An AI Workflow for Creating Presentations

A reliable AI presentation workflow moves through five stages in order: outline the structure first, draft slide content one section at a time, write speaker notes separately from slide text, add visual guidance, then review and export. Skipping straight to "make me slides about X" is what produces a wall-of-text deck that needs a full rebuild.

Picture it's 9 p.m. and tomorrow's lesson still needs a slide deck — a title slide, five content slides, and a wrap-up. Trying to generate all of that in one request tends to produce a rough draft that reads more like an essay chopped into pieces than a set of slides built to be presented.

Quick Answer: Build a presentation with AI in five stages: outline the slide count and structure first, draft each slide's content separately, write speaker notes as their own request, add visual/layout guidance, then review before exporting. Treating slide generation as one big request is the most common reason a deck comes back needing a total rewrite.

What an AI-Assisted Presentation Workflow Actually Looks Like

A presentation isn't one document — it's several different types of content stacked together: an outline, slide text, speaker notes, and visual layout. Each of those benefits from its own focused prompt rather than one request trying to do all four at once.

This staged approach is the same underlying principle covered in AI Prompting & Content Workflows for Teachers (2026 Guide) — break a large content type into its real sub-parts, then prompt for each one on its own.

Stage 1: Outline Before You Generate a Single Slide

An outline decides slide count, section order, and each slide's one job — decisions that are much cheaper to fix at the outline stage than after ten slides already exist.

A Working Outline Prompt

• "You are a Grade 5 science teacher. Outline an 8-slide presentation on the water cycle: a title slide, 5 content slides (one concept per slide), a summary slide, and a discussion-question slide. List only the slide titles and one-sentence purpose for each, not full content yet."

Why Slide Count Belongs in the First Prompt

Stating slide count upfront prevents a common failure mode: a model that either crams too many ideas onto too few slides, or pads a simple topic out to twice the length it needs. Fixing an outline's structure takes one sentence; fixing ten already-drafted slides takes much longer.

Reviewing the Outline Before Moving On

A quick scan of the outline for logical order and one-idea-per-slide framing catches structural problems while they're still cheap to fix. Reordering two outline bullet points is a much smaller edit than reordering two fully drafted slides.

Stage 2: Drafting Slide Content One Section at a Time

Once an outline exists, draft slide text in small batches — two or three slides per prompt — rather than requesting the entire deck's content at once. Smaller batches are easier to review and far easier to revise without losing the parts that already worked.

The One-Idea-Per-Slide Principle

Cognitive-load research associated with Richard Mayer's multimedia learning principles has long argued that overloading a single slide with too much text or too many ideas at once makes it harder for an audience to process, not easier. A slide-content prompt should explicitly ask for this constraint, not assume the model will apply it on its own.

A Content-Slide Prompt Template

• "Draft the text for slides 2-4 of this outline. Each slide gets a title and 3-5 short bullet points, no full sentences, no more than 8 words per bullet. Flag if any single slide seems to be covering more than one main idea."

That last instruction — flagging overloaded slides — catches a real, common failure: a "concept" from the outline that actually needed two slides, not one.

Stage 3: Speaker Notes and Talking Points

Speaker notes and slide text serve different purposes, and asking for both in the same request tends to blur them into one another. Slide text is what the audience reads; notes are what the presenter says.

Why Notes Need a Separate Prompt From Slide Text

A single combined request often produces slide bullets that are really just shortened notes, or notes that just repeat the bullets word for word. Requesting notes as a distinct, second pass — after the slide text is finalized — keeps each doing its own job.

A Speaker-Notes Prompt Template

• "Write speaker notes for these 5 slides. Each note should expand on the slide's bullets in full sentences, roughly 3-4 sentences per slide, written in a conversational tone appropriate for explaining this to a Grade 5 class."

Notes written this way give a substitute teacher, or the same teacher a year later, enough context to deliver the lesson without having to reconstruct what each terse bullet point was supposed to mean.

Stage 4: Visual and Design Guidance

An AI tool can describe what a slide's visual should show even when it isn't producing the final graphic itself, and that description is worth requesting explicitly.

• "For the slide on evaporation, suggest what a supporting diagram should show — labeled arrows for the process steps, not a decorative stock image."
• "Note where a slide would benefit from a simple table instead of bullet points, based on this content."

Why "Suggest, Don't Assume" Matters Here

Slide text alone doesn't communicate layout. A prompt that explicitly separates "what to write" from "what a visual should show" produces guidance a teacher can act on when assembling the final deck, rather than a vague implication buried inside the bullet text itself.

