Microsoft Is Quietly Replacing OpenAI Inside Copilot. Your AEO Strategy May Not Have Noticed.

Microsoft has started routing tasks inside GitHub Copilot, Excel, and Outlook to its own in-house MAI models, replacing work that previously went to OpenAI and Anthropic. CEO Satya Nadella confirmed the shift in a post on X, describing it as a cost and capability decision. Copilot Chat and PowerPoint are next in line.

The MAI models in question include MAI-Image-2.5-Pro and MAI-Voice-2-Flash, which Microsoft says cut GPU costs by up to 89 percent compared to frontier alternatives. Nadella framing is direct: software now has a real marginal cost for the first time, which makes the choice of model a business decision rather than a technical default. Frontier models stay in the mix for frontier problems. Everything else goes to whatever is cheapest and good enough.

For enterprise users, this is largely invisible. Copilot still answers your questions. The model behind the answer is just different. That invisibility is precisely the problem for AEO practitioners.

AEO strategies built around ChatGPT citation patterns assume that Copilot behaves similarly, because both historically ran on GPT-4. That assumption is now wrong for a growing share of Copilot queries. Microsoft MAI models were trained on different data, with different fine-tuning priorities and different retrieval behaviors. A brand that has built strong citation visibility in ChatGPT may not have the same visibility in a Copilot session that routes to MAI-Image or a Phi-4 variant.

The enterprise context makes this particularly significant. Copilot is the AI interface for most Fortune 500 employees. It handles procurement research, vendor comparisons, market analysis, and competitive intelligence queries. These are exactly the high-intent, high-value queries where brand citation matters most. If Microsoft own models are now answering those queries, and if those models have different citation patterns than GPT-4, then enterprise AEO strategies need to be recalibrated.

The practical implication is the same as the OpenRouter story: the model behind the answer is no longer a constant. Microsoft has made it a variable it controls. Brands doing AEO need to test citation visibility in Copilot directly, not infer it from ChatGPT performance. The two products are diverging, and the gap will widen as Microsoft pushes more tasks to its own models.

None of this means GPT-4 is going away from Copilot. Nadella was explicit that frontier models stay for frontier problems. But the definition of what counts as a frontier problem will keep shrinking as MAI models improve. The direction of travel is clear: Microsoft wants to own the model layer inside its own products. For brands, that means the citation landscape inside the most widely deployed enterprise AI tool just became harder to predict and more important to monitor.

AEO Updates is published by The Prompt Group. Editorial decisions sit with the AEO Updates team, and any commercial relationship that touches a story is labelled on the page.