**TL;DR** — In April 2026 we ran the same 60-query basket through Bing Copilot that we run through ChatGPT, Perplexity and Gemini. Across our cohort Copilot is now the fourth-largest source of AI-search referrals — small in absolute terms (about 11% of AI-search sessions) — but its conversion rate beats ChatGPT and the citation overlap with the other three engines is below 25%. If you are not auditing Copilot specifically, you are leaving the cheapest unique-citation opportunity on the table.
Why most teams skip Copilot — and why that is a mistake
Bing's market share never recovered to the levels it had in 2018, and most SEO programs stop at "we will inherit some Bing traffic if we do Google well." That heuristic was reasonable in 2022. It is not reasonable in 2026. Microsoft has wired Copilot into Edge, Windows 11 search and the default new-tab experience — and Copilot's citation panel pulls from a ranking layer that materially diverges from Bing classic. We logged 60 queries weekly through April; Copilot's cited URL set overlapped with Bing classic's top 10 only 41% of the time. The two surfaces share infrastructure, not output. Treating them as the same lane in your audit is the same mistake as treating Google organic and AI Overviews as the same lane was in 2024.
The under-investment shows up in our competitive teardowns. Across 14 client engagements where we audited the competitor set on Copilot, three of the top five Google-organic competitors per query were absent from Copilot's citation panel entirely. The pages exist, they index, they just do not get cited. The composer is choosing differently — and the gap is exploitable. Two of those engagements involved clients who, with no new content work, gained meaningful Copilot citation share inside a quarter just by re-shaping pages they already had. The work was not finding new keywords; it was reading what Copilot was already willing to read.
What gets cited on Copilot
Three patterns showed up consistently across the audit. First, Copilot is the most aggressive of the four engines about citing primary sources — vendor documentation, official specs, government data and well-known foundation publications. 58% of its cited URLs were on domains that meet a strict "primary source" definition, vs. 31% on ChatGPT and 22% on Perplexity. If your page synthesises other sources without itself citing primary upstream, Copilot is the engine that will skip you most often. Adding two or three primary-source citations to a page that previously had none was, in our sample, the single highest-leverage edit for moving Copilot share.
Second, Copilot rewards structured content more than the others. Pages with strong heading hierarchy, definition lists and tables of comparable data showed up in Copilot's citation panel disproportionately. We pulled the top 50 cited URLs across the basket and 39 of them had at least one HTML `<table>` element with comparable data. The signal is not unique to Copilot — Google rewards it too — but the magnitude on Copilot is roughly twice as strong. The composer appears to read structured comparisons as "answer-ready," and a page that buries its comparison inside prose loses citations to a page that puts the same numbers in three rows of a table.
Third, Copilot has a noticeable preference for Microsoft-adjacent surfaces and the broader LinkedIn graph. LinkedIn articles, GitHub READMEs, MS Learn pages and the Microsoft documentation domain show up far more on Copilot than on the other three engines. None of these signals is news on its own; the combination is what makes Copilot's citation profile distinct enough to warrant its own audit lane. Practically, if your authors have a strong LinkedIn presence and any code-adjacent GitHub footprint, Copilot is the engine where that footprint pays back the fastest.
The conversion shape is the surprise
Here is the surprise from 60 days of GA4 data. Copilot referrals had a 5.8% conversion rate across our cohort — higher than ChatGPT (3.6%), comparable to Perplexity (5.9%) and well above Google organic (2.7%). The volume is small (11% of AI-search sessions in our sample), but the qualified-pipeline share inside that 11% is roughly 19% of total AI-search-attributed pipeline. It is a high-margin trickle, and the margin only shows up when you separate it from Bing classic in the report.
The user shape matters here. Copilot users skew heavily toward enterprise — Edge as default browser, Windows 11 deployment, Microsoft 365 integration. That means the click that lands on your page is more likely to be an evaluator inside a procurement workflow than a random consumer. We saw this most clearly on the B2B SaaS clients in our cohort: Copilot-attributed sessions had a 2.3× longer median session duration than Google organic and were 1.8× more likely to view the pricing page within the first three pageviews. For consumer DTC categories the multiplier shrinks — the Copilot user is still skewed enterprise even on consumer sites — but the absolute volume drops to a level where the audit work is harder to justify.
How to actually audit yourself on Copilot
The mechanical steps are quick, but the discipline matters. Run the same 30-query basket you use for ChatGPT, Perplexity and Gemini through Copilot weekly — use Edge with a clean profile to avoid personalisation noise, sign into a baseline Microsoft account if you want enterprise-shape results, and capture the citation panel for each query. We use a small Playwright script to automate this and dump the results into a spreadsheet alongside the other engines. The cross-engine matrix should now have four columns, not three, and the report should refuse to combine them.
When you find a query where Copilot cites you and the other engines do not, treat it as a "Copilot-only win" — those URLs are usually structurally distinct (more tables, clearer headings, primary-source citations) and worth pattern-matching across other underperforming pages. When you find a query where the other three engines cite you and Copilot does not, the fix is usually one of three: a missing `<table>` of comparable data, a missing primary-source citation in the body, or a heading structure that hides the answer below an H3 the composer does not reach. None of these are heavy lifts, and all of them improve the page for the other three engines too.
What changes in the weekly report
We added a fourth column to the AI-search section of every client weekly: ChatGPT, Perplexity, Gemini and Copilot, separately. The conversion-rate gap and the per-engine pipeline split mean a combined "AI search" total hides the cheapest channel inside it. We also pulled Copilot referrals out of GA4's default `bing` bucket using the channel-grouping setup we wrote about last week — without that, Copilot looks like classic Bing organic and the conversion-rate signal gets attributed to the wrong slot. The change took an afternoon; the resulting clarity in the weekly review has been worth a quarter of arguments about why the AI total looked smaller than it should.
- 01Run your 30-query basket through Bing Copilot weekly with a clean Edge profile. Capture the citation panel separately from Bing classic; the two diverge by ~60% and should not share a column in your report.
- 02Audit your top 20 commercial pages for `<table>` of comparable data and primary-source citations. These are the two structural levers that move Copilot citation share fastest in our sample.
- 03Add Copilot as its own GA4 channel using the same custom-grouping pattern as ChatGPT, Perplexity and Gemini. Default GA4 lumps it with Bing organic and the conversion-rate signal is buried.
- 04For B2B SaaS specifically, treat Copilot citation share as a leading indicator of enterprise pipeline. The session-duration and pricing-page-view multipliers in our cohort are too large to ignore.
Where this argument breaks
Copilot's citation preferences shift faster than the other three engines' — Microsoft has shipped four visible composer changes in the past six months, and the patterns above are accurate as of April 2026 but will likely move by Q3. For consumer DTC categories where the audience is not on Edge or Windows 11, the absolute Copilot volume sits below the noise floor and the audit work is overhead. For Chinese-language markets, Copilot's coverage is thin enough that the audit does not currently pay; we revisit that quarterly. Outside those carve-outs, Copilot is the unmonitored fourth engine that most clients did not know they were already getting traffic from — and the small absolute number on the chart hides a disproportionately qualified mix.