What model powers Microsoft Copilot?

Microsoft’s consumer Copilot (copilot.microsoft.com).

As of July 2026, Microsoft Copilot’s consumer app doesn’t name the model behind any given answer. A “Smart” router picks from OpenAI’s GPT-5 family per prompt, and Microsoft removed the model picker in favor of “response styles.” Microsoft names GPT-5.6 the “preferred” model only for its separate Microsoft 365 Copilot (9 July 2026). So this is a vendor-stated family, not an observed per-answer model.

Current model
GPT-5 family (Smart router)
Evidence
Vendor-stated
As of
Aug 2025
What we capture
copilot.microsoft.com, the free default

The vendor discloses this default; the engine never labels which model produced a given answer, so we report the vendor’s word with its date, not an observation.

How Copilot picks its model

Microsoft Copilot doesn’t expose a model choice. An orchestration layer — the “Smart” router Microsoft introduced with GPT-5 in August 2025 — reads your prompt and routes it to a model in the GPT-5 family: a fast model for routine questions, a deeper-reasoning model for complex ones. On Office surfaces Microsoft also blends in its in-house MAI models.

Microsoft calls GPT-5.6 the “preferred model” for Microsoft 365 Copilot as of 9 July 2026. That is the closest thing to a named default, but it is Microsoft’s statement about the work Copilot — the consumer Copilot never labels which model produced a given answer, and Microsoft removed the model picker in favor of “response styles.” Because there is no per-answer model to read, Discoverably reports the family Microsoft discloses for the consumer app — the GPT-5 family — not a specific model or an observation.

Model history

Every family-level change to the model behind Microsoft Copilot, newest first, each linked to its primary source.

GPT-5 family (Smart router)
Routing changeVendor-stated

Microsoft brings GPT-5 to Copilot with a real-time model router (the Smart orchestration layer).

Microsoft ↗ (opens in a new tab)
GPT-5.6 (Microsoft 365)
Model changeVendor-stated

Microsoft names GPT-5.6 the “preferred model” for Microsoft 365 Copilot (a separate product from the consumer app), replacing GPT-5.

Microsoft ↗ (opens in a new tab)
  1. GPT-5.6 (Microsoft 365)

    Microsoft names GPT-5.6 the “preferred model” for Microsoft 365 Copilot (a separate product from the consumer app), replacing GPT-5.

    Microsoft ↗ (opens in a new tab)
  2. GPT-5 family (Smart router)

    Microsoft brings GPT-5 to Copilot with a real-time model router (the Smart orchestration layer).

    Microsoft ↗ (opens in a new tab)

What decides whether Microsoft Copilot names your brand

A routed, Bing-grounded engine: a small model decides whether to search, then aligns the answer with high-ranking Bing content — so Bing’s ranking signals are, to first order, Copilot’s visibility signals.

How answers form & what drives brand mentions

Routed, Bing-grounded. VENDOR-DOCUMENTED A small model routes the request, decides whether web grounding is needed, and selects a model; for information-seeking text that needs more, Copilot searches, uses top web results, and aligns with high-ranking content. Bing’s index powers Copilot; classic crawlability/metadata/links remain the substrate, and for AI answers Microsoft recommends clear structure, self-contained passages, precise claims, evidence, and relevant schema. Bing’s ranking parameters (relevance, quality/credibility, freshness, popularity, location) are, to first order, Copilot grounding factors.

First-party measurement exists. VENDOR-DOCUMENTED Bing Webmaster Tools’ AI Performance report (public preview, Feb 2026) shows citations across Copilot, cited pages, sampled grounding queries, and a Citation Share metric — Microsoft calls it “an early step toward GEO tooling,” and explicitly says a citation record does not reveal placement, authority, or role. This is the only native per-engine citation dashboard among the six besides Google’s.

Freshness plumbing. VENDOR-DOCUMENTED Microsoft ties IndexNow directly to Copilot answer currency — timely change notifications refresh Bing’s index faster and cut stale references in Copilot responses.

Modality. VENDOR-DOCUMENTED (perishable) As of 2025-12-19, voice prompts did not trigger web search and voice answers did not carry citations.

Factor weights

Weight and confidence are our editorial estimate of each factor’s influence, and of the evidence for its direction — not a vendor’s ranking weight. No engine publishes a brand-selection formula.

FactorWeightConfidenceBasis
Bing indexation/crawlability (Bingbot access, BWT verification)High (binary gate)HighVENDOR-DOCUMENTED
Bing organic rank for the query + sub-queries (relevance/quality/freshness/popularity)HighHighVENDOR-DOCUMENTED substrate; transfer to citation order OUR INFERENCE
Freshness + IndexNow adoptionMed-HighHighVENDOR-DOCUMENTED (tied to Copilot grounding accuracy)
Passage structure & semantic clarity (parseable sections, direct answers)Med-HighMed-HighVENDOR-DOCUMENTED guidance; magnitude OUR INFERENCE
Authority/credibility signals (clear sourcing, consistent entity)MedMedVENDOR-DOCUMENTED ranking parameter; brand mechanics undisclosed
Structured data / schemaMedMedVENDOR-DOCUMENTED consumption for understanding; no AI-citation weighting claim (indirect effect OUR INFERENCE)
Bing Places / local entity data (local intents)Med (local), Low otherwiseMedOUR INFERENCE from Bing’s location parameter
Community/UGC + third-party coverageMedMedINDEPENDENT STUDY (cross-engine patterns; Copilot breakouts thin)

What you can do (ranked)

  1. Treat Bing SEO as first-class, not a Google afterthoughtengine-specificVENDOR-DOCUMENTED

    Verify in Bing Webmaster Tools, submit sitemaps with accurate lastmod, confirm Bingbot health, fix Bing-specific indexation gaps. Copilot cannot cite what Bing hasn’t indexed.

  2. Adopt IndexNowengine-specificVENDOR-DOCUMENTED

    Microsoft ties it to Copilot answer currency; a small ping-on-publish integration.

  3. Structure pages for parsingcross-engine · vendor-endorsed hereVENDOR-DOCUMENTED

    Clear headings, self-contained one-question sections, comparison tables, current dates.

  4. Instrument with the AI Performance reportengine-specificVENDOR-DOCUMENTED

    Use exposed grounding queries + Citation Share as the content feedback loop.

  5. Build corroborated authority, not claim volumecross-engineVENDOR-DOCUMENTED consumption; OUR INFERENCE on effect

    Relevant backlinks, independent reviews, partner references, consistent entity/schema.

  6. Maintain Bing Places for localengine-specificVENDOR-DOCUMENTED

    Location is a documented Bing input; test from target markets.

Grounding, citation & locality

VENDOR-DOCUMENTED Grounded text answers list hyperlinked citations below the text (a handful per answer; no vendor count). Not every response grounds — conversational/creative queries can answer model-only, producing uncited parametric mentions. Locality: Bing inherits device/IP/manual location; consumer Copilot inherits Bing’s market system.

Sources

Three steps to get started.

A couple of hours a month is enough to make real progress.

Enter your website.

We look up your business and the questions your customers actually ask AI.

Review what we found.

Confirm your business, your questions and your competitors. Then we ask the live apps your customers actually use: ChatGPT, Gemini, Google AI Overviews, AI Mode, Copilot and Perplexity. Not the developer API, and not the model's training data. Both give a different answer from the one your customer sees.

Get your free report.

Where you appear, who wins instead, and how AI describes you, in plain English with clear next steps. A one-time snapshot to start from.

See what AI says about you.

Get a free report: no sign-up, no credit card, results in minutes.  ·  See a sample report ›