Walking Through a Full Deck: A Worked Example

Seeing all five stages applied to one real 8-slide deck makes the workflow concrete rather than abstract. Say the topic is the water cycle for a Grade 5 science class, due tomorrow morning.

1. Generate the outline first — 8 slides, one-sentence purpose each, reviewed for logical order before moving on.
2. Draft slides 1-3 (title, evaporation, condensation) in one batch, checking for one-idea-per-slide framing.
3. Draft slides 4-6 (precipitation, collection, a summary diagram) in a second batch, using the same bullet-style instruction.
4. Draft the closing discussion-question slide separately, since its format — an open question, not bullet points — is genuinely different from the rest.
5. Request speaker notes for all 8 slides in one follow-up pass, now that the slide text itself is finalized.
6. Ask for visual guidance slide by slide, noting where a diagram or table would replace a wall of bullet points.
7. Do a full read-through, checking science accuracy, slide count, and pacing before the file goes anywhere near a projector.

That sequence takes longer than one giant "make me a water cycle presentation" request. The payoff shows up in revision time: a five-stage draft built this way usually needs small, targeted fixes rather than a full regeneration, and the same seven-step shape works for almost any single-topic lesson, not just this one.

Adjusting the Workflow by Presentation Purpose

Not every classroom presentation serves the same purpose, and the five-stage workflow bends slightly depending on who the deck is actually for.

Building a Student-Facing Review Deck

A deck meant for students to click through on their own needs a different content-slide prompt than a lecture deck: "Draft slides 1-5 as a self-paced review deck. Each slide should pose a short question or prompt a student can answer before revealing the next slide, rather than presenting a full explanation." Speaker notes are largely unnecessary here, since no one is presenting live.

The Same Five Stages, a Different Subject

The outline-then-draft structure holds regardless of subject — a world-language class needs the same staged approach, just with vocabulary-heavy slide content instead of a science concept. How to Write AI Prompts for Spanish covers how slide and worksheet prompts shift when the content itself is vocabulary and conjugation practice rather than a science process.

Building a Reusable Slide-Outline Library

A handful of saved outline shapes, reused across many topics, is what makes this workflow faster the second time and every time after that. Without one, each new topic starts the outline stage completely from scratch.

• Concept-explanation shape: title, 4-6 concept slides (one idea each), a summary slide, a discussion-question slide — reusable for most single-topic science or social studies lessons.
• Compare-and-contrast shape: title, one slide per item being compared, a side-by-side summary slide — reusable whenever a lesson sets two things against each other.
• Process/sequence shape: title, one slide per step in order, a full-sequence recap slide — reusable for anything with ordered steps, from the water cycle to a writing process.

Discussion Slides Feed Into a Larger Question Bank

A deck's closing discussion-question slide doesn't have to stand alone. Saving those questions alongside a larger item bank — see How to Generate 50 Quiz Questions in 5 Minutes With AI for building that bank at scale — turns a handful of one-off discussion prompts into a reusable review resource by the end of a unit.

Stage 5: Review, Export, and Reuse Next Year

A finished draft still needs a human read-through, and the right export format saves a real reformatting step afterward.

EduGenius can generate slide content aligned to a saved class profile and export directly to PPTX, which is designed to skip the step of manually rebuilding a generated outline inside separate slide software. That same saved profile can carry over the next time a similar lesson topic comes up.

A saved, working outline is worth keeping past this one lesson — the same three-slide "concept" structure often reuses well the next time a related topic comes up, the same way a How to Batch-Generate Flashcards With AI term list reuses across a unit's review sessions.

Accessibility Considerations for AI-Generated Slides

A generated deck that reads well on screen doesn't automatically work for every student, and a couple of prompt additions close most of the gap.

• Request alt-text descriptions for any suggested image or diagram: "Include a one-sentence alt-text description for the diagram suggested on this slide."
• Ask for plain, direct language on any slide meant to be read independently, since dense academic phrasing is harder to process for students working in a second language or with a reading disability.
• Avoid relying on color alone to convey meaning in a suggested chart or diagram — ask the model to note this explicitly when it proposes a visual.

None of this replaces a school's own accessibility review process. It does mean the first draft arrives closer to usable, rather than needing every slide retrofitted for accessibility after the fact.

UNESCO's guidance on inclusive digital learning materials makes a related point about instructional content generally: accessibility considered from the start of a workflow is consistently less disruptive than a separate remediation pass applied to a finished document.

References

• Richard Mayer — cognitive-load and multimedia-learning research on how audiences process slide content.
• ISTE — guidance on effective use of AI-generated instructional materials in the classroom.