Microsoft Copilot Optimisation

Microsoft Copilot generates answers by retrieving content from the Bing index. The optimisation principles overlap closely with ChatGPT Search, for the share of its web answers Microsoft says Web IQ serves, though Copilot’s dependence on the Bing index is far more complete. Copilot also has distinct integration points, a B2B-weighted audience, and the only first-party citation measurement tool available from any major AI search platform. Copilot is one surface within the wider Microsoft search ecosystem, so the visibility foundations are the same ones that govern Bing itself.

How does Copilot retrieve content?

When a user asks Copilot a question, Copilot does not simply search for that exact query. It generates a set of internal reformulations, which Microsoft calls grounding queries and reports in Bing Webmaster Tools as “the internal queries Copilot generated that led to your content being retrieved”,¹ and retrieves passages from pages in the Bing index. A user asking “what’s the best approach to international hreflang” might trigger grounding queries like “hreflang implementation guide,” “hreflang errors common mistakes,” and “international SEO hreflang best practices.”

If the reformulations run as wide as that example, a page covering a topic across several angles has more chances to match one of them than a page written to a single exact phrase. The AI Performance report is where to check what the grounding queries for your pages actually were.

The Bing index is the entry point

Copilot cannot cite content that is not indexed in Bing. A page that ranks well in Google but has not been submitted to, or crawled by, Bing is invisible to Copilot. The first step for any Copilot optimisation effort is confirming that key pages are indexed in Bing.

Microsoft has reframed what that index is for, and the reframe explains the grounding queries above. Its May 2026 post on the evolving role of the index puts it directly: “Traditional search asks: which pages should a user visit? Grounding asks: what information can an AI system responsibly use to construct a response?”² The index is being asked to supply evidence for an answer rather than candidates for a click, which is why a page’s rank for the user’s literal question matters less here than whether its sections answer the reformulations cleanly.

Microsoft’s Web IQ, the grounding API built on the same index, makes the unit explicit: it uses “passage-level retrieval”, and describes its aim as higher grounding satisfaction in domains “where source quality, freshness, and attribution directly impact decision-making”.³ That passage-level selection reaches beyond Copilot, because Web IQ also serves some of ChatGPT’s web answers.

Submit your sitemap through Bing Webmaster Tools and use the URL Inspection tool to verify indexing status for priority pages. The Bing Webmaster Tools guide covers this in full.

Content formatting

The unit Microsoft has documented is the passage: Web IQ, built on the same index, uses “passage-level retrieval”.³ Microsoft has also given formatting advice, though not measurements. In an October 2025 post on the Microsoft Advertising blog, Krishna Madhavan, a principal product manager at Microsoft Bing, wrote that “Assistants like Copilot break content down, a process called parsing, into smaller, structured pieces”, and recommended strong headings, question-and-answer pairs, lists, tables and “self-contained phrasing”.⁴ The post gives no data for any of it, so the advice below rests on the reader first and on Microsoft’s description second.

Answer first. Place the core answer in the first one to two sentences of each section, not at the end of a long explanation. A section that opens with its answer is a self-contained passage; one that builds to it is not.

Question-formatted headings. H2 and H3 headings phrased as questions (“How does Copilot choose which sources to cite?”) tell a scanning reader what each section answers, and match the question-shaped queries a conversational interface receives. Microsoft’s post says headings “act like chapter titles that define clear content slices”, and that assistants “can often lift” question-and-answer pairs “word for word” into answers;⁴ heading hierarchy covers what Google and OpenAI do and do not document about headings.

One point per paragraph. A paragraph covering a single claim reads as one unit; a discursive paragraph covering several is one unit to a reader and several to anything that quotes it.

Summary blocks on longer pages. A short summary section near the top of a long article gives a reader the answer without the scroll. Whether Copilot prefers it to body content is not documented.

Structured data

Microsoft is the one AI search operator that has said schema helps at all, and its statement is narrower than the trade coverage of it. On the Bing Webmaster Blog in November 2025, Microsoft’s Fabrice Canel and Krishna Madhavan wrote that “structured content, like schema-marked product pages, FAQs, and comparison tables, helps AI systems interpret and summarize your content more effectively”, adding that this “increases the chances of your being cited”.⁵ The subject is structured content, with schema as one form of it, and no type is named as a requirement. The widely repeated “Microsoft confirms schema helps its LLMs” from SMX Munich in March 2025 was a LinkedIn summary of Canel’s talk; the only words attributed to him there are about freshness, “Gen AIs value fresh content in particular, partly as a reference check of their LLM training data”, with IndexNow recommended to push updates.⁶

FAQPage schema no longer produces rich results in Google Search (removed May 2026). Whether any AI system reads it as an extraction aid is undocumented; if the questions are on the page as headings with the answers beneath them, the markup adds a machine-readable copy of what is already visible, at no cost.

