The AI Slide Deck Workflow: Why Your Decks All Look the Same and How to Fix It
Most AI-generated slide decks share a problem: they look like every other AI-generated slide deck. The usual response is to try a different tool. A better response is to notice that building a deck is two separate jobs, and that no single tool is strong at both.
This post covers a workflow that splits the job in two: a four-column outline document that holds your instructional design decisions, and PowerPoint's Copilot rendering that outline into a finished deck. It comes from my own testing of Gamma, Claude, and Copilot, and it includes the licence and file requirements that tripped me up.
Why AI Slide Decks Look the Same
Purpose-built deck generators are fast, but they are opinionated. Gamma, in my testing, produced an image on one side and text on the other, slide after slide. It is one template, so every deck reads as a Gamma deck rather than yours. Claude sorted information sensibly, pulled out key points instead of dumping source material onto the slide, and used icons well, but it cannot generate images, so the result had no background visuals at all.
The pattern is not unique to my setup. A 2Slides analysis of AI deck design lists the same tells: uniform bullet structures, repeated stock icons, identical title formatting, and safe mid-palette colours. Its conclusion is that the difference between an AI deck and a human deck is the time spent replacing defaults with specifics.
That is a useful observation, but it treats the symptom. The cause is that one tool is being asked to make decisions about content and about visuals at the same moment.
The Job Is Actually Two Jobs
The thinking is the instructional design work: what goes on each slide, in what order, how much text belongs there, and what the instructor says that never appears on screen. The making is production: layout, images, colour, and brand.
Every tool I tested was strong at one of these and weak at the other. Handing the whole job to a single tool meant accepting either its template or an empty design. The fix is to do the thinking in a general-purpose tool where you control it, capture that thinking in a document, and hand the document to a tool that renders.
The Four-Column Outline
Ask an AI to draft the deck as a table, not as a presentation. Use four columns: slide number, the text that appears on the slide, an image prompt describing a background image specific to that slide, and speaker notes with the detail the instructor needs.
Each column does a job. The slide text column keeps content at slide length because you are writing into a cell, not a page. The image prompt column breaks the sameness, since every slide carries a description of its own visual. The speaker notes column catches everything that would otherwise migrate onto the slide.
Here is one row from an outline for a report writing session. The slide text is "Three things every use of force narrative must establish." The image prompt is "A desk at the end of a shift, a report half written on a monitor, a notebook open beside it, low evening light, muted tones, no readable text." The speaker notes hold the three elements, the one most often omitted, and a question to ask the room before revealing the third.
Notice that the notes column is already most of a facilitator guide entry. A workflow aimed at better slides produces better documentation as a side effect.
Why Text-Heavy Slides Hurt Learning
Keeping text off the slide is not only an aesthetic preference. An AhaSlides review of presentation research describes Richard Mayer's redundancy effect: when a presenter reads text the audience is also reading on screen, comprehension drops compared with either the spoken word or the text alone.
A slide that can be read aloud as a coherent paragraph should be cut, and its content moved to the notes. Neither Gamma nor Copilot has any concept of a room full of people trying to read and listen at once, so this judgment stays with you. The four-column structure makes it routine, because the notes column gives the content somewhere else to live.
Building the Deck With Copilot in PowerPoint
Save the outline document to OneDrive. This step is load-bearing: Copilot in PowerPoint can only build from a file it can see, and a file on your desktop will not appear in the picker. Then open a new blank presentation, click the Copilot icon, and choose to create a presentation from a file. Add style direction in the prompt: palette, tone, and formality.
When I ran my own newsletter through this workflow, the result was twelve slides with twelve different images, each applied as the slide background rather than dropped into a picture box. The slide text matched my outline word for word, and the speaker notes came through unchanged. Copilot honoured every column and invented nothing.
Licensing is where documentation and reality diverge. I use a personal Microsoft 365 plan with Copilot Pro, and the workflow worked. Tom's Guide describes the same requirements: a Word file in OneDrive and Copilot Pro. A Microsoft Tech Community thread shows availability is inconsistent across builds, with some users missing the file picker and pasting a OneDrive link into the Copilot sidebar as a workaround. Test it on your own licence before believing either the documentation or the forum posts.
Also keep in mind that the standalone Copilot app and the Copilot inside PowerPoint are different surfaces. On my setup, the desktop app could not produce a PowerPoint file at all.
What the Deck Still Needs From You
The output is a draft. Check it against your learning objectives, because neither tool knows what the session is supposed to accomplish. Expect a few slides where the layout fought the content. Rewrite the speaker notes in your voice, since what comes back is information rather than facilitation. For single-slide edits, open that slide and prompt the Copilot panel directly. It does not get everything right, but changing an image is something it handles well.
The larger risk is speed. A well-produced deck on a weak structure is still weak, only more convincing. The ATD research on AI use among instructional designers found that 37 percent of talent development professionals say AI has greatly reduced course design time, and that outlining, storyboarding, and writing objectives are among the main uses. Faster production makes the upstream work more important: define the performance gap, write measurable objectives, then build.
If your agency does not run Copilot, keep the outline step and change the production step. The outline is useful whatever builds the deck. I have not tested those alternatives as substitutes, so treat them as directions to explore.
Want to Build These Skills With Your Team?
If you want your training team to start applying structured AI workflows to slide and course design, I offer private 4-hour virtual workshops designed specifically for training developers. We work through your department's actual projects using multiple AI platforms, so participants leave with practical skills and working materials, not just theory.
Format: Private virtual sessions for up to 20 participants
Investment: $2,000 USD / $2,500 CDN per workshop
Contact me at kerry.avery@shaw.ca or visit the workshops page on this site.
Sources
A Note on AI Use
This post was researched and drafted with AI assistance, then reviewed and edited for accuracy and voice. All practical recommendations reflect my own instructional design experience.
Comments