Microsoft Bing Copilot

Microsoft Bing Copilot is an umbrella term for Microsoft’s sequence of Bing-based generative search products. The current Bing product surface is Copilot Search in Bing: a search-and-answer experience that returns summarized responses, visible citations, related topics, and conventional web links. For GEO, the key distinction is that Microsoft says this experience is grounded on Bing search results, including results for additional queries issued on the user’s behalf.

That means two gates stand between a publisher and a visible citation. First, the page must be discoverable, crawlable, and eligible for Bing’s index. Second, the system must retrieve and select it as support for a particular answer. Indexing creates eligibility; it does not guarantee citation.

1. What Microsoft Bing Copilot is

The name has covered several related but non-identical experiences. Keeping them separate prevents a result measured on an older product from being presented as evidence about the current one.

Name or surface	Role	Primary retrieval or context source	How to interpret it for GEO
Bing Chat	Historical conversational search product launched in 2023	Bing search plus Microsoft’s original Prometheus orchestration	Useful for interpreting early audits, not a current product label
Microsoft Copilot	Microsoft’s broader consumer assistant brand	Depends on the feature and context in use	Do not assume every Copilot interaction is the Bing search experience
Copilot Search in Bing	Current generative search surface inside Bing	Bing search results and additional Bing queries	Primary surface covered by this entry
Search in Copilot	Dedicated search mode in the consumer Copilot app, introduced in 2025	Microsoft’s web search and grounding infrastructure	Related current surface, but not the Bing results-page product
Copilot in Microsoft Edge	Browser assistant that can answer about the open page or search more broadly	Prompt, conversation history, and permitted browsing context	Page-context retrieval can bypass open-web discovery for the page already in view
Microsoft 365 Copilot and Copilot Search	Enterprise productivity and organizational search products	Microsoft 365 data, Graph, connectors, permissions, and optional web data	Separate from consumer Bing citation measurement

Microsoft launched the new AI-powered Bing and Edge in February 2023. Its September 2023 announcement, updated in November, said Bing Chat and Bing Chat Enterprise would become Microsoft Copilot. Bing then previewed a generative search layout in July 2024 and launched Copilot Search in Bing in April 2025.

In November 2025, Microsoft also introduced a dedicated Search mode inside consumer Copilot. It brought prominent citations and a consolidated source view to copilot.com, the mobile apps, and Edge. That is a related distribution surface, but it is not the same interface as Copilot Search in Bing.

The model layer and the retrieval layer should be analyzed separately:

Bing-indexed candidates → query-dependent retrieval → model synthesis → visible citations

A model supplied by or related to OpenAI does not imply the same candidate pool as ChatGPT Search. Google AI Overviews likewise sits on top of a host search engine, but its index, submission channels, presentation, and controls are Google’s rather than Bing’s. This distinction is central to generative engine optimization: optimize and measure the actual retrieval surface, not the model family name.

This does not mean that every Copilot answer uses only the Bing index. The broader Copilot family can also use model knowledge, Edge page context, Microsoft 365 work data, connectors, and newer grounding services. The narrower and directly documented claim is that Copilot Search in Bing is grounded on Bing search results.

2. How it works

Microsoft’s Bing generative search preview described a system that combines the foundation of Bing search results with large and small language models. It interprets a query, reviews sources, dynamically matches content, and generates a tailored layout while retaining conventional search results. The current product page adds an important detail: Copilot Search can issue additional search queries on the user’s behalf.

The observable path is therefore:

user question → query interpretation and follow-up queries → Bing retrieval → source selection → answer synthesis → inline citations and source list

Stage	What Microsoft documents	What remains undisclosed
Discovery and indexing	Bing finds URLs through links, sitemaps, IndexNow, and crawling	Complete crawl scheduling and index-selection logic
Candidate retrieval	Copilot Search uses Bing results for the original and additional queries	Full query rewriting, ranking features, and candidate counts
Grounding and synthesis	The system reads, compiles, and reasons over web information	Model routing and the weight assigned to each source
Citation display	Answers can include inline linked passages, cited sources, and a source list	The threshold that turns a retrieved result into a visible citation
Search continuation	Related topics and follow-up searches preserve the discovery context	How personalization and session context affect source choice

The interface itself distinguishes sources used in the response from other links related to the search. Microsoft’s product tutorial labels one view “Sources” and explains that an “All links” view can include pages that did not inform the answer. A page’s presence in the broader result set, its retrieval for an answer, and its visible use as a citation are separate events.

