Copilot’s answers begin somewhere specific, and it is not the language model most people picture. They begin in Bing’s search index. When you ask Copilot a web question, it queries Bing, retrieves candidate pages, and then writes a synthesized answer with numbered citations drawn from what it found. That order matters more than any other single fact about how Microsoft Copilot chooses sources, because it means the citation contest is decided first in Bing and only finished in the model. Miss the Bing step and the model never gets the chance to quote you.
It starts with the Bing index, not a language model
The instinct to think of Copilot as a pure AI that somehow knows things is misleading. For web answers, Copilot is a retrieval system wearing a conversational interface. The GPT-based model does the writing, but the sources come from Bing, and a page that Bing has not indexed well is a page Copilot cannot reach. This is why brands that dominate Google and neglect Bing get skipped by Copilot despite excellent content. The model never saw them, because the index it draws from did not surface them.
Once you accept that Bing is the gatekeeper, the priorities reorder themselves. Your first job is not to write the most brilliant page. It is to make sure Bing has indexed your pages and ranks them for the queries you care about. Bing Webmaster Tools, sitemap submission, and IndexNow are the levers. They are unglamorous, and they are the difference between being a candidate Copilot can consider and being invisible to the whole system. Everything downstream depends on clearing this gate first.
This ordering has a blunt consequence worth stating: your Google performance does not transfer automatically. A page that dominates Google can be nearly absent from Bing, and since Copilot grounds in Bing, that page is nearly absent from Copilot too. Brands discover this the hard way, assuming their strong Google presence covers them everywhere, then finding Copilot cites competitors they beat handily on Google. The two indexes are built and ranked separately, and Copilot inherits Bing’s version of the web, not Google’s. Checking your Bing presence directly, rather than assuming it mirrors Google, is the single most clarifying thing you can do.
The retrieval-then-synthesis pipeline
Copilot works in two stages, and understanding the split clarifies where citations are won. First comes retrieval: Bing returns a set of candidate pages relevant to the query. Then comes synthesis: the model reads those candidates, selects the passages that best answer the question, and writes an answer that attributes them. Retrieval decides who is in the running. Synthesis decides who gets quoted.
Different work wins each stage. Retrieval rewards classic search fundamentals in Bing, relevance, indexing, and enough authority to rank. Synthesis rewards clarity and quotability, passages that state the answer plainly enough to lift and attribute. A page can win retrieval and lose synthesis by being relevant but vague, or it can be beautifully written and lose retrieval by being invisible in Bing. To be cited consistently, you have to win both stages, which means doing the Bing plumbing and the on-page clarity work, not one or the other.
Separating the two stages also clears up a common frustration. Brands often cannot tell whether their Copilot problem is a visibility problem or a quality problem, and they end up guessing. The pipeline gives you a clean diagnostic. If your page does not surface in Bing for the query at all, you have a retrieval problem, and the fix lives in indexing, ranking, and the Bing plumbing. If your page surfaces in Bing but Copilot never quotes it, you have a synthesis problem, and the fix lives in how the page is written. Same symptom, opposite cures, and knowing which stage you are failing is the difference between wasted effort and targeted work.
The two stages also fail silently in different ways, which is why they go undiagnosed. A retrieval failure is invisible because nothing appears, so there is no error to notice, just an absence. A synthesis failure is invisible because the page ranks and gets traffic, so it looks healthy while quietly losing every citation to a clearer competitor. Neither failure announces itself, and both get blamed on vague notions of the algorithm being unknowable. It is not unknowable. It is a two-stage pipeline, and how Microsoft Copilot chooses sources becomes legible the moment you test which stage is dropping you.
What makes a passage win the citation
In the synthesis stage, Copilot is choosing among candidate passages, and a few qualities decide the winner. The passage that answers the exact question most directly tends to win, because Copilot is filling a specific slot in its answer and wants a clean fit. The passage from a source Copilot can identify and trust wins over one from an ambiguous or low-credibility source. And the passage that aligns with what other retrieved sources say wins over an uncorroborated outlier, because agreement lowers the risk of repeating it.
This means your writing choices directly shape your synthesis odds. Lead sections with the answer. Make claims specific and verifiable. Ensure your key points are corroborated across the credible web that Bing indexes, so Copilot finds agreement when it assembles the answer. None of this is exotic. It is the same discipline that helps across AI search, applied with the awareness that for Copilot, the candidates were all pulled from Bing. Win the clarity contest among Bing’s results and you win the citation.
