There is a very comfortable story doing the rounds right now: Microsoft Copilot is a flop. Bad product, people abandon it, ChatGPT wins. The data even seems to back it up. And yet we think that story is reading the scoreboard of a game Microsoft stopped playing months ago.
Let us start by conceding the obvious, because it is true.
Yes, on product, Copilot is losing
When people are free to choose, they don't choose Copilot. The independent tracking from Recon Analytics tells it without mercy: among U.S. paid AI subscribers, Copilot dropped from 18.8% to 11.5% in six months, while ChatGPT holds above 55% and Gemini already overtook it. Put a worker in front of both Copilot and ChatGPT and roughly three out of four go to ChatGPT. And of those given access to Copilot, independent estimates put regular use at barely a third.
The reason isn't mysterious, and it isn't the model. Copilot isn't a model at all, it is a layer: underneath it can run the same engines as ChatGPT, or Claude, or now Microsoft's own MAI models, launched at Build in June. The user almost never knows which brain answered, and that is precisely the tell. The problem is the experience: you ask it for something, the answer lands short of what you needed, and you quietly think "faster if I just do it myself." Open Copilot inside Excel, then open the ChatGPT or Claude extension in that same sheet, and the gap is not a nuance, it is a chasm.
There is even a structural reason we have some respect for naming honestly. Microsoft was among the first, and by far the most aggressive, in betting its whole office suite on AI, and that took nerve. But it bolted that AI onto fifteen years of its own plumbing, SharePoint and Microsoft Graph and grounding layers that chew on every query before the model sees it. Hanging a modern agent off that is like dropping a Ferrari engine onto a dirt track. The engine is fine. The road is the problem.
So if the game were "best product", this would be over. It isn't over. And that is the part worth thinking about.
Microsoft was never playing the product game
Here is our actual opinion, and it is not the popular one: Microsoft knows Copilot isn't the favourite, and it has decided it doesn't need to be.
Microsoft has never won on product. It didn't win the browser because it was better, or the office suite because it was better. It won because it was already there, on the machine, in the contract, in the budget that finance had already approved. That is the only game it has ever truly mastered, and it is the game it is playing again now: not "have the best AI", but "be the AI that is already inside".
It barely hides it any more. To ship its newest Copilot features, Microsoft leans on OpenAI and Anthropic at the same time, and has now added its own MAI models to the mix, betting openly that its advantage is not the model but the data: your data, your tenant, your Office. That is the whole strategy in one sentence, and Microsoft's own moves say it plainly. The model underneath is interchangeable. The lock-in is not.
And on that board, the numbers flip from embarrassing to formidable. Microsoft says more than 90% of the Fortune 500 already use Microsoft 365 Copilot in some form, and Microsoft 365 carries over 400 million paid seats. No competitor on earth has a starting line like that. When a leadership committee sits down to "decide on AI", it almost never picks the best tool on the market. It picks the one that is already a button in the corner of the Excel everyone uses: no new vendor to vet, no extra security review, no workflow to rebuild. "It's already integrated, and it's Microsoft" beats any benchmark in a boardroom, every time.
The product reviews are written by power users. The cheques are signed by procurement. Microsoft figured out long ago which of those two it actually needs.
Why that is the uncomfortable part
It would be easier if Copilot were simply bad and that were the end of it. It isn't, and that is exactly what makes it dangerous.
The people who live inside AI every day, the ones who set the direction for everyone else, have already walked. They're on ChatGPT, on Claude, on whatever is sharpest this week. But direction and lock-in are not the same thing. Microsoft is betting it can chain enterprises through integration faster than rivals can win them over through quality. If that bet pays off, "the best tool" never even gets a seat at the table, because the decision was made the day the company renewed its Microsoft licence.
There is a real risk for Microsoft in all this, to be fair: confusion. Copilot, Copilot Chat, Copilot Studio, Agent Mode, Agent 365, Work IQ, a blur of agents and add-ons announced by the dozen. When a vendor ships hundreds of features and nobody can explain what half of them do, "already integrated" starts to wobble. Distribution buys you the door. It does not, on its own, buy you trust.
How we read it at GROS
At GROS we work with AI models every day to produce images and video, and this is precisely why we refuse to marry one. The lesson in the Copilot story is not "Microsoft good" or "Microsoft bad". It is that the tool that ends up running inside a company is rarely the best one; it is the one that arrived first and made leaving expensive.
So our position is simple. Don't choose the AI that comes bundled. Choose the one that does the job, and stay free enough to switch when a better one appears, because it always does.
The real question was never which AI is best. We already know that answer, and it changes every few months. The question is which one wins in the end: the better product, or the one that was already inside your company before you decided anything at all.
Sources
- Recon Analytics. AI Choice 2026: Why Licenses Don't Equal Adoption (U.S. AI Survey, Jul 2025 - Jan 2026): https://www.reconanalytics.com/ai-choice-2026-why-licenses-dont-equal-adoption/
- Microsoft. Ignite 2025: Copilot and agents built to power the Frontier Firm (18 Nov 2025): https://www.microsoft.com/en-us/microsoft-365/blog/2025/11/18/microsoft-ignite-2025-copilot-and-agents-built-to-power-the-frontier-firm/
- The New Stack. Microsoft's Copilot makes Anthropic's Claude and OpenAI's GPT team up (14 Mar 2026): https://thenewstack.io/microsofts-copilot-llm-team/
- Nerd Level Tech. Microsoft MAI Models: In-House AI in Copilot (Jun 2026): https://nerdleveltech.com/microsoft-mai-models-in-house-ai-copilot