Sitemaps and IndexNow operate before final source selection. They can help Bing discover a page or a recent change, but neither submission channel is documented as a guaranteed Copilot citation signal. A clean diagnostic funnel is: discovered, crawled, indexed, retrieved, cited, and clicked.

Copilot in Edge introduces one important difference. Microsoft’s Edge guidance says a response may use the current page, page title, open tabs, and browsing history when appropriate and permitted. If a user asks for a summary of the page already open, that page is supplied through browser context rather than discovered from a fresh open-web query. Measure those page-context answers separately from Bing search answers.

3. Crawlers and user-agents

bingbot is the primary documented crawler for Bing’s standard web crawling. Microsoft’s official crawler list also names specialized agents for advertising, previews, and video, but it does not assign those agents a general Copilot Search indexing role.

As of August 28, 2026, that official list does not identify a dedicated Copilot crawler or user-agent. This is a date-bounded documentation finding, not a claim that Microsoft could never introduce one. Publishers should recheck the live list during each review cycle.

Control or observation	Effect	Important limit
Allow verified bingbot access	Lets Bing retrieve content for its web index	Access does not guarantee indexing or citation
Block bingbot in robots.txt	Prevents compliant crawling of blocked paths	A crawl block is not a reliable request to remove an already known URL from the index
Use a supported noindex directive	Requests exclusion from Bing search and associated grounding eligibility	Bing must be able to crawl the directive to see it
Submit an XML sitemap	Supplies canonical URL and update hints at scale	Submission does not guarantee crawling, indexing, or ranking
Send an IndexNow notification	Tells participating engines that a URL changed	A successful request means receipt, not inclusion

Do not copy the controls for OpenAI’s GPTBot or OAI-SearchBot from ChatGPT Search. Those agents express policies for OpenAI surfaces; they do not control Bing’s index. The relevant access path here is Bing’s crawler and indexing stack.

User-agent strings can be spoofed. Use Bing’s verification method, compare forward and reverse DNS where Microsoft instructs it, review server logs, and confirm that a WAF or CDN is not serving a challenge page to the real crawler. The broader AI crawler workflow explains how to test status codes, rendering, canonicalization, and bot-specific policy without relying on an unverified string.

4. Citation preferences

Microsoft publishes eligibility and content guidance, not a complete Copilot citation-ranking formula. Its current Webmaster Guidelines connect Bing and Copilot experiences to a shared crawling, indexing, and ranking foundation. They also distinguish content controls that affect indexing or reuse. Those statements support operational requirements, but they do not reveal a fixed weight for authority, freshness, markup, or passage structure.

Evidence layer	Supported conclusion	Unsupported leap
Official crawl and index guidance	A page needs a viable Bing discovery and indexing path to join the web candidate pool	Every indexed page is considered for every relevant answer
Official AI-content guidance	Clear subject focus, structure, evidence, and current information help systems understand and reference content	One heading pattern or schema type guarantees a citation
Interface behavior	Copilot Search shows sources used for the answer separately from other links	Every high-ranking blue link becomes a cited source
Independent audits	Source quality, citation support, and domain mix can be measured for a dated query set	A study’s percentages are permanent platform-wide rates

For publishers, the most defensible practices are straightforward. Put the answer in indexable HTML, and make each important claim complete enough to stand alone. In the relevant passage, identify the entity, date, jurisdiction, unit, and source; link to original evidence; and make factual updates visible. These practices improve interpretability and support content that can be cited cleanly, but Microsoft does not promise that any one of them is a ranking factor.

Freshness has an infrastructure component. Microsoft’s sitemap guidance for AI-powered search recommends comprehensive sitemaps for coverage and IndexNow for rapid change notification. Its explicit warning is just as important: no submission tool guarantees when or how content will appear in AI-powered results.

Compared with Google AI Overviews, Bing offers IndexNow as a change-notification channel. That difference can affect discovery latency, but it is not evidence that an IndexNow submission receives a citation boost. Compared with ChatGPT Search, a similar answer model can still retrieve a different source set because Bing’s index and query processing determine the eligible web pool.

