Key takeaways
Prometheus is Microsoft's proprietary model, announced February 2023, that combines the fresh Bing index, ranking, and answer results with OpenAI's most advanced GPT models. A component called the Bing Orchestrator runs internal search queries, then grounds the GPT answer in retrieved Bing results and adds clickable citations. Pages that get cited earn AI-answer visibility instead of a classic click.
- Prometheus combines the Bing index, ranking, and answers with OpenAI's GPT models (Microsoft-confirmed, Feb 21, 2023).
- The Bing Orchestrator iteratively generates internal queries and feeds Bing search results to the model so answers are grounded and fresh.
- Grounding lets Prometheus answer recent questions, reduce inaccuracies, and integrate clickable citations into the answer.
- Same crawl-index-rank pipeline serves Bing Search, Copilot, and the Grounding with Bing Search API - so classic Bing SEO directly drives citation eligibility.
- Earning a citation means AI-answer surface area; use IndexNow, clear structure, verifiable facts, and clean snippet directives to qualify.
What is Prometheus?
Definition
Prometheus is Microsoft's orchestration model, announced in February 2023, that combines the fresh Bing index, ranking, and answer results with OpenAI's most advanced GPT models. It uses a component called the Bing Orchestrator to run internal queries against the index and then grounds the GPT answer in those retrieved results, adding clickable citations.
In other words, Prometheus is not a chatbot but the retrieval-augmented layer that sits between the Bing index and the language model. It is the architecture that lets Bing Chat and Copilot answer in prose while pointing to real, verifiable sources. The same pattern underpins the broader Bing ranking algorithm stack that decides which pages are eligible to appear and be cited.
Prometheus at a glance
- Built by
- Microsoft, with OpenAI's GPT models
- Announced
- February 7 and 21, 2023
- Core component
- The Bing Orchestrator
- Key technique
- Grounding GPT answers in Bing results
- Powers
- Bing Chat, Copilot, Copilot Search
- Crawler
- Bingbot (shared with Bing Search)
- Output for publishers
- Clickable citations in answers
- Core idea
- Retrieve, ground, then cite
What Prometheus actually is
Prometheus is the name Microsoft gave to the orchestration model behind the new Bing, introduced when the company relaunched Bing with AI on February 7, 2023. In its February 21, 2023 engineering post, Microsoft described Prometheus as "a first-of-its-kind AI model that combines the fresh and comprehensive Bing index, ranking, and answers results with the creative reasoning capabilities of OpenAI's most-advanced GPT models."
The key idea is that the GPT model on its own has a fixed knowledge cutoff and can hallucinate. Prometheus pairs it with the live Bing index so answers can draw on current web content and link back to real sources. Microsoft positioned the underlying model as "much more powerful than GPT-3.5, which powers ChatGPT" and customized for search. Prometheus is therefore best understood not as a single chatbot, but as the retrieval-augmented layer that sits between the Bing index and the language model.
How the Bing Orchestrator and grounding work
The mechanism at the heart of Prometheus is a component Microsoft calls the Bing Orchestrator. When a user asks a question, the Orchestrator "generate[s] a set of internal queries iteratively" - it decides what to look up, runs those queries against the Bing index, and selects the most relevant results.
Those results are then fed to the GPT model through a technique called grounding. In Microsoft's words, "the model reasons over the data provided by Bing and hence it's grounded by Bing data, via the Bing Orchestrator." Microsoft says this "provides relevant and fresh information to the model, enabling it to answer recent questions and reducing inaccuracies."
- Retrieve - the Orchestrator issues internal queries to the Bing index.
- Ground - retrieved passages are passed to the GPT model as context.
- Cite - Prometheus "integrate[s] citations into sentences in the Chat answer so that users can easily click to access those sources and verify the information."
This retrieve-ground-cite loop is the conceptual ancestor of every grounded AI search product Microsoft ships today, including Copilot and the Grounding with Bing Search API. Which results the Orchestrator can retrieve in the first place is itself decided by Bing's ranking models, including RankNet and the SPTAG vector index.
From Prometheus to today's Copilot stack
The Prometheus brand name is rarely used in current Microsoft documentation, but the architecture it introduced is now baked into the entire Bing and Copilot ecosystem. The underlying GPT models have advanced from the launch model through GPT-4 and later versions, while the orchestration-plus-grounding pattern has stayed constant.
Practically, the same crawl, index, and rank pipeline now feeds several surfaces: classic Bing Search, Copilot Search in Bing (launched April 2025), the standalone Copilot assistant, and the developer-facing Grounding with Bing Search API in Azure AI Foundry, which replaced the retired Bing Search APIs in August 2025. Because there is no separate Copilot crawler - Bingbot serves them all - the work you do to rank in Bing is the same work that makes you eligible to be cited in Copilot. Microsoft is unusually explicit about this overlap, bundling "Bing search experiences, Copilot, and grounding API results" into a single set of webmaster guidelines.
