Why ChatGPT is crushing Microsoft Copilot

Microsoft’s Copilot AI assistant has a fraction of the downloads OpenAI’s ChatGPT does on mobile devices, a pretty clear indication it’s losing to ChatGPT on Windows, too. In fact, Copilot is often ignored entirely in comparisons of the most widely used AI chatbots, where Anthropic’s Claude and Google’s Gemini get much more attention as true ChatGPT competitors.

I sat a few feet away from Microsoft CEO Satya Nadella when the company first unveiled its grand Copilot strategy in 2023, including its plan to add Copilot to all Windows PCs. The company was flying high, and the Microsoft employees in the room seemed excited about their burgeoning generative AI (genAI) plans. But even with the most popular desktop operating system in the world as a platform, that Copilot push probably hasn’t gone as well as Microsoft expected.

It’s shocking, given that Microsoft was an early investor in OpenAI. With its early Bing Chat launch, Microsoft pulled ahead of both OpenAI and Google, offering the first chatbot that could search the web. Microsoft even had early access to a more advanced GPT model than the one you could use in ChatGPT at the time.

Then everything changed.

Copilot vs. ChatGPT: The current reality

A recent Bloomberg story offers just the latest indication of what we already know: ChatGPT is far ahead of Copilot in app downloads; even DeepSeek — despite data security concerns around that service sending data to Chinese servers — is now beating Copilot in mobile app downloads. Those downloads arguably mirror what people are doing on their PCs, too, as most users rely on the same chatbot across devices.

We’ve heard a lot lately about the difficulty Microsoft has had in encouraging companies to adopt Copilot when their employees already prefer ChatGPT. (Nadella earlier this year told Microsoft employees that the reception to DeepSeek R1’s launch set a new bar for AI success.)

Plain and simple, Copilot isn’t moving the needle in the way Microsoft would prefer.

The Bing Chat mirage

Microsoft’s initial investment in OpenAI back in 2019 totaled $1 billion. The company has now invested $13 billion in the organization.

That investment alone has given it some unusual opportunities m— and early on, Microsoft embraced them. OpenAI released ChatGPT in November 2022; Microsoft launched Bing Chat in February 2023, just a few months later.

Bing Chat was the high-water mark for Microsoft’s genAI ambitions. While the name was bad — “Bing” doesn’t have positive cultural associations — this was a unique product. In a lot of ways, Bing Chat was better than ChatGPT: it had integrated web search, a feature ChatGPT didn’t have at the time. It was also powered by an early version of GPT-4, a more advanced large language model that you couldn’t use in ChatGPT at the time.

When Microsoft released Bing Chat, Google hadn’t even released its own chatbot yet. Google’s Bard, which later evolved into Gemini, wouldn’t show up until March 2023.

In spite of all of that, Bing Chat was actually an incredible PR disaster at launch. It seemed as if Microsoft didn’t do enough testing, and the “Sydney” personality that Microsoft created for its chatbot consequently did strange things in extended conversations. Dialogues like this one in The New York Times showed Bing Chat saying “I want to be alive” and making other such statements that Microsoft probably didn’t want printed in the media.

Microsoft responded by capping messages in Bing Chat threads — initially letting you only send five messages to Bing Chat before you had to start a new thread — which made the chatbot much less useful than ChatGPT for many purposes. Strike one.

As Microsoft made Bing Chat safer, the shift away from the chaotic Sydney personality was also a bit of a loss. Bing Chat got negative PR, yes — but it also got serious viral attention. Personally, the conversations I had with Bing Chat were the first time I found an AI chatbot somewhat entertaining. When Microsoft scrambled to make Bing Chat more buttoned up and professional, it lost that viral angle.

There was likely a middle ground where Bing Chat could have remained intriguing if you clicked through a disclaimer, while making the default chatbot experience less crazy for the average AI-assisted web search. Microsoft didn’t take it. Strike two.

The Copilot branding disaster

All launch-related hurdles aside, the name “Copilot” in and of itself feels very focus-grouped, while the name “ChatGPT” seems technical and unprofessional. If you and I were branding consultants sitting around a conference room, we’d say Copilot as a name held a big advantage.

But, in the real world, Copilot is a messy moniker. Microsoft’s Copilot for consumers, Microsoft 365 Copilot, Copilot+ PCs, and Copilot Pro subscriptions are all different products. They’re an absolute maze, and I won’t even attempt to sort it all out here because it would take too many words.

In comparison, ChatGPT is a known entity. It’s a single interface both for business and consumer use. When OpenAI offers other products, like a video generator, it names them things like “Sora.” It doesn’t re-use the ChatGPT name for many different products, as Microsoft does with Copilot.

