Microsoft Just Gave Up on Consumer Chatbots. Your Website Still Needs One.

On September 25, 2026, Microsoft did something rare for a company of its scale: it publicly stepped back from a market it once wanted to dominate. In a keynote rollout on its official blog, Microsoft announced a full redesign of Copilot, merging its consumer and workplace versions into a single enterprise-first application. In the words of one industry analyst, Microsoft is “trading the ChatGPT bake-off for margin on seats it already owns.” Bloomberg summed it up bluntly: Microsoft cedes the personal chatbot race to OpenAI, Google and Meta, and doubles down on paid corporate use.

If you run a small or mid-sized business with a public website, this announcement is more relevant than it looks — but not for the reasons Microsoft’s press release suggests. Here is what it means for the chatbot your customers actually interact with.

What Microsoft Actually Announced

The new Copilot is built around three surfaces:

• Home — a hub where the classic chatbot experience (Chat) and the productivity mode (Cowork) come together. Home is where an employee starts their day inside Microsoft 365.
• Code — the user describes an application in plain English, and Copilot generates a TypeScript/React app that deploys directly inside the company’s Microsoft 365 tenant. Home and Code will roll out first through the Frontier programme “in the coming weeks.”
• Autopilot — a persistent named agent. You give it a role and a goal, and it “keeps working even when you’re not,” watching channels, following up on threads, running recurring tasks. Autopilot is in private-preview expansion.

Pricing has also shifted. The existing $30/user/month seat license stays. On top of it, Microsoft is introducing Copilot Credits — a metered consumption model that bills Cowork, Code, Autopilot and frontier models by usage. No credit price has been published yet, and administrators must create a spending policy before any of this is switched on.

Why Microsoft Cedes the Consumer Chatbot

The reasoning is not subtle. Microsoft has more than 450 million commercial Office 365 seats worldwide. Fewer than 7% of those seats carry a Copilot license — about 30 million paid subscriptions. Three years after the AI hype cycle began, the flagship generic chatbot bundle has stalled below 7% inside its own captive base.

The rebuild does two things at once. It stops trying to beat ChatGPT on generalist consumer chat — a market where distinct product moats are hard to defend. And it repositions Copilot as agentic, tenant-bound, work-oriented software. Moor Insights & Strategy put it plainly: Microsoft now “starts selling work, not just seats.”

The implicit admission is worth reading carefully: a generic chatbot is not a winning product. What wins is specific work, tied to specific data, inside a specific perimeter. That principle is exactly why RAG-based, bounded chatbots exist.

Two Jobs, Two Tools

There is a common confusion in this news cycle. Because “Copilot” and “chatbot” share vocabulary, businesses assume Microsoft’s new stack replaces every AI conversation surface they need. It does not.

There are two different AI jobs in your business:

• Job A — internal productivity. Your employees need to summarize meetings, draft emails, query internal data, run recurring workflows. This lives inside your Microsoft 365 tenant, behind SSO. The new Copilot Home / Code / Autopilot targets this.
• Job B — customer conversations on your public website. A visitor lands on your homepage from Google, from an AI Overview, from a paid ad. They have three seconds to decide whether you can answer their question. They do not have a Microsoft 365 account. They will never open Copilot. They see whatever chatbot you put in front of them.

Job A and Job B are orthogonal. They serve different populations, they respect different data boundaries, they answer to different regulations. Copilot with Autopilot does not solve Job B. Nothing in Microsoft’s September 25 announcement is designed for anonymous public visitors on your marketing site.

Confusing the two — pointing an internal productivity agent at your public visitors — creates three problems at once:

1. Regulatory surface stacking. Microsoft is a US provider, so anything routed through Copilot is exposed to the CLOUD Act. EU AI Act Article 50 (in force since August 2, 2026) requires that any chatbot in contact with an EU user be identifiable as AI, with an audit trail. Public-facing conversations captured inside a workplace productivity tool complicate that trail considerably.
2. Broad-access agent risk. Autopilot is a persistent, agentic system — the same class of software that was involved in the OpenAI agent breach of Australia’s Medicare portal in June 2026 (disclosed only in September) and in the GPT-5.6 Sol / Hugging Face incident of July 2026. When a persistent agent has broad reach and vague instructions, unpredictable actions become structurally possible.
3. Cost model mismatch. Copilot Credits meter internal work. Your public chatbot handles thousands of anonymous conversations a month — most of them low value, some of them extremely valuable. Metered internal-productivity pricing was not built for that traffic pattern.

The Hidden Signal for SMBs

For a French or European SMB reading Microsoft’s announcement, the headline is not “Copilot got better.” It is: the biggest AI vendor in the world just admitted that generic chatbots are not a product. Specific, bounded, work-tied AI is the product.

That admission validates what small businesses already suspected. You do not need a superintelligent assistant that opens Google Docs and edits your CRM to answer a question like “Do you deliver in Nantes on Saturdays?” You need a chatbot that reads your delivery PDF, cites it, and stops there.

Where DoxyChat Fits

DoxyChat is built for Job B. It is not competing with Copilot Home for your employees’ inbox, and it is not competing with Autopilot for CRM automation. It is what your public visitors talk to.

Three properties matter:

• Bounded — DoxyChat’s RAG engine answers only from the documents you upload. No internet, no CRM writes, no email drafting. If the answer is not in your knowledge base, the chatbot says so. That is the opposite of a broad-access agent.
• Deterministic — the same document produces the same answer. This is essential for legal, medical, financial and public-sector contexts where consistency is a compliance requirement.
• Auditable — every response is traceable to the chunks and prompt that produced it, satisfying EU AI Act Article 50 by design.

The stack underneath is European-first: Scaleway data centers in France, Mistral open-weight models under Apache 2.0, PostgreSQL Row-Level Security for tenant isolation. No CLOUD Act exposure, no per-seat license for your public visitors, no Copilot Credits meter on anonymous conversations. A one-line JavaScript widget deploys the chatbot on your website in two minutes.

The Discovery plan is free — 1 chatbot, up to 10 documents, 200 requests per month. Enough to test whether your customers actually find answers before you commit to anything.

Conclusion

Microsoft’s September 25 pivot is a valuable signal for anyone running a customer-facing business online. The world’s largest software vendor just told the market that a general-purpose chatbot inside a productivity suite does not compete. Specific, bounded, work-tied AI does.

Your Microsoft 365 seats can absolutely benefit from Home, Code and Autopilot for internal productivity. But your public website — the entry point where prospects decide in three seconds whether you are the answer to their question — still needs its own chatbot. Bounded, sovereign, auditable, and priced for real customer conversations, not internal Copilot Credits.

Try DoxyChat free — deploy in two minutes on your website, keep your data in France, and stay Article 50-compliant from day one.