My Updated Presentation on Artificial Intelligence and GenAI, Plus My First Thoughts on the Claude Add-In for PowerPoint, and Yet Another Head-to-Head Comparison Between Claude, Gemini, and ChatGPT – Ryan Schultz

I have used (and reported on) my use of generative AI tools extensively in this blogpost, but my writing is all my own, and I have not edited it or polished it using any GenAI tools.

I am (as you can tell by this absurdly long blogpost title) trying to do three related things here. If you want, you can skip to the very end, where there will be an executive summary, where I have some thoughts to share about all this.

First, I wanted to share an updated version of the original slide presentation on artificial intelligence and generative AI, which I shared in a December 2025 blogpost. I used to think that keeping track of the many metaverse platforms I blog about was a task similar to herding cats, but let me tell you, it was a breeze compared to trying to stay abreast of all the rapidly changing and accelerating developments in generative AI!

Below is my updated PowerPoint slide presentation, complete with my speaker notes, for you to download and use as you wish, with some stipulations. I am using the Creative Commons licence CC BY-NC-SA 4.0, which gives the following rights and restrictions):

Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International

This license requires that reusers give credit to the creator. It allows reusers to distribute, remix, adapt, and build upon the material in any medium or format, for noncommercial purposes only. If others modify or adapt the material, they must license the modified material under identical terms.

BY: Credit must be given to you, the creator.

NC: Only noncommercial use of your work is permitted. Noncommercial means not primarily intended for or directed towards commercial advantage or monetary compensation.

SA: Adaptations must be shared under the same terms.

(The tool I used to determine the appropriate Creative Commons licence can be found here: https://creativecommons.org/chooser/.)

So, with all that said, here is my PowerPoint presentation (please click on the text link or the black Download button under the picture, not the picture itself):

Second, today I installed a brand-new add-in from Anthropic’s Claude GenAI tool, which is supposed to work with Microsoft PowerPoint. This is an initial review.

And I have an actual real-world use case against which I will be trying out this new add-in: the design of an actual keynote presentation which I will be giving in a couple of weeks. (I am also using it in the third section, but in a different test of all three of ChatGPT, Claude, and Gemini.)

Now, before I get into this, I should explain that I have tried in the past with all three GenAI tools on which I currently have paid accounts (OpenAI’s ChatGPT, Anthropic’s Claude, and Google’s Gemini) to create a PowerPoint slide deck presentation design—only to get highly disappointing and completely unusable results back. So I was not expecting much here, particularly as this is a still a research beta version of the PowerPoint add-in.

My initial prompt to the Claude add-in to Microsoft PowerPoint was:

Please create a new PowerPoint slide presentation design with the title of the presentation being: “Your Metaverse Is Too Small: How the Biases and Preconceptions of Virtual Worlds Hinder Their Use in Education.” The theme of the talk is educational uses of virtual worlds, social VR, and the metaverse in general. I want to have some nice background images to use in some of my slides, as well as a visually pleasing title slide. I’d prefer blue as a colour in the slide deck theme, thanks!

And Claude chugged away on my request, keeping my posted on what it was doing:

And it even prompted me to be sure I wanted to delete the Claude add-in help slide!

The set-up for the title slide took a long, loooong time, much longer than I would taken to click on the Designer button in the PowerPoint toolbar and just select one of the default options, and a colour scheme. Eventually, I just gave up on waiting and went off to work on another task, leaving Claude to beaver away. After fifteen minutes, I realized that I still had to explicitly okay the clearing of the original slide design (inset Homer Simpson “D’oh!“), which I did, so that the work could continue.

If I could summarize the result in one word, it would be: meh (again, shout-out to The Simpsons):

I mean, I could easily do better than this myself. And two dots do not make, as I asked for, “some nice background images to use in some of my slides, as well as a visually pleasing title slide.” Here’s my section title slide:

Again, extremely underwhelming, and frankly, not an improvement at all over my previous failed attempts to generate a PowerPoint slide presentation design using any of the GenAI tools (Claude, ChatGPT, or Gemini). Mind you, I have deliberately stayed away from using the image-generation tools in these products; I can spot a GenAI-produced image from a mile away by this point, having been playing around with these tools, off and on, since they first came out in 2022.

Claude continued to generate all the standard versions of PowerPoint slides in this theme, ending with a final slide that, I must confess, I kind of liked the look of (although, again, I would have preferred some sort of background image):

This is where the process got interesting, as I finally decided to stop having to manually okay each individual step, and just gave Claude carte blanche to do whatever it felt was best. (I mean, the worst that could happen was that it come up with something I hated so much that I threw it away and started over.)

