Creating carousel posts for platforms like LinkedIn and Instagram is often a tedious process involving endless back-and-forth, reworking, and fine-tuning. But can AI streamline the workflow?
To find out, we put four different AI tools to the test: ChatGPT, Gemini, Claude, and Copilot. Using a standard carousel dimension and a pre-designed two-slide brand template in PowerPoint, we challenged these tools to build out the remaining slides based on draft content.
Here’s how they stacked up.
Copilot Played It Safe
Copilot built out the rest of the deck without a hitch. It matched our layout, matched our look, and stayed on brand. That’s also all it did. Text sizing shifted from slide to slide with no clear logic, and nothing was highlighted or emphasized.
The result was safe, not smart.
ChatGPT Made One Smart Choice
Same prompt, same setup. ChatGPT took about five minutes and made one improvement Copilot didn’t think of: it changed the color of the “link in comments” text so it would get noticed.
Beyond that, the differences were mostly interpretation, not innovation. It formatted the same content differently than Copilot did, with some text ending up too small to read comfortably.
Claude Asked Before It Built
Claude didn’t just start generating. It asked more questions before touching a slide, and the difference showed up fast. It kept small text consistently sized across the whole deck, colorized the lines it judged most important instead of treating every word the same, and added its own color-treatment slides at the end without being asked.
It got a little busy in places and needed some editing for accuracy, but of the three PowerPoint-native tools, this was the one closest to something like a final product.
Gemini Had the Right Ideas but the Wrong Process
Gemini refused to build the whole deck from one prompt. It would only generate one slide at a time, meaning we had to manually recreate the same prompt for every slide in the deck.
The instincts underneath that workflow were sound: it broke content into bullet points, and like Claude, made a real decision to color specific text instead of defaulting everything to one color.
However, we didn’t finish the full deck. The one-slide-at-a-time process wasn’t worth the time it demanded.
Claude Design Actually Understood Design
Claude Design lets you upload your brand guidelines and past work, and it builds a design system from that instead of just filling in a template. We’d already set up a custom skill in Claude for carousel posts, so we gave it our content and let it design freely, which meant what came back looked nothing like our PowerPoint version.
The result ran more colorful than we would have chosen, arguably too colorful, and it added visual step elements we never asked for. Our logo didn’t load in correctly either. Still, it understood layout in a way none of the other tools did, formatting one section like an equation and instinctively treating the “link in comments” line like a button instead of just text.
Editing inside Claude Design feels closer to a web builder than a marketing tool. Text edits were simple, but color edits took more patience than expected.
Exporting Claude Design Work for Your Existing Tools
To move a Claude Design file into another tool, export as PowerPoint, not PDF, and skip the Canva connector, which hasn’t worked reliably for us. Choose custom fonts rather than universal fonts, so the right typeface carries over instead of defaulting to Arial.
From there, Canva imported cleanly and kept the layout logic intact, though some background elements didn’t survive the trip. Adobe Express required more manual font fixing, even with the right font already loaded there.
So, is AI good at design?
I’ll let you decide. Copilot and ChatGPT get you safe, on-brand slides fast, with zero creative risk. Claude, inside PowerPoint, made the most thoughtful calls on hierarchy and emphasis. Gemini has sharp instincts trapped in a workflow that won’t scale. And Claude Design is the real shift, if your team already works inside Claude and has brand guidelines set up. It understands layout the way a designer does, not the way a template does. It just needs a human pass on color and logo placement before it ships.
None of these tools hand you a finished carousel post. Most of them hand you a real starting point. But for a process that used to eat an afternoon, that’s not nothing.