Why citations are the new currency
In a classic blue-link result, you earn a click. In a Prometheus-grounded answer, the model often satisfies the user inside the answer box and attributes the fact to a small set of cited sources. If your page is one of those citations, you earn AI-answer surface area: a named, linked appearance inside the response, even when no one clicks through.
Microsoft has formalized this shift. Its webmaster guidance now tells site owners to measure visibility beyond clicks - tracking impressions, indexing status, and citation eligibility separately. In 2026 Microsoft added an AI Performance report inside Bing Webmaster Tools that surfaces total citations, average cited pages, the sample grounding queries the AI used to retrieve your content, and page-level citation counts. That makes Bing the only major engine that shows publishers, with first-party data, whether Prometheus-style grounding is actually citing them. Off-page factors such as social engagement may shape which pages reach that retrieval pool, though Microsoft has never published a weight for them.
History of Prometheus: a timeline
Prometheus evolved from an early push-discovery protocol and a February 2023 announcement into the grounding layer behind Copilot and a first-party publisher dashboard.
-
2021
IndexNow launches
Microsoft and Yandex co-launch IndexNow, the push-based URL discovery protocol that later becomes central to keeping Copilot answers fresh.
-
2023
Prometheus announced
On February 7, 2023 Microsoft relaunches Bing with AI; the February 21 engineering post names Prometheus and the Bing Orchestrator as the grounding layer between the Bing index and OpenAI's GPT models.
-
2025
Copilot Search and API consolidation
Copilot Search in Bing launches (April 2025); the legacy Bing Search APIs are retired (August 2025) in favor of Grounding with Bing Search in Azure AI Foundry.
-
2026
AI Performance dashboard
Microsoft adds an AI Performance report to Bing Webmaster Tools, giving publishers first-party data on total citations, cited pages, and the grounding queries the AI ran.
The signals that drive Prometheus citations
Because grounding draws on the same index as Bing Search, the inputs that decide citation eligibility are largely the inputs that decide ranking. Some are documented by Microsoft; others are acknowledged in principle but never quantified, so we flag the difference plainly.
| Signal | Why it matters |
|---|---|
| Crawlability for Bingbot | Bingbot powers Bing Search, Copilot, and the Grounding with Bing Search API. If Bingbot cannot crawl or render a page, it cannot be cited. Confirmed by Microsoft's single bundled guidelines. |
| Freshness via IndexNow | Microsoft repeatedly frames IndexNow as the route to keep Copilot answers current; per Microsoft, 22% of clicked Bing URLs originate from IndexNow submissions (Feb 2026). Microsoft-stated. |
| Clear structure and verifiable claims | Microsoft's stated guidance for improving AI citation odds: depth, clear headings/tables/FAQ, claims backed by evidence, single-topic pages, key information surfaced early. Microsoft-stated. |
| Snippet directives | NOCACHE and NOARCHIVE reduce or block Copilot citation depth; the Bing-specific data-snippet attribute (Oct 2025) lets you specify citable text. Microsoft-documented. |
| User-engagement signals | Bing has historically acknowledged that user-engagement metrics feed its ranker more openly than Google does. The principle is Bing-acknowledged; the precise weight is practitioner inference, not a published figure. |
| Social signals | Bing has long said it considers social engagement, unlike Google. That it factors in is Bing-acknowledged in principle; the magnitude and the idea that likes or shares move citations are industry inference, not an official metric. |
The practical takeaway is that the documented signals deserve your effort first. For the contested ones - engagement and social - treat the principle as real but resist any tool that quotes a precise weight, because Microsoft has never published one. The Clarity-derived UX signals guide walks through exactly where that line sits.
How to earn Prometheus citations
Because grounding draws on the same index as search, the fundamentals of earning a citation are mostly strong, well-structured, verifiable content - plus a few Bing-specific levers.
-
Submit new and updated URLs through IndexNow
Microsoft frames IndexNow as the primary way to keep Copilot answers fresh, and reports a large share of clicked Bing URLs originate from IndexNow submissions.
-
Lead each page with a self-contained, factual answer near the top
Grounding lifts atomic, verifiable passages; Microsoft's guidance tells creators to surface key information early and make facts explicit rather than implied.
-
Add clear headings, tables, and FAQ blocks supported by evidence
These are the exact structural and credibility signals Microsoft names for improving the odds of being cited in AI answers.
-
Audit robots and meta directives for NOCACHE and NOARCHIVE
Both directives reduce or block Copilot citation depth; removing them on cite-worthy content restores eligibility, and data-snippet can specify the exact citable text.
-
Track citations in the Bing Webmaster Tools AI Performance report
It is the only first-party view of total citations, cited pages, and the grounding queries the AI used - measure visibility beyond clicks, as Microsoft now advises.