Claude was still working away while I took my lunch break, giving feedback along the lines of “Build stunning title slide design.” 🙄 (I’ll be the judge of what’s considered stunning, Claude. Calm the fuck down.)

Here’s the final result, my “stunning” title slide (insert RuPaul’s Drag Race shade death rattle):

So, yes, this is, once again, a complete and utter failure. I will probably still use this as a basic slide design, but obviously I will be locating and using my own images to illustrate it. This is now the second new tool in a week (first Claude Cowork and now Claude PowerPoint add-in) which has utterly failed at the tasks given it. I am not impressed.

Third, and finally, thank God, I had much better luck was in issuing all three general-purpose GenAI tools the exact same text prompt, a technique I had used before here (and one which I found very useful in comparing and contrasting the responses):

I am writing a keynote presentation on the mistakes companies make when creating, designing, and marketing the following product category: virtual worlds, social VR/AR, and metaverse platforms in general. Please give me a list of failed or shut down metaverse platforms, along with reasons why they might have failed. Please cite both academic and industry sources of information in your answer.

In all cases, I used the latest models as specified in Ethan Mollick’s latest AI Guide:

• ChatGPT’s GPT 5.2 Thinking with the Extended Thinking option;
• Claude Opus 4.2 Extended Thinking with the Research option; and
• Gemini 3 Thinking with the Deep Research option.

Unlike the last comparison, I’m not going to go into great detail on the results (because I will be using some of these results, once they are double-checked against more authoritative sources, in an actual keynote presentation I will be delivering later this month). Instead, I will my general overall impression of each report (and all three did provide a detailed report with citations).

Please note that I deliberately left it up to the specific GenAI tool to define what “failed” or “shut down” means, how far back and how thoroughly to search for failed platforms, and what metaverse platforms to include or exclude from its final report. As always, I find the differences between the reports to be an interesting way to compare and contrast the results, so below I will give some basic statistics:

While ChatGPT was the most thorough in listing failed metaverse platforms, and seems to have gone the furthest back in time (including There.com, which launched back in 2003!), it also had the fewest number of citations, and most of them were historical, platform-related announcements (e.g. a 2020 announcement of the shutdown of the then-social-VR platform High Fidelity by its CEO) rather than peer-reviewed academic journal articles (although there were a couple of those, too). While Claude had more citations, a review of those showed mostly blogs and news websites, with almost no references to actual academic research papers (probably because much of that content is locked behind academic publisher paywalls, although there were a few academic references to free sources such as ResearchGate and PubMed Central/PMC). Of the three, Gemini’s 33 citations used included the most resources which I would consider academic, from a good range of different publishers (as well as more informal websites). Interestingly, Gemini also included a list of resources which it looked at, but chose not to include in the final report, something which neither ChatGPT nor Claude offered! Gemini for the win here.

Gemini was also notable for the strong, overarching narrative structure to its report, something which I had also noticed in previous queries using this GenAI tool. Gemini is clearly very good at telling a cohesive story! However, Claude was also notable for listing, in a separate section of its report, what it called “cross-cutting failure themes” in its 13 examined metaverse failures (which is definitely a phrase I will be stealing for my final keynote presentation!). By comparison, the final report from ChatGPT, while thorough, was jargon-heavy, poorly-formatted, and seemed to lack the final polish of its competitors. For example, there were three separate sections titled “failure themes and comparative analysis,” “theme-to-platform mapping,” (?!??) and “top 10 failures by primary cause.” It was, by far, the worst of the three reports generated, just in terms of sheer (lack of) organization and narrative. Again, Gemini for the win!

EXECUTIVE SUMMARY: So, here are my final thoughts.

• It is getting harder and harder (in fact, almost a full-time job) to keep on top of what is fast becoming an arms race between the top three general-purpose generative AI tools (ChatGPT, Claude, Gemini), not to mention an ever-growing legion of more narrowly-focused applications, which might be better at certain specific tasks, such as writing programming code or generating music.
• While Claude seems to be good at putting new agentic (e.g. Claude Cowork) and add-in tools (Claude for PowerPoint) into the hands of its users first, my personal experience with these new tools has been very disappointing, even comically bad. However, Claude’s chatbot interface works well for generating detailed answers with citations (although slightly edged out by Gemini).
• I am impressed by Gemini’s consistent ability to create a strong narrative structure within its generated reports, something in which ChatGPT in particular is noticeably lacking. It also came first in a key metric: actual citations to academic literature, not just freely-accessible websites (blogs and news articles).
• If I were forced to rank the three GenAI tools by just this one head-to-head-to-head comparison (i.e. the third part of my blogpost), I would rank them as follows:
  • 1st: Google Gemini.
  • 2nd: Anthropic Claude.
  • 3rd. OpenAI ChatGPT.

Related Blogposts