Bing and Copilot Are Testing Microsoft-Decision-1 Right Now

Microsoft rolled out a new AI model called Microsoft-Decision-1 and, within days, it was already being tested inside Bing Search and Copilot. According to Bing & Copilot Experimenting With Microsoft-Decision-1, the model is described by Microsoft as a "decision-scoring model" that "delivers top performance in latency and quality on structured decision tasks to outperform both LLMs and other decision models." That's a meaningful distinction from the large language models most people think of when they hear "AI in search."

This is early-stage experimentation, not a full rollout. Bing is not using Microsoft-Decision-1 for all queries or all users. But the speed of deployment, from announcement to active testing in a major search engine within a single week, signals how quickly Microsoft is moving.

The Details: What Microsoft-Decision-1 Actually Does

Most of the AI models you hear about in search are generative: they produce text, summarize content, answer questions. Microsoft-Decision-1 is built for something different. Microsoft describes it as designed for "routing, classification, prioritization, verification, and workflow control." Think of it less as a writer and more as a traffic cop.

The model is available in Microsoft Foundry and through OpenRouter, meaning developers can already start building with it outside of Bing. That matters because it tells you this is not a one-off internal tool. Microsoft is positioning Decision-1 as infrastructure: a layer that sits inside agents, applications, and workflows to make faster, better-structured choices.

Rangan Majumder, who posted about the experiment on X, said: "We're experimenting with it in Copilot and it's pretty good! Very impressed with how quickly they improved it as well." When Barry Schwartz of Search Engine Roundtable asked the Bing team directly, Krishna Madhavan confirmed on X: "we are always experimenting!"

So the framing from the Bing side is casual, almost understated. But the underlying technology is anything but. A model purpose-built to outscore both LLMs and other decision models on structured tasks is a serious infrastructure bet.

What This Means For You

Here's the question I'd be asking if I were running SEO for a brand with meaningful Bing traffic: what does it look like when a search engine's query-routing layer gets smarter?

A decision-scoring model sitting upstream of the results layer could change how queries get classified, how intent gets bucketed, and how different result types get prioritized. If Bing uses Decision-1 to determine which queries get a standard blue-link result versus a Copilot-generated answer versus a product card, that's a direct impact on organic visibility.

Most SEOs have underweighted Bing for years. That's understandable given Google's market share, but Bing powers a meaningful share of desktop search, and Copilot has been integrated into Windows, Edge, and Microsoft 365 at scale. If Decision-1 makes Copilot's answers faster and more accurate, more users may accept those answers without clicking through at all. Zero-click pressure from a new direction.

There's also an agent workflow angle here. Microsoft explicitly built Decision-1 for agents and automated workflows. If your brand is thinking about AI agent visibility, this is the kind of model deciding whether your content gets surfaced, routed, or skipped entirely inside those automated pipelines. That's a new frontier for what some of us are calling generative search optimization.

The Classification and Routing Risk

I've watched query classification quietly reshape results for years without ever showing up in a confirmed algorithm changelog. A routing model makes those decisions faster and at scale. If a topic you rank for gets reclassified as "best answered by Copilot," your organic result may stop appearing for a segment of Bing users before you ever see it in your rank tracker.

That's not a reason to panic. It's a reason to monitor. Watch your Bing impressions and click-through rates in Search Console. If you see impression volume hold steady while clicks drop, something upstream changed in how results are being presented.

What to Do Now

1. Baseline your Bing traffic now. Pull Bing-specific data from Microsoft Clarity or Bing Webmaster Tools and note current click-through rates by query type. You want a clean before picture before any experiment becomes a wider rollout.
2. Segment your queries by intent. Decision-1 is built for classification and prioritization. Informational, transactional, and navigational queries may get routed differently as the model matures. Know which bucket your top Bing keywords fall into.
3. Check your Copilot citation footprint. Open Copilot and run a few searches for topics where you currently rank on Bing. Is your content being cited, summarized, or ignored? This is your AI visibility baseline for Microsoft's ecosystem.
4. Review structured data coverage. A decision-scoring model built for structured tasks is going to reward structured content. Make sure your schema markup is clean, your FAQs are marked up, and your product or service pages give the model something clear to classify.
5. Keep an eye on OpenRouter integrations. Microsoft-Decision-1 is available to developers outside Microsoft's own products. Third-party agents and apps that route queries through this model could become a secondary distribution channel worth tracking.

Background and Context

This move fits a broader pattern Microsoft has been building toward since integrating GPT-4 into Bing in early 2023. The company has steadily layered more AI into both the search results page and the Copilot interface, and each new model layer adds complexity to how queries get processed before a result is ever served.

What's different about Decision-1 is the specialization. Rather than making the language model bigger or smarter, Microsoft is adding a dedicated decision layer. That architectural choice, separating language generation from decision-making, reflects where serious AI infrastructure is heading. You see similar thinking in how Google separates its AI Overview generation from its core ranking signals.

The fact that this model is also available through OpenRouter means the broader developer ecosystem can route their AI applications through the same decision logic Bing is using. Over time, that could make Microsoft's classification framework the de facto standard for a class of AI-powered tools, well beyond search.

And yes, this is all still experimental. "We are always experimenting" is a phrase that could mean anything from a one-week A/B test to a staged global rollout. But in my experience, the gap between "experimenting" and "default behavior" in search has gotten shorter every year. The time to pay attention is now, not after the rollout email lands in your inbox.

If you're tracking AI visibility across both Google and Bing, AI visibility tracking can help you catch these shifts before they show up as a revenue problem.