5. API and integration

Microsoft exposes several interfaces with “Copilot,” “search,” or “retrieval” in their names. They answer different questions and should not be merged into one API story.

Interface	Intended use	Citation or retrieval output	Relationship to consumer Copilot Search ranking
Copilot Search in Bing	Consumer search and answer experience	Visible answer citations, sources, and related web links	This is the product being measured, not a general-purpose ranking API
Grounding with Bing Search and Web IQ	Give developer agents current external-web context	Grounding returns cited answers; limited-access Web IQ returns structured, citation-ready context	No one-to-one equivalence is documented
Microsoft 365 Copilot Retrieval API	Retrieve permission-aware chunks from SharePoint, OneDrive, and Copilot connectors	Enterprise content chunks for grounding	No; it queries organizational data and indexes
Bing Webmaster Tools, sitemaps, and IndexNow	Submit, diagnose, and measure publisher URLs	Crawl/index reports, change notifications, and aggregated AI citation reporting	No; these are publisher interfaces, not answer-generation APIs

Grounding with Bing Search is the closest programmable proxy. Microsoft documents a query formulation, search execution, synthesis, and source-attribution pipeline. The standard tool searches broadly across the public web; the custom-search option restricts results to configured public domains that Bing has indexed. Region, model support, configuration, tool calls, and agent instructions can all change the result, so an API experiment must record them.

The tool is useful for testing whether Bing can retrieve a page and whether a configured agent cites it. It is not evidence of the exact candidate set, ranking, model routing, or presentation used by consumer Copilot Search. Microsoft retired the legacy Bing Search APIs on August 11, 2025 and directed customers to Grounding with Bing Search. As of the review date, Microsoft does not document a public API that promises a one-to-one reproduction of the consumer surface.

Microsoft Web IQ is a newer, limited-access developer surface. It returns ranked, citation-ready context across web pages, news, images, video, licensed sources, and structured sources through REST, MCP, or an SDK. Microsoft says it is built on Bing search infrastructure but is designed for agentic retrieval rather than a conventional results page. It shows that Microsoft’s grounding services provide more than conventional blue-link results, but it still does not expose consumer Copilot ranking.

The Microsoft 365 Copilot Retrieval API is further removed from consumer Bing. It retrieves relevant text chunks from Microsoft 365 and connected organizational sources while respecting user permissions. Treat it as an enterprise grounding interface, not as a web citation API.

Publisher measurement has improved. In February 2026, Microsoft put AI Performance in Bing Webmaster Tools into public preview. It reports total citations, average cited pages, sampled grounding queries, page-level citation activity, and trends across supported Microsoft AI experiences. The aggregates do not indicate placement, authority, or rank within a specific answer, so retain direct prompt observations alongside dashboard trends.

6. History and timeline

Date	Product event	Why it matters for GEO
February 2023	Microsoft launched the new Bing preview and Edge integration, describing Prometheus and an OpenAI model customized for search	The first Bing Chat audits belong to this early product era
September to November 2023	Microsoft announced the Copilot umbrella; Bing Chat and Bing Chat Enterprise were renamed	“Bing Chat” and “Microsoft Copilot” may refer to related but different snapshots
July 24, 2024	Bing previewed generative search for a small share of queries	Generative answers moved into a search-results layout that retained classic results
April 4, 2025	Copilot Search in Bing launched as a dedicated generative search experience	Current source and citation presentation gained its canonical product name
July 31, 2025	Bing published combined sitemap and IndexNow guidance for AI-powered search	Publisher discovery and freshness practices were tied explicitly to AI results, without a guarantee
August 11, 2025	Microsoft retired the legacy Bing Search APIs	Developer testing moved toward Foundry Grounding with Bing Search rather than a classic search-results API
November 7, 2025	Microsoft introduced dedicated Search mode inside consumer Copilot	Prominent citations and source aggregation expanded beyond the Bing results-page surface
February 10, 2026	AI Performance entered public preview in Bing Webmaster Tools	Publishers gained first-party citation counts, cited URLs, grounding-query samples, and trends
June 2, 2026	Microsoft introduced Web IQ in limited access	Developers gained an agent-oriented grounding service with structured, citation-ready external context

The 2025 launch announcement said Copilot Search was available worldwide except Russia and China at launch. Availability and features can vary by market, device, and browser version, so geographic tests should record the actual interface reached rather than assuming a global experience is uniform.