Schema.org entity markup (Organisation, Person, Product) is where the effort goes. Keep it consistent with the visible page, the About page and your LinkedIn organisation name; that consistency is the part that holds whatever any LLM does with the markup.

LinkedIn and B2B queries

Copilot draws on LinkedIn content for professional and B2B topics. Profound measured 1.4 million citations across six engines including Copilot, over 15 November 2025 to 15 February 2026, and found LinkedIn the most-cited domain for professional queries.⁷ That finding is cross-platform rather than Copilot-specific: it comes from a synthetic prompt basket of Profound’s design, and the study publishes no per-engine breakdown, so nothing in it isolates a Copilot effect, and its window closes before the decline in LinkedIn’s citation share covered on LinkedIn search. Microsoft’s ownership of LinkedIn and Copilot’s integration with Microsoft 365 are plausible reasons to expect a stronger effect here than elsewhere, but no published measurement tests it.

What follows from that is narrower than it looks. If you publish on LinkedIn, long-form articles are the format that gets cited: two independent panels agree on the hierarchy of articles over posts over profiles, and LinkedIn search sets out both. But the hierarchy tells you what to write, not whether to be there. LinkedIn’s citation share is set by retrieval arrangements you cannot see and has moved sharply in both directions inside a year, so publish there for the audience and referral traffic it earns you directly and treat any Copilot citation as upside.

The entity consistency is the part that holds regardless of who is citing what: the organisation name on LinkedIn, the site’s About page, and Schema.org Organisation markup should all match.

Measuring with the AI Performance report

Since February 2026, Bing Webmaster Tools includes an AI Performance report.¹ Microsoft scopes it to “Microsoft Copilot, AI-generated summaries in Bing, and select partner AI experiences”, so the totals are not Copilot’s alone.¹⁸ It shows:

• Total citations: how many times your pages have been cited across those surfaces during the selected period
• Grounding queries: the internal queries Copilot generated that led to your content being retrieved
• Cited pages: which specific URLs are being cited and how citation volume is trending

A June 2026 update added further dimensions: intent labels, topic groupings, citation share (your citations as a proportion of the total for a topic), and a compare view for period-on-period analysis.⁸

As at October 2026 this is the only first-party citation measurement tool available from any major AI search platform. Google Search Console has carried a dedicated Search Generative AI performance report since June 2026, available to all websites worldwide from 31 August 2026, but it reports impressions by page, country, device and date rather than citations, and carries no query-level data.⁹

The report shows citation volume and trend data, without click-through data from citations.

Copilot’s integration points

Copilot surfaces in the Edge sidebar, as a standalone product at copilot.microsoft.com, within Windows search, and throughout Microsoft 365 applications. Each context brings a different audience and query type.

Edge sidebar queries arrive while a user is already reading a webpage; Microsoft 365 queries arrive inside a work task: summarising documents, answering work-related questions, generating drafts. Microsoft publishes no breakdown of what each context cites, and the AI Performance report does not split by surface, so the difference in audience is documented and the difference in citation behaviour is not.

Plan for the audience, then. Reference material that answers a work question is the content a Microsoft 365 user is most likely to need; a consumer page has a smaller share of that audience whatever the retrieval does.

Footnotes

1. Introducing AI Performance in Bing Webmaster Tools — Microsoft ↩ ↩² ↩³

2. Evolving role of the index: From ranking pages to supporting answers — Bing, 6 May 2026 ↩

3. Microsoft Web IQ ↩ ↩²

4. Optimizing Your Content for Inclusion in AI Search Answers — Microsoft Advertising, Krishna Madhavan, Principal Product Manager, Microsoft Bing, 8 October 2025. ↩ ↩²

5. How AI Search Is Changing the Way Conversions are Measured — Bing Webmaster Blog, Fabrice Canel and Krishna Madhavan, 20 November 2025: “Structured content, like schema-marked product pages, FAQs, and comparison tables, helps AI systems interpret and summarize your content more effectively.” ↩

6. Microsoft Bing/Copilot use schema for its LLMs — Search Engine Land, 20 March 2025, relaying David Mihm’s LinkedIn account of Fabrice Canel’s SMX Munich talk; the freshness sentence is the only wording attributed to Canel. ↩

7. LinkedIn is the most-cited domain for professional queries in AI search — Profound, 9 March 2026, covering 15 November 2025 to 15 February 2026. Two Profound datasets: the ChatGPT rank movement comes from real user prompts; the “#1 for professional queries” finding from a synthetic prompt basket of Profound’s design, run across six engines, 1.4 million citations. ↩

8. New AI Visibility Insights in Bing Webmaster Tools: Intents, Topics, Citation Share, Compare — Microsoft ↩ ↩²

9. Introducing Search Generative AI performance reports in Search Console — Google Search Central. Carries Google’s note: “As of August 31, 2026, we’ve rolled out these insights to all websites worldwide.” ↩