Watching the development of AI chatbot features can feel a lot like listening to “Anything You Can Do” from Annie Get Your Gun. You know the song, “Anything you can do, I can do better. I can do anything better than you.”
For instance, when OpenAI debuted the Canvas feature for ChatGPT, it boasted that it would make for a more flexible, collaborative way to create and edit text and code. When Google unveiled a feature with the exact same name, you can guess how they marketed it.
Gemini’s approach to an interactive workspace where users can draft, refine, and tweak content does have the distinction of being available for free, while you’ll need to pay $20 a month for ChatGPT Plus to use Canvas with OpenAI’s chatbot.
Both claim to make creativity easier, but I wanted to see how they handled some real-world instances where those abilities would come in handy. One of the main elements of both versions of Canvas is that you can edit and adjust specific bits of text and code, so theoretically, I could have iterated forever. I kept things relatively simple for these experiments, lest I spend forever chasing down that perfect phrase or code snippet.
Calendar
I started with a common tool that needs extensive editing: a schedule. I went with a generic request for a basic weekly schedule. In their respective Canvas forms, I asked ChatGPT and Gemini to “Create a weekly schedule for me. I need to balance work (9 am – 5 pm, Monday to Friday), exercise, meal preparation, and personal time. I prefer to exercise in the mornings and dedicate evenings to personal projects.”
I then ran them through a series of edits, asking them to change the exercise time, add a writing session, expand the meal prep time, and add a dinner with friends to Friday. Both performed admirably in that regard.
Finally, I asked them to “display this schedule as a visual calendar with different colors for each activity type.” This is where there was some differentiation.
While Gemini turned the calendar into a chart, it only suggested which colors to apply to which schedule bits. ChatGPT, meanwhile, formed a long list of events with an extra column for them to have a color with a small circle of that color. Neither really did what I wanted, but they did at least understand what I wanted in theory.
Learn to talk
I next decided to see how the two Canvases would handle enhancing a helpful guide. First, I asked, “How can I encourage my one-year-old son to start talking? Give me practical tips that don’t feel like a parenting manual.”
Both AIs put together a nice range of suggestions for activities and practices to help teach my son new words. The real test came when I asked them both to add visuals to their guides.
ChatGPT simply wrote text suggestions for images to go with different parts of the guide. Gemini really stood out, however, with not only actual photos showing some of the activities but also images that were linked to videos on TikTok and other sites demonstrating what the guide described.
History AI
My wife is an amazing cook, and I’ve often thought about putting together a small website where people could find some of her best recipes. I figured testing Canvas was a great opportunity to see how the AI tools handled web development.
I asked them both to: “Create a simple HTML page to showcase my wife’s recipes, including sections for appetizers, main courses, and desserts. Keep it clean and visually appealing.”
Both quickly wrote some code, and both Canvas versions have an option to preview what that code would look like. You can see ChatGPT’s below on the left and Gemini’s on the right.
Gemini had a decidedly more creative approach, even with the bare minimum of information in the prompt. With some more detail and creative ideas, the HTML from either could be adjusted, but when it comes to the initial prompt, Gemini clearly had more of a flair for web design.
That seemed to be the real takeaway from the tests. Both forms of Canvas were great, with some iterating, but Gemini seemed to think more visually than ChatGPT. In terms of text, I found ChatGPT to generally be better, as is the case even without using Canvas.
So, for myself, ChatGPT is where I’m likely to start, but if I really want to add some multimodal enhancement, I’ll ask Gemini to lend its visual taste to the answer.