Model changes are intentionally absent from most of this timeline. Microsoft can change the synthesis model without replacing Bing as the documented grounding foundation. A model-version milestone matters to GEO only when it changes retrieval, citation presentation, access, or measurable source selection.

7. Measured citation behavior

Independent studies show why dates and product labels belong beside every metric. The studies below examine different systems, query sets, and definitions; their percentages should not be merged into a single Copilot benchmark.

Research	Collection period and surface	Query domain	Reported result	Boundary
Liu et al., 2023	Early Bing Chat, alongside NeevaAI, Perplexity, and YouChat	Fact-seeking queries	Bing Chat averaged 58.7% citation recall and 89.5% citation precision in the paper’s evaluation	Early product snapshot; recall measures whether claims had citations and precision measures whether citations supported claims, not publisher visibility
Li and Sinnamon, 2024	Data collected in June 2023 from Bing Chat, ChatGPT, and Perplexity	48 public-interest queries across four topics, repeated over seven days	The audit found sentiment differences and uneven commercial, geographic, and source-type representation	Small topical sample from the earliest Bing Chat era; not a current ranking study
Allaham and Diakopoulos, 2026	User-interface audit of Copilot, ChatGPT, Gemini, and Perplexity	712 real-world queries across politics, health, and environment	Across all engines, the study collected 2,848 responses and 26,266 unique cited URLs; Copilot averaged 8.44 cited sources per response and 27.8% of its scraped citations were classified as AI-generated	Cross-engine source audit, not an optimization experiment; some cited pages could not be scraped and AI-origin detection is probabilistic

Liu et al. separate citation recall from citation precision. A high precision score does not mean most answer claims were cited; it means the citations that appeared tended to support their associated claims under the study’s method. Conversely, the 58.7% recall result does not measure whether a particular publisher could earn a citation. It measures claim-level citation coverage in the answer.

Li and Sinnamon examined which sources function as authorities in public-interest answers. Bing Chat averaged 5.13 sources per response in their sample, and 65% of references across Bing Chat and Perplexity were US-based even though the queries were issued in Canada. These findings justify auditing source diversity. They do not establish that changing a page’s category or tone causes Bing to cite it.

The 2026 synthetic-sources audit is closer to the current Copilot era. It classified about 16% of scraped citations across all four engines as AI-generated. For Copilot alone, the figure was 27.8%. Only 19,154 of the 26,266 unique cited URLs were successfully scraped, and the Pangram detector can miss or misclassify material; the authors frame the result as a lower bound. The paper supports monitoring source provenance; it does not provide a platform-wide Copilot quality score.

For operational measurement, pair academic audits with first-party diagnostics. Bing Webmaster Tools can show that a URL was cited across supported AI surfaces, but its aggregate counts do not reconstruct a particular answer. A reproducible test should save the prompt, date, market, surface, answer text, cited URL, citation context, and whether the brand was merely mentioned. Repeat prompts because retrieval and citation selection can change between runs.

8. Optimizing for Bing Copilot

The strongest Copilot-specific workflow follows the two-gate model: establish Bing eligibility, then improve and measure answer-level selection.

Action	Why it matters	Verification signal	Limit
Allow and verify bingbot	Bing needs to retrieve the public page for normal indexing	Valid crawler identity, successful status, rendered primary content	A successful crawl does not guarantee indexing or citation
Keep the page’s main answer in accessible HTML	Retrieval systems need the claim, context, and attribution in a usable response	Bing URL Inspection and rendered-page review	Rendering alone does not establish relevance
Submit a complete XML sitemap	Helps Bing discover canonical URLs and update hints	Sitemap accepted; important URLs discovered	Submission does not guarantee inclusion
Send IndexNow only for real additions, changes, or deletions	Shortens the notification path for freshness-sensitive pages	HTTP receipt plus later crawl/index observation	IndexNow explicitly does not guarantee indexing
Write self-contained factual passages	Makes a selected passage easier to quote with correct entity, date, and unit	Citation preserves the intended claim and attribution	This is an editorial principle, not a disclosed ranking weight
Cite original evidence and show update dates	Helps readers and systems verify current claims	Source links resolve and claims match the cited evidence	Authority remains query- and topic-dependent
Test stable query clusters repeatedly	Copilot can issue follow-up queries and vary source selection	Retrieval, citation, and mention rates by query and date	One run is anecdotal
Keep ChatGPT and Google controls separate	Different indexes and interfaces can produce different candidate pools	Same prompt set measured by surface	Cross-engine numbers are not interchangeable