-
Keep entity references consistent across text, images, and video
Microsoft tells creators to reduce ambiguity across formats so the model can reliably resolve which entity a page describes before citing it.
Prometheus myths vs. reality
Prometheus sits at the intersection of search, AI, and marketing folklore, so it attracts a lot of confident but wrong claims. Here are the most common ones and what is actually true.
Myth Prometheus is just a rebranded version of ChatGPT.
Reality Prometheus is an orchestration layer, not a single model. Microsoft confirmed it combines the live Bing index, ranking, and answers with OpenAI's GPT models via the Bing Orchestrator and grounding - capabilities ChatGPT alone does not have.
Myth Microsoft Clarity behavioral data feeds Bing rankings and citations.
Reality Microsoft's own Clarity documentation describes it strictly as a "behavioral analytics tool" for heatmaps and session recordings. There is no Microsoft statement that Clarity data is a Bing ranking or citation signal. Treat Clarity-feeds-ranking as unverified industry speculation.
Myth Social shares are a major, quantified Bing ranking factor.
Reality Bing has acknowledged it considers social engagement, unlike Google, but Microsoft has never published a weight. The existence of the signal is Bing-acknowledged; its magnitude and any effect on AI citations are practitioner inference, not official.
Myth You need a separate strategy to get cited by Copilot.
Reality Microsoft uses one crawl-index-rank pipeline and one crawler, Bingbot, for Bing Search, Copilot, and the Grounding with Bing Search API. Strong classic Bing SEO is what makes a page eligible to be cited.
Myth Citations do not matter because they do not send clicks.
Reality Citations are the new visibility currency. Microsoft itself now tells site owners to measure impressions and citation eligibility "beyond clicks," and ships an AI Performance report so publishers can track citation surface area directly.
Frequently asked questions
Prometheus is Microsoft's orchestration model, announced in February 2023, that combines the fresh Bing index, ranking, and answer results with OpenAI's most advanced GPT models. It uses the Bing Orchestrator to run internal queries and grounds GPT answers in retrieved Bing results with clickable citations.
Microsoft relaunched Bing with AI on February 7, 2023 and detailed Prometheus in a Bing engineering post on February 21, 2023. The name describes the model that powered the original Bing Chat, which has since evolved into the broader Copilot and grounding stack.
The Bing Orchestrator iteratively generates a set of internal search queries, runs them against the Bing index, and selects the most relevant results. Those results are then fed to the GPT model as grounding context, so the answer reflects current web data and can cite real sources.
Grounding is the technique where the GPT model reasons over data the Bing Orchestrator retrieves from the Bing index, rather than relying only on its training data. Microsoft says grounding provides fresh information, lets the model answer recent questions, and reduces inaccuracies in the response.
Earn citations the way you earn rankings: let Bingbot crawl the page, submit updates via IndexNow, write self-contained verifiable facts, use clear headings and tables, and avoid NOCACHE or NOARCHIVE. The same crawl-index-rank pipeline that ranks pages decides citation eligibility.
There is no Microsoft confirmation that Clarity feeds Bing rankings or citations. Microsoft's Clarity documentation describes it only as a behavioral analytics tool for heatmaps and session recordings. The claim that Clarity influences ranking is industry speculation, not official guidance, and should not be presented as fact.
Bing has long said it considers social engagement, which distinguishes it from Google. However, Microsoft has never published the weight of social signals, and no official source ties likes or shares to AI citations. Treat the existence of the signal as acknowledged but its magnitude as inference.
Not exactly. Prometheus is the orchestration-plus-grounding architecture introduced for Bing Chat. Copilot is the consumer product that now uses an evolved version of that architecture. The Prometheus brand is rarely named today, but the retrieve-ground-cite pattern it introduced underpins Copilot, Copilot Search, and the Grounding with Bing Search API.
The bottom line
Bottom line
Prometheus turned the Bing index into the live grounding source for an answer engine, and that idea now runs the whole Copilot stack. The brand name has faded, but the retrieve-ground-cite loop is what decides whether your pages appear inside AI answers. Earn it the same way you earn rankings: keep Bingbot crawling, push freshness through IndexNow, write self-contained verifiable facts, and clear the snippet directives out of the way.
References
- Building the New Bing (Prometheus announcement) - Bing Search Quality Insights
- Reinventing search with a new AI-powered Microsoft Bing and Edge - The Official Microsoft Blog
- Microsoft explains how Bing AI Chat uses ChatGPT and Search with Prometheus - Search Engine Land
- Introducing Copilot Search in Bing - Bing Blogs
- Bing Webmaster Guidelines
- Grounding with Bing Search - Microsoft Learn (Azure AI Foundry)
- Clarity Overview - Microsoft Learn
- SPTAG: Space Partition Tree And Graph (Microsoft Research / Microsoft Bing) - GitHub
- IndexNow Documentation