Numbers (“max 3 bullets”) and limits (“max 12 words per bullet”) get respected. Vague guidance (“keep it short”) does not.
  3. One task per turn. Asking for “outline + design + speaker notes” in a single prompt produces mediocre versions of each. Run them as three turns and each output is sharper.

Iterative refinement, not single-shot generation

Treat the first output as a draft to negotiate with , not a finished product. The pattern that consistently produces strong decks:

  1. Generate the outline only . Critique it. Ask for two alternative orderings.
  2. Pick an outline. Ask ChatGPT to expand slide-by-slide , one slide per turn for the most important slides.
  3. Once the content is locked, ask for format conversion (Marp, HTML, or python-pptx code).
  4. Finally, ask for speaker notes and a Q&A pack based on the finished deck.

Each step uses the previous step’s output as anchored context. The result feels coherent because the model has been steered, not just prompted.

Slide design rules ChatGPT will actually follow

ChatGPT is bad at “design well.” It is excellent at “follow these specific rules.” Translate design principles into rules and the model will apply them consistently.

The 6×6 rule and one-idea-per-slide

Add this clause to any deck-generation prompt:

✓Design rules to follow strictly:
- One core idea per slide. If a slide has two ideas, split it.
- Max 6 bullets per slide, max 6 words per bullet (the 6x6 rule).
- Every slide title is a full sentence stating the takeaway, not a topic label.
- For numerical claims, lead with the number ("$3.2M saved" not "Cost savings").

The “title as takeaway” rule alone transforms decks. Compare:

  • Topic label: “Q4 Sales Performance”
  • Takeaway title: “Q4 sales beat target by 18%, driven by enterprise renewals”

The second version means a reader skimming title-only already understands the deck.

Pyramid principle for executive decks

For audiences with no patience — executives, busy clients — instruct ChatGPT to follow Barbara Minto’s pyramid principle :

✓Structure the deck using the pyramid principle:
- Slide 1 states the conclusion and the recommended action.
- Slides 2-N support the conclusion with evidence, in MECE order.
- Each supporting slide opens with the claim, then the data.

This produces decks that work whether the executive reads slide 1 only or sits through the whole thing — both common scenarios.

Visual hierarchy hints

ChatGPT can’t draw, but it can describe images precisely for image-generation tools or stock searches. Add this:

✓For each slide, also output an "Image brief" line in this format:
Image brief: <subject>, <style>, <composition>, <mood>, <color palette>

You’ll get briefs like “Image brief: warehouse worker scanning inventory, photographic, eye-level shot, focused and competent, warm neutral palette.” Paste these into Midjourney, GPT-4o image generation, or a stock photo search and the visuals match the message.

Marp theming for branded, repeatable output

Marp ships with a few built-in themes, but the real power is custom theme CSS . Once you have a theme, every deck your team generates looks like it came from the same brand.

A minimal custom theme

Ask ChatGPT for a starter theme with your brand colors:

✓Write a Marp theme CSS file with these brand parameters:
- Primary: #0B3D91 (deep blue)
- Accent: #F5A623 (amber)
- Background: #FFFFFF
- Text: #1A1A1A
- Font: "Inter", system-ui, sans-serif
- Logo placeholder in bottom-right of every slide except the title.
- Title slide has a full-bleed primary-color background, white text.
- Section dividers have a left-border accent color stripe.

ChatGPT returns valid Marp CSS that drops into a file like theme.css . Reference it from any Marp Markdown deck:

✓---
marp: true
theme: theme
paginate: true
---

Layout directives worth memorizing

Marp’s directive syntax is small but powerful. The four you’ll use most:

DirectiveEffect
<!-- _class: lead -->Centers content; great for section dividers
<!-- _backgroundColor: #0B3D91 -->Per-slide background override
![bg right:40%](path.jpg)Image fills 40% of the slide on the right
<!-- _paginate: false -->Hide page number on this specific slide

Tell ChatGPT “use _class: lead for section breaks and bg right:40% for one image-led slide per section” and the deck gets visual rhythm without you touching CSS.

Versioning Marp decks with Git

Because Marp decks are plain Markdown, they version-control beautifully. A practical workflow:

  • One deck per branch while drafting.
  • Theme CSS lives in a shared decks/themes/ folder so every deck stays on-brand.
  • Reviewers comment on Markdown diffs in a pull request — far better than tracking .pptx changes.

This is the workflow that scales when slides become a recurring deliverable rather than a one-off.

