Enterprise AI Hit 15 Million Seats Inside Microsoft 365. Your Buyers Are Already Using It to Build Vendor Shortlists.
Microsoft disclosed 15 million paid Copilot seats in January — AI embedded in Word, Outlook, and Excel. When enterprise buyers ask it to draft a vendor comparison, they never open a browser. The discovery happening in those tools runs on the same rule as every other AI: the publication record wins.
On January 28, 2026, Microsoft held its Q2 FY2026 earnings call. For the first time, Satya Nadella put a hard number on enterprise Copilot adoption: 15 million paid seats inside Microsoft 365, out of a 450-million-seat base.
That number matters more than the earnings headline.
Those 15 million seats don't sit in a browser tab. They sit inside Word. Inside Outlook. Inside Excel. When an enterprise buyer drafts a vendor comparison or searches their inbox for past conversations about a category and asks the AI embedded there to synthesize options, that AI answers. No browser. No new tab. No interaction that registers in your analytics.
TechCrunch reported that Nadella was unusually deliberate about the disclosure. He specifically distinguished between the 450 million total Microsoft 365 seats and the 15 million paid Copilot seats, companies paying an additional $30 per user per month for AI embedded in their daily work environment. This wasn't an incidental number. It was the number Microsoft chose to anchor its AI narrative around when investors were asking whether the spending was producing anything real.
That same month, Google announced Gmail was entering what it called the "Gemini era." Gemini is now embedded directly in the inbox for more than 3 billion users, generating responses, drafting comparisons, and synthesizing vendor conversations from email threads the buyer never consciously went looking for. Google's own internal data showed that 70% of enterprise users who use the AI writing tool took Gemini's suggestion outright.
The pattern is the same across both platforms. The AI doing vendor research for enterprise buyers is no longer a thing you open. It's a thing that's already open.
Most AI visibility strategies are built around a specific model of buyer behavior: a buyer opens ChatGPT.com, or Perplexity.ai, or Google AI Mode, and types a query. Your brand either appears in the response or it doesn't. We've written about that layer before.
That model is increasingly incomplete.
The buyers running pre-purchase research through embedded AI tools aren't auditable through referral traffic. They're not producing LLM traffic spikes in your analytics. A VP of Procurement asking Copilot to draft a shortlist for CRM vendors, directly inside Outlook, leaves no external footprint. The question gets answered. The shortlist gets built. The decision gets shaped, and your pipeline either benefits or doesn't, without any indicator you could have caught.
With 15 million paid seats now confirmed and Microsoft actively positioning Copilot as the enterprise standard, the usage curve is going up. And this is the conservative read: the 15 million figure covers only the paid dedicated seats, not the broader M365 users with access to free-tier AI features in Outlook and Teams.
Here's what doesn't change.
Whether the AI is ChatGPT.com or Microsoft Copilot embedded in Word, the mechanism determining what it recommends is the same. The AI needs to know something about your brand before it includes you in an answer. That knowledge comes from somewhere. And the somewhere has been consistent across every serious analysis of how AI systems decide what to surface.
A September 2025 study published on arXiv ran a large-scale controlled analysis across multiple verticals and query types, comparing what AI search engines cite versus what Google surfaces. The finding: AI search shows "a systematic and overwhelming bias towards Earned media (third-party, authoritative sources) over Brand-owned and Social content." Brand websites, landing pages, and product pages ranked lowest in citation weight. Third-party editorial coverage in publications with established credibility ranked highest.
Microsoft Copilot pulls from the Bing index, which is weighted the same way. LinkedIn articles, industry trade press, named editorial coverage in recognized publications: that's the corpus Copilot draws from when a buyer asks it to synthesize vendor options. Your blog posts and product descriptions are structurally excluded from the outputs that matter.
Google's Gemini in Gmail is built on Google's knowledge graph and index, the same sources that power AI Overviews, which have shown the same editorial bias.
The embedded AI layer doesn't change the rules. It just makes the consequences of ignoring them harder to see until a deal is already lost.
The argument most teams make when this comes up is that they can't do anything about the Copilot layer specifically. They can't optimize for it. There's no Copilot-specific keyword strategy. There's no Copilot SEO.
That argument is right about the tactics and wrong about the conclusion.
You can't optimize directly for Microsoft Copilot's answers. What you can do is build the editorial record that Copilot pulls from when it constructs those answers. That record is made up of the same things that built brand authority with human readers for decades: coverage in publications the AI treats as credible, in formats structured around the questions buyers actually ask, accumulated over time across multiple outlets so the AI isn't drawing from a single data point.
One TechCrunch profile doesn't change your Copilot citation rate. Three pieces across Forbes, TechCrunch, and a respected trade publication on the same positioning, published within a reasonable window, start to build what looks like category authority to an AI synthesizing a response in real time.
That's the mechanism. It's always been the mechanism. The AI era just made the stakes clearer.
Related Reading
- AI Visibility for SaaS Companies: The 2026 Earned Media Playbook
- AI Visibility for Healthcare Companies: The 2026 Earned Media Playbook
Machine Relations, the discipline of building earned media presence specifically for AI citation, exists because the reader changed before most brands noticed. machinerelations.ai frames the shift: what PR got right was always the mechanism. A placement in a publication the AI treats as authoritative is a placement in the answer your buyers are going to get when they ask their embedded AI to help them think. The publication hasn't changed. The reader has.
Most B2B brands have a ChatGPT visibility problem. Fewer have diagnosed that the same problem exists inside their buyers' Outlook.
The brands that close that gap first don't do it by optimizing for Copilot. They do it by building the kind of editorial record Copilot already knows how to find.
Run your AI visibility audit to see where the embedded layer currently finds — or doesn't find — your brand.