The Enterprise Grounding Stack
Here is a model to hold the whole system in view. Call it the Enterprise Grounding Stack, four layers Copilot moves through before it names you. The bottom layer is index presence: Bing has your page and can retrieve it. The second layer is retrieval relevance: your page surfaces for the query among the candidates. The third layer is passage quality: your content states the answer in a clean, quotable form. The top layer is trust and corroboration: your source is credible and your claim is echoed elsewhere.
Most brands that fail to get cited are missing the bottom layer without realizing it, because they measured their success in Google and never checked Bing. They have great passage quality and real authority sitting on a page the grounding stack cannot retrieve. Diagnosing which layer you are missing is the fastest way to fix your Copilot visibility. A page missing the bottom layer needs Bing work. A page that retrieves but never gets quoted needs clarity work. The stack tells you which, so you stop applying the wrong fix to the right problem.
The Bing plumbing that decides retrieval
Since the bottom layer of the grounding stack is Bing presence, it is worth being concrete about the plumbing that controls it, because this is where most brands quietly lose. Start with Bing Webmaster Tools, the equivalent of Google Search Console for Bing, where you can submit your sitemap, see which pages are indexed, and read the coverage issues Bing has found. Many brands have never opened it, which means they have no idea whether Bing can even see the pages they hope Copilot will cite. That blind spot is the single most common reason a strong site is invisible to Copilot.
IndexNow is the second lever, and it matters more for Bing than most realize. It lets you notify Bing the moment you publish or update a page, so the page can be crawled and indexed far faster than waiting for Bing to find it on its own. For time-sensitive content, that speed difference decides whether you are in the candidate pool while a topic is current or arrive after the answer has already been assembled from someone else’s page. Wiring up IndexNow is a one-time setup that pays off on every future publish.
The rest of the plumbing is ordinary technical hygiene viewed through a Bing lens: crawlable pages, clean sitemaps, reasonable load speed, and robots rules that do not accidentally block the crawler you want. None of it is exotic, and all of it is easy to neglect when your attention is fixed on Google. The brands that get cited by Copilot are, almost without exception, the ones that did this unglamorous Bing plumbing while their competitors assumed Google performance would carry over. It does not carry over on its own, and the gap is the opportunity.
Where entity signals tip the synthesis
Once retrieval is handled, the synthesis stage often comes down to entity clarity, and this is where two comparable pages get separated. When Copilot has several relevant passages and must choose which to attribute, it favors the source it can identify with confidence. A page from a clearly-named organization, written by a credible author, with structured data spelling out who published it, is a safer attribution than a passage from an anonymous page whose ownership is murky. Confidence in the entity translates into willingness to cite.
This is why entity work quietly lifts your whole site rather than one page. The clearer and more consistent your brand is across your About page, your author bios, your structured data, and your presence elsewhere on the web, the more readily Copilot treats any given page from you as a trustworthy source. Ambiguity does the reverse, introducing a small hesitation that, when a competitor’s entity is clearer, tips the citation their way. In a synthesis step choosing between similar passages, that hesitation is often the whole margin.
The move is to make your identity unmistakable everywhere Copilot might encounter it. Use one consistent brand name. Give your authors real, verifiable credentials. Mark up your organization and content so machines can parse who stands behind each claim. These signals cost little and compound across every page, and they are exactly what a careful system leans on when it decides whose words are safe to repeat. Entity clarity does not replace clear writing or Bing visibility, but when those are equal, it is frequently what determines how Microsoft Copilot chooses sources between you and the next candidate.
Read Copilot’s citations like a scoreboard
The best way to learn how Microsoft Copilot chooses sources is to watch it choose. Ask Copilot the questions your content should answer, and study the numbered sources it returns. Who got cited? Are they stronger in Bing than you, clearer on the page than you, more corroborated than you? Each answer is a scoreboard showing exactly where you are losing the grounding stack, and the gaps repeat in patterns you can act on.
Then close the gaps in order. Fix Bing visibility first, because it gates everything. Sharpen the passages Copilot would want to quote. Strengthen your entity and earn the corroboration that makes your claims safe to repeat. Re-run the same questions weeks later and watch the citations shift toward you. Copilot’s source selection is not a mystery to accept. It is a system with visible outputs you can read, diagnose, and beat, one layer of the grounding stack at a time.