Bing runs its own ranking stack, and while some components have direct Google equivalents, others don't - which is exactly why the same page can rank differently on each.
RankNet. Introduced by Chris Burges and a team at Microsoft Research in 2005, RankNet was one of the first neural learning-to-rank models put into production in commercial search. It learns from pairs of results, training a neural net to predict which of any two pages should rank higher for a query. It seeded a lineage at Microsoft - later LambdaRank and LambdaMART - and the learning-to-rank approach it pioneered is still foundational to how Bing orders results. For a brand, this means relevance is learned from relative comparisons across the whole result set, so the practical goal is to be the clearly better answer than the pages you sit beside, not to hit an absolute threshold.
Space Partition Tree And Graph (SPTAG). SPTAG is Microsoft's approximate-nearest-neighbor vector index, open-sourced in 2019 and used inside Bing to power semantic retrieval. It represents queries and documents as high-dimensional vectors and finds the closest matches at scale, surfacing related results when literal keyword matches are sparse. This is the layer that lets Bing understand meaning rather than only matching strings. In practice Bing's semantic layer is applied more conservatively than Google's, which is why exact-match keyword targeting still moves rankings on Bing where it has largely stopped working on Google.
Prometheus. Announced in early 2023, Prometheus is Microsoft's orchestration model that pairs a large language model with the live Bing index behind Bing Chat and Copilot. A component Microsoft calls the Bing Orchestrator turns a conversational question into multiple targeted search queries, runs them through Bing's index and ranking, then grounds the model's answer in the retrieved results and attaches citations. The answer Copilot returns is only as current and accurate as the pages Bing retrieves and trusts for it. For a brand, being cited here earns a different kind of visibility - named surface area inside the AI answer rather than a blue-link click - and eligibility traces straight back to ranking well in the underlying Bing index.
Clarity-derived UX signals. Bing has publicly stated that user-engagement signals - clicks, dwell time, and pogo-sticking back to the results page - feed into how it judges result quality. Microsoft Clarity is Bing's free UX analytics product, capturing heatmaps and session recordings of exactly that behavior across millions of sites. Microsoft has not confirmed that Clarity's own data is wired directly into the ranking algorithm, so the honest framing is that engagement is the signal and Clarity is the lens that makes it visible. For a brand, the implication is the same either way: pages that hold attention and answer the query on the first click are the ones Bing treats as high quality, and Clarity is where you diagnose the dead clicks, rage clicks, and quick bounces working against you.
Social signals. Bing has been openly willing to use social signals where Google has not. As far back as 2013 Microsoft described pulling Facebook and Twitter data into results, and Bing has repeatedly said social activity factors into how it reads a site's authority - a position Google's own spokespeople have explicitly rejected for direct ranking. The mechanism is reputational rather than a raw share count: visible, engaged social presence corroborates that a brand and its content are real and active. For a brand, that makes shares and mentions on X, LinkedIn, and Facebook a measurable lever on Bing in a way they are not on Google, and a reason to treat social distribution as part of the SEO program rather than separate from it.