Generate native .pptx files with python-pptx

When colleagues need to edit text directly in PowerPoint — sales reps tweaking pricing, executives rewording a line — Marp’s image-based PPTX export breaks down. The right tool is python-pptx , and ChatGPT writes it for you.

The Advanced Data Analysis route

On a paid ChatGPT plan, you can have ChatGPT both write and run Python in a sandbox. Ask:

✓Use python-pptx to generate a .pptx file with the following slides.
The output must be a fully editable PowerPoint file — every text element
must be in a real text frame, not an image.

Slides:
1. Title: "Q1 Operations Review"
   Subtitle: "<presenter name>, May 2026"
2. Section divider: "Headline metrics"
3. Three KPI cards (revenue, NPS, on-time delivery) with target vs actual.
4. ...

Use a 16:9 layout. Apply our brand color #0B3D91 for accents.
Run the code and give me the .pptx file to download.

ChatGPT writes the script, runs it in the sandbox, and hands back a download link. The resulting file opens in PowerPoint, Keynote, or Google Slides as a fully editable deck.

Reusable template scripts

The deeper play: build a template function that your team reuses for every deck. Ask ChatGPT to scaffold something like:

✓def make_deck(slides, theme, output_path):
    """
    slides: list of dicts with keys 'layout', 'title', 'body', 'image'
    theme: dict with 'primary_color', 'accent_color', 'font'
    """
    # ChatGPT fills in the python-pptx implementation

Save the script in your team’s shared drive. When someone needs a deck, they fill in the slides list (often by asking ChatGPT to generate it from a brief) and run the function. Output is consistent, on-brand, and immediately editable.

When to use each format

FormatBest forEditable by colleagues?
HTML + reveal.jsConference talks, web-shared decksRequires technical comfort
Marp MarkdownInternal decks, frequent updates, Git versioningAnyone who can edit text
python-pptx → .pptxClient-facing decks, hand-offs to non-technical editorsYes, in PowerPoint directly
Marp PPTX exportQuick PowerPoint conversion, view-onlyNo, slides are images

Pick the format based on who edits next, not on what feels familiar.

Templates for specific slide types

Some slides recur across nearly every business deck. Pre-built sub-prompts for these save real time.

The competitive landscape slide

✓Generate one slide titled "How we compare on the dimensions that matter."
Format: a 2x2 matrix.
- X-axis: <dimension 1>
- Y-axis: <dimension 2>
- Plot us, <competitor 1>, <competitor 2>, <competitor 3>.
- Output as a Marp slide using a markdown table; describe quadrant
  positions in a footnote line.

The roadmap slide

✓Generate one slide titled with the headline takeaway about our roadmap.
Format: a horizontal timeline with three phases (Now / Next / Later).
- Each phase: 2-3 specific deliverables, each with an owner.
- Output as Marp; use a 3-column layout with bold phase labels.

The “ask” slide for proposals

✓Generate the closing slide for a proposal deck.
Title must state the specific decision requested (e.g.,
"Approve $120K in Q3 to launch Phase 1").
Body: three bullets — what we'll do, when, what success looks like.
End with a single line: "Decision needed by <date>."

These templates aren’t precious; the point is to build your own library for the slide types you produce repeatedly. Five well-tuned sub-prompts will speed up nine out of ten decks you make this year.

Quality control: critique before you ship

ChatGPT can review its own output surprisingly well — if you ask in a specific way. After generating a deck, run this critique pass:

✓Critique this deck as if you were a senior partner at a strategy
consultancy reviewing junior work. Be direct.

For each slide:
- Does the title state the takeaway, or just label a topic?
- Is there exactly one core idea, or has it tried to do two things?
- Would a busy executive reading title-only understand the argument?
- Is any claim unsupported by data shown on the slide?

Then list the three slides that most need rework, in priority order.

Apply the critique to the top three slides and ship. This single pass catches the issues that make decks feel amateurish — vague titles, slides doing too much, claims without evidence — and it takes about two minutes.

Summary

The leap from “ChatGPT can make slides” to “ChatGPT makes our team’s decks better and faster” is mostly about reusable building blocks: a structured prompt template, a small set of design rules the model follows reliably, a Marp theme that enforces brand, and a python-pptx scaffold for hand-off-ready PowerPoint files.

None of these techniques are exotic. They take a couple of hours to set up once and pay back on every deck for the rest of the year. Start with one — the structured prompt template is the cheapest win — and add the next when you feel the friction of not having it.

The point of using AI for slides isn’t to skip the thinking. It’s to spend your thinking on the message , and let the model handle everything else.