How Microsoft Copilot Cites Brands: A 2026 GEO Playbook

Microsoft Copilot now ships by default on roughly 400 million Windows PCs and is folded into every Microsoft 365 subscription. For most B2B brands, that makes it the highest-distribution AI assistant on the planet. And yet most marketing teams still treat Copilot as an afterthought.

This post is the playbook. We cover where Copilot pulls its facts from, how it picks one brand over another, the new Bing Webmaster Tools dashboard that finally exposes the data, and the concrete steps to move your share of Copilot citations up and to the right.

What is Microsoft Copilot actually citing from?

Copilot is not a single model with a single index. It is an orchestration layer that pulls from three distinct knowledge sources, and which one wins depends on the surface you are using.

The first is the Bing index. When you ask a public, web-grounded question (the kind a journalist or analyst would type into a search bar), Copilot retrieves a candidate set of pages from the Bing search index, ranks them, and lets a large language model summarise the top results. This is similar in spirit to how Perplexity works, but the candidate pool is different because Bing's crawl is different from Google's.

The second is the Microsoft Graph. If you are signed in to Microsoft 365 and you ask a question inside Word, Outlook, or Teams, Copilot can ground the answer in your tenant's documents, emails, chat history, and calendar events. This is the part that competitors cannot touch. It is also the part where third-party brand citations leak in unexpectedly, because your customer's pitch deck citing you is now a retrievable source.

The third is the live web. For time-sensitive queries, Copilot can fetch fresh pages outside the indexed corpus. Microsoft has been expanding live-web usage since the November 2025 update that brought "Best of AI Search" into Copilot, and the citation surface has grown with it.

Why this matters: optimising for Copilot is not a single SEO task. It is three separate jobs. Be indexable, fresh, and authoritative in Bing. Be quoted by customers, partners, and analysts whose documents live in Microsoft 365 tenants. Be the page a real-time crawl would land on for your category. If you only solve for the first job, you are leaving most of Copilot's citation surface on the table.

How Copilot decides which brand to recommend

When a Copilot user asks "what's the best invoicing tool for freelancers" the model does not pick a brand at random. It runs a sequence Microsoft has documented in pieces across the Bing webmaster blog and the Copilot product blog.

Step one is query rewriting. Copilot generates what Microsoft calls grounding queries. These are internal search phrases, often two or three of them, written for retrieval rather than for human readability. A user prompt of "best invoicing tool for freelancers" might be rewritten as "freelancer invoicing software comparison 2026" plus "self-employed invoicing app reviews." Your job, then, is to rank for queries you may never see, because they live inside the model and not in any analytics dashboard.

Step two is retrieval and ranking. Copilot pulls candidate pages from the Bing index and scores them on a familiar mix of relevance, freshness, and credibility signals. Pages that load fast, have clear headings, and have inbound links from sources Bing already trusts get pushed up.

Step three is source filtering. This is where most brands quietly lose. Copilot prefers sources that are easy to parse: clear H2 structure, definitions near the top, comparison tables, FAQ blocks. Pages built as image-heavy landing pages with the meat buried in carousels rarely make the cut.

Step four is synthesis with attribution. The model writes the answer and, since the November 2025 update, surfaces inline numbered citations. The brands that get named are the ones that survived steps one through three with enough specificity that the model can attribute a clean fact to them.

Two practical implications fall out of this. First, generic content gets filtered out. If your blog post says "good invoicing tools should be easy to use" you provided no fact for Copilot to attribute. The competitor who said "Wave invoicing has no monthly fee and supports multi-currency billing in 180 countries" provided one. They get cited. You get summarised. Second, the entity has to be unambiguous. Copilot needs to know that Bold GEO is a company, that it tracks AI citations, and that the URL it should attribute to is boldgeo.co. This is the same entity-clarity problem we cover in our piece on the seven factors of AI visibility, but Copilot is especially strict about it because of its enterprise-data integration.

Where Copilot citations actually show up

The web chat experience at copilot.microsoft.com is only one of seven surfaces where a Copilot citation can land in front of your buyer.

There is Copilot in Word, where a sales rep asks the assistant to draft a comparison memo and Copilot inlines citations for each tool it mentions. There is Copilot in Outlook, where a prospect asks "draft a reply to the vendor evaluation thread" and the model surfaces relevant emails plus public sources. There is Copilot in Teams meeting summaries, where the assistant transcribes "we should consider Tool X" and links the brand. There is Copilot in PowerPoint, where the model now suggests vendor logos and references based on the slide's topic.

Then there is the LinkedIn integration. Copilot's profile and message-generation features pull from LinkedIn data and surface brand mentions in suggested intros and account research summaries. For B2B specifically, this is the highest-leverage citation surface most marketing teams have never measured.