Use Bing’s sitemap and IndexNow controls as discovery and freshness infrastructure. Use AI crawler checks to confirm access. Use the broader GEO loop to choose real audience questions, create answerable evidence, and measure citation outcomes. Each tool addresses a different failure mode.

Avoid claims that llms.txt, a schema type, a fixed paragraph length, or IndexNow directly boosts Copilot citations. None of the reviewed Microsoft documentation establishes such a guarantee. Structured data, clear headings, tables, and FAQs can make content easier to interpret, but their effect on retrieval and citation must be tested against a controlled baseline.

9. Why Bing Copilot matters for GEO

Bing’s search infrastructure connects conventional web indexing to generative answers across several Microsoft surfaces. That gives publishers more observable controls than a model-only view suggests, while leaving final source selection outside publisher control.

Layer	Publisher lever	Observable outcome	What it cannot prove
Eligibility	Crawl access, canonical URLs, index directives, sitemaps	Bing can crawl and index the page	That Copilot will retrieve it
Freshness	Accurate lastmod data and IndexNow change notifications	Bing discovers a new or changed version	That freshness receives a ranking boost
Selection	Relevant, explicit, evidenced answer passages	A URL or passage appears in a cited answer	A stable causal ranking formula
Distribution	Bing, Copilot Search, Edge, and supported partner experiences	Reach across several Microsoft entry points	Identical behavior on every surface
Measurement	Prompt logs plus Bing Webmaster Tools AI Performance	Citation counts, cited pages, query samples, and direct answer evidence	Exact rank or importance inside an answer

Google AI Overviews shares the broad pattern of a generative answer layered onto a search engine, but publishers work through Google’s index and tooling. ChatGPT Search may use related model technology, but it is a separate OpenAI surface with separate access controls and source behavior. Neither comparison changes the core Bing rule: the host retrieval system matters more than model lineage when diagnosing source visibility.

The durable GEO value of Microsoft Bing Copilot is therefore not that it is “another ChatGPT.” It is that Bing’s index serves as the documented bridge from the open web into Copilot Search’s generated answers. Treat indexing as the eligibility stage, citation as a second selection stage, and measure each independently.

References

• Microsoft, Copilot Search in Bing.
• Microsoft Bing, Introducing Copilot Search in Bing, April 4, 2025.
• Microsoft Bing, Introducing Bing generative search, July 24, 2024.
• Microsoft, Announcing Microsoft Copilot, your everyday AI companion, September 21, 2023, updated November 15, 2023.
• Microsoft, Reinventing search with a new AI-powered Microsoft Bing and Edge, your copilot for the web, February 8, 2023.
• Microsoft Bing Webmaster Tools, Which crawlers does Bing use? and Bing Webmaster Guidelines.
• Microsoft Bing, Keeping Content Discoverable with Sitemaps in AI Powered Search, July 31, 2025.
• Microsoft Bing, Introducing AI Performance in Bing Webmaster Tools Public Preview, February 10, 2026.
• IndexNow, Frequently asked questions.
• Microsoft Learn, Grounding agents with Bing Search tools.
• Microsoft Learn, Bing Search APIs retiring on August 11, 2025.
• Microsoft, Microsoft Web IQ.
• Microsoft, Bringing the best of AI search to Copilot, November 7, 2025.
• Microsoft Learn, Microsoft 365 Copilot Retrieval API Overview.
• Microsoft Support, Getting started with Copilot in Microsoft Edge.
• Liu et al., Evaluating Verifiability in Generative Search Engines, Findings of EMNLP 2023.
• Li and Sinnamon, Generative AI Search Engines as Arbiters of Public Knowledge: An Audit of Bias and Authority, 2024.
• Allaham and Diakopoulos, Synthetic Sources?: Auditing Generative Search Engine Citations for Evidence of AI-Generated Sources, 2026.