Strike three.

Microsoft’s Copilot ‘friend’ obsession

Copilot is in a strange spot, and at this point, even I don’t fully understand what Microsoft wants it to be. On one hand, Microsoft would like Copilot to be your new AI friend. That was the result of one of the Copilot app’s many redesigns over the past few years

Bloomberg writes that Microsoft executives “see Copilot’s emphasis on being a personable companion as a potential advantage with a younger cohort that tends to use AI tools as sounding boards.” That’s a strange statement when ChatGPT is now ahead among users in that same age range!

But while Microsoft superficially wants Copilot to offer a more personal companion, Copilot’s standard prompts seem to result in a more buttoned-down, corporate tone. That’s my experience, of course, but the usage numbers back me up and show that many people are drawn to other assistants. In vibe terms: The vibe is off.

Yet another strike. Is anyone even still counting?

The AI enthusiast angle

For more professional use, Copilot is still behind ChatGPT in the experience it offers. ChatGPT and Gemini both provide model transparency: You can go into the model switcher and choose exactly what model you want to use for a particular prompt. With Copilot, Microsoft leaves the LLM a black box. You send a message and Microsoft has a model router that sends your input to whichever LLM Microsoft’s tools think are best. (And it could just be the model that will save Microsoft money by being cheaper to run — you don’t know.) You don’t have control, which also makes it tougher to troubleshoot why something isn’t working as expected.

Also, while Microsoft used to offer early access to OpenAI’s technological advancements — remember how Bing Chat had GPT-4 before ChatGPT — Copilot now seems noticeably behind. New ChatGPT features are often added to Copilot months later. ChatGPT’s Windows app is ahead of Copilot in a lot of ways, and that’s clearly not the perception Microsoft needs to win folks over.

If you’re interested in genAI, you’re likely going to choose a more advanced, customizable tool — especially if you’re ready to spend money. With both Copilot Pro and ChatGPT Plus costing $20 per month, you probably aren’t going to spend money on both — and ChatGPT Plus is the logical choice unless you want AI features built into Office apps like Word, Excel, PowerPoint, and Outlook. 

That and that alone remains Copilot Pro’s killer feature.

Microsoft’s Copilot challenge

All considered, then, sure: Copilot is…fine. The underlying LLMs are as good as OpenAI’s, and Copilot Pro can be quite helpful in Microsoft Office. I’ve been experimenting with using it for tedious spreadsheet tasks like analyzing data, writing formulas, applying conditional formatting, and so on. It’s great. Copilot Vision works well, too, and is an exclusive feature.

But, overall, Microsoft’s strategy of implementing OpenAI’s technologies in Copilot means the company doesn’t really stand out when competitors like OpenAI, Google, Anthropic, and even Meta are delivering their own models.

The company does have a few bright spots. The recently launched Copilot Vision feature, for instance — it lets you talk about a Windows app with the genAI tool in real time — is something ChatGPT doesn’t offer on Windows.

Additionally, that Office integration is truly special — but it’s tough to say how much that matters in the long term. Between OpenAI’s interest in delivering an office suite, MCP servers that will let your chatbot of choice interact with your apps, and the way every company seems to be integrating AI features into their services, AI built into Office might be nice for people already using Office, but not something that lets Microsoft stand out in the larger market. That’s especially true given pricing: Microsoft is charging organizations $30 per user for Copilot on top of their Microsoft 365 subscription fees. Meanwhile, Google is offering Gemini features to all paid Workspace users.

In the long term, Microsoft will likely need to break with OpenAI and develop its own AI models to power its software. That seems like it’s already happening, and the relationship between OpenAI and Microsoft is reportedly not good.

A bizarre break with Microsoft’s Windows past

Bloomberg explains Microsoft’s push to turn Copilot into a “friend” that’s less focused on traditional chatbot workflows. Mustafa Suleyman, the CEO of Microsoft AI, reportedly believes that “people will end up using distinct AI tools depending on whether they’re at work or home.”

That’s a surprising statement considering Microsoft’s history. One reason PCs won is because you could use the same Windows operating system and software at work and at home. The same operating system that ran Microsoft Office also ran PC games and computer software. Your computer-use knowledge transferred between your home and work PC; you didn’t have to learn two different things.

In an age where more people work from home and work and personal lives are becoming more porous, it makes even less sense to use a different genAI tool at home than at work. I think Microsoft is wrong: People want to use the sameAI tools in both places.

The real question now is how much time is left — if any — for a proper course correction.