Can AI visibility tools track Microsoft Copilot answers?

There are two ways to measure Copilot, and they measure different things.

• Prompt sampling. A tool sends Copilot a list of buyer questions and records whether your brand or URL shows up. It sees the answer text, so it can catch wrong facts. It only sees the questions it asked.
• Bing Webmaster Tools' AI Performance report. Microsoft counts how often Copilot and Bing's AI summaries cite your pages. It sees every citation it samples across real usage, but not the answer wording.

Neither replaces the other. A prompt tracker tells you what Copilot says. Microsoft's report tells you how often Copilot leans on your pages and for which queries. Copilot grounds its answers in Bing's index, which is why Bing data matters here. For the mechanics, see how Copilot grounds answers in Bing.

A prompt tracker is a controlled survey. You pick the questions, the tool asks them on a schedule, and you get mention and citation rates for that set.

That is useful, with three limits:

• Your prompts are not the real queries. AI assistants rewrite what users type. OpenAI documents that ChatGPT search rewrites a prompt into one or more targeted queries before searching, and Copilot's grounding queries show the same pattern. Your prompt list is a proxy for an input you cannot see.
• Results vary by run. Retrieval is decided per query, so the same prompt can return a grounded answer one day and a different one the next.
• It is a sample. Coverage depends on how many prompts you track and how often.

The strength is the answer text itself. Only a prompt tracker shows you that Copilot named a competitor, or stated your old pricing. When it does, fixing wrong facts in Copilot starts with your own site.

Microsoft launched the AI Performance report in public preview on 10 February 2026 and expanded it on 16 June 2026. It covers Microsoft Copilot, AI-generated summaries in Bing, and select partner integrations.

It exposes:

• Total Citations and Average Cited Pages
• Page-level citation activity
• Grounding queries, which Microsoft describes as the key phrases the AI used when retrieving content that was referenced in answers
• Intents, Topics, Citation Share and a Compare view

Grounding queries are the rare part. No other engine shows site owners the queries its AI ran. That makes them the best input for choosing which prompts to track elsewhere.

The catch: as of our August 2026 evidence review, the Bing Webmaster API has no endpoint for any of this. Its reference documentation does not mention citations, grounding or Copilot. Microsoft has indicated API access during 2026, so check the current state. Until then, any tool that uses this data needs you to export it by hand. Our Bing Webmaster Tools setup playbook walks through verification, and tracking Copilot grounding queries covers what to do with the export.

OtterlyAI, which sells AI visibility tracking, ran a public 2026 experiment with two sites of roughly 1,000 AI-generated posts each. Google deindexed both. Bing kept them, and Copilot kept citing them.

For one site, Otterly's prompt tracker logged 3,934 citations across seven engines, 2,718 of them from Copilot. Microsoft's AI Performance report credited the same site with 42,817 citations, more than ten times the tracker's total.

The time windows may not match, so this is not an exact ratio. It still shows what a prompt sample is: a slice. Where an engine publishes its own count, that count sees more. Expect your tracker and Bing to disagree, and read why Bing's citation counts differ from tracking tools before treating either as wrong.

The same test carried a second lesson. Ranking in Bing produced effectively zero ChatGPT citations, even with ChatGPT's crawlers visiting. Measure each engine separately.

• Citation Share is not a leaderboard. Microsoft calls it an observational metric, not a ranking system. It does not expose competitor domains or traffic share. It can fall while your absolute citations rise.
• Grounding queries are sampled. Microsoft scopes its sampling note to that metric. Do not assume the whole report is sampled.
• Cited is not the same as used. Engines retrieve more pages than they cite, so citation counts undercount influence.
• Zero citations can be an access problem. Bing documents that content marked data-nosnippet is still indexed but excluded from snippets and AI summaries. A template that applies it by accident removes you quietly.
• No rank exists. Neither method gives a position inside an answer. Treat any "Copilot rank" number with suspicion.

1. Verify your site in Bing Webmaster Tools and confirm your key pages are indexed.
2. Export grounding queries monthly. Use them to build or prune your prompt list.
3. Run a prompt tracker on that list to read Copilot's actual wording about you.
4. Check that bingbot can reach your pages through your CDN or firewall. The AI Crawler Checker requests a page as each AI crawler and shows which ones get turned away.
5. Compare trends, not totals, between the two sources.

What you control is access, indexing and the facts on your own pages. What Copilot chooses to say is not in your hands, and no tool can promise a citation.