Finally, there is Copilot in the Edge sidebar, Copilot inside Bing search, and the Copilot mobile app. Each has its own slight ranking variation but all three lean on the Bing index as the primary source.

The point: a single Copilot citation can compound across half a dozen daily workflows your buyer already lives in. Compare that to ChatGPT, where the citation lives and dies inside a chat window the buyer has to deliberately open. Copilot meets buyers where they already are.

How do you actually optimise for Microsoft Copilot?

There are four moves we have seen consistently lift Copilot citations across the brands we track at Bold GEO.

Move one: own your Bing Webmaster Tools account. As of February 2026 the AI Performance dashboard is in public preview and is the first piece of first-party data on AI citations any platform has shipped. It exposes total citations, average cited pages per day, the actual grounding queries Copilot used, and page-level performance. If you do not have this dashboard connected, you are flying blind. The official Bing announcement walks through the setup.

Move two: implement IndexNow. Microsoft built the IndexNow protocol specifically to push new and updated URLs to Bing in near real time. Copilot's live-web layer relies heavily on the freshest version of a page, so freshness is the single biggest under-rated lever for Copilot performance, and IndexNow costs roughly half an hour of engineering to wire up.

Move three: structure your pages for the synthesis step. Every commercial page should have a clear H1 that names the entity, an H2 that defines what you do, a comparison or specification table that gives the model a clean fact to attribute, and an FAQ block that maps to grounding-query phrasing. We covered the deeper version of this in our post on structured data for LLMs, and the same structural moves apply twice as hard to Copilot.

Move four: earn third-party authority. Copilot weighs co-citation heavily, so a mention in a Forrester report, a G2 review, or a Microsoft Reactor session compounds faster than three of your own blog posts. Review aggregators (G2, TrustRadius, Capterra) all sit deep in the Bing index and are reliably retrieved. So is LinkedIn Pulse content from credible authors. So is your own About page, which is why we keep telling people to fix their About page before doing anything else.

A short word on what does not work. Stuffing keywords into your meta description does not work. Hiding fact-dense paragraphs at the bottom of a long blog post does not work. Putting your value proposition only inside an image does not work, because Copilot's retrieval layer cannot read it. These are the same Bing-era SEO mistakes most teams already know to avoid, but they kill Copilot performance faster because the synthesis step is intolerant of ambiguity.

The Copilot vs ChatGPT difference for marketers

ChatGPT and Microsoft Copilot share Microsoft's investment, but they are not the same product, and the marketing playbook diverges in three ways.

Distribution. ChatGPT lives inside chat.openai.com, the OpenAI apps, and embeds. Copilot lives inside Word, Outlook, Teams, PowerPoint, Edge, Windows search, LinkedIn, and Microsoft 365 Copilot for Sales. If your buyer is using a Microsoft stack at work (a category that includes most of the Fortune 500), Copilot will reach them more often than ChatGPT.

Source preference. ChatGPT, in browse mode, leans on Bing too, but its non-browse responses lean on the pre-training corpus and the SearchGPT index. Copilot leans almost entirely on the live Bing index, which is more aggressive about freshness and gives a meaningful edge to brands that publish, update, and ping IndexNow on a real cadence. Search Engine Land covered the November 2025 shift in detail when Microsoft brought AI Search into Copilot with emphasis on citations.

B2B specificity. Microsoft Graph integration means Copilot has line of sight into the actual documents inside an enterprise. That is not a citation source you can directly optimise, but it is a source you can influence. If your customers are quoting you in their pitch decks and your partner co-marketing pieces sit inside their SharePoint, Copilot is going to retrieve those.

The simplest way to think about it: ChatGPT rewards brands that are loud everywhere on the web. Copilot rewards brands that are loud on the web plus deeply embedded inside the workflows of Microsoft 365 customers. The strategies overlap by maybe sixty percent. The other forty percent is the work that gets you cited on Copilot specifically.

If you want a full comparison of how the major models pick brands, our breakdowns of how ChatGPT recommends brands and how Gemini selects sources cover the rest of the stack.

Where to start this week

If you do nothing else, do three things in the next seven days. Connect your domain to Bing Webmaster Tools and switch on the AI Performance dashboard so you have a real baseline. Ship IndexNow to your blog and your top ten commercial pages so updates land in the index within minutes instead of weeks. Then run five test prompts against Copilot for the queries you most want to win, screenshot the citations, and use that as your before snapshot. Re-run the same prompts every Friday for the next month. The pattern of which sources Copilot pulls from for your category becomes clear inside three or four cycles, and the gaps it reveals are usually a more honest brief than any keyword report.

To benchmark where your brand currently stands across all five major AI assistants including Copilot, use Bold GEO's AI visibility dashboard →