How can I see which of my pages AI search engines cite?
99Visibility Research · Updated October 21, 2026 · 5 min read
Three sources show page-level AI citation data. Bing Webmaster Tools' AI Performance report lists cited pages for Copilot and Bing AI summaries. Search Console's generative AI report shows impressions by page for AI Overviews and AI Mode, with no clicks or position. AI APIs from OpenAI, Anthropic and Perplexity return the sources behind each answer you request. Prompt trackers only sample answers to prompts they chose.
The short answer: two first-party reports, one API method, one sample
No single tool shows every page every AI engine cites. You combine sources, and each one sees a different slice.
| Source | Surfaces covered | Page-level data | Main limit |
|---|---|---|---|
| Bing AI Performance report | Copilot, Bing AI summaries, select partners | Cited pages, citation counts, grounding queries | Microsoft surfaces only, no API as of August 2026 |
| Search Console generative AI report | AI Overviews, AI Mode | Impressions by page | No clicks, CTR or position |
| API source lists | Answers you request | Sources consulted and sources cited | Not the consumer app |
| Prompt trackers | Whatever prompts they run | Citations in sampled answers | A sample of prompts someone chose |
Notice the gap. No first-party report shows which pages ChatGPT cites in the consumer app. For that engine you rely on API calls and tracker samples, which is why asking whether OpenAI offers a citation report is a common next question.
Bing's AI Performance report: the only page-level citation count
Bing Webmaster Tools opened its AI Performance report as a public preview on 10 February 2026 and expanded it in June 2026. It covers Microsoft Copilot, AI summaries in Bing and select partner integrations.
It shows total citations, average cited pages, citation activity per page, and grounding queries. Grounding queries are the phrases the AI actually used to retrieve your content. No other surface exposes them.
Three limits matter:
- No API. As of August 2026 you export it by hand.
- Sampling. Microsoft says the grounding-query data is a sample of overall citation activity.
- Citation Share is not a ranking. Microsoft calls it an observational metric, not a competitive scoreboard. It can fall while your absolute citations rise.
Bing data also says nothing about other engines. In Otterly's 2026 experiment, Copilot kept citing two sites after Google deindexed them, while ChatGPT almost never cited them even though they ranked in Bing.
Search Console's generative AI report: impressions, not citations
Google's report covers AI Overviews and AI Mode. You can break it down by page, country, date and device. It reports impressions only: no clicks, no CTR, no position. The Search Analytics API does not cover it, so this data stays in the interface.
Global rollout completed around 11 August 2026. A Google-side bug depressed impressions in this report from around 12 August 2026, so check the current state before reading any trend across that window.
Google says AI Overviews and AI Mode "may use different models and techniques" and return different links. Read them separately.
If a key page shows zero, check eligibility before anything else. Google requires a page to be indexed and eligible to show with a snippet. A stray nosnippet or max-snippet:0 robots meta tag removes it. So does the Search Console generative AI opt-out, which child properties can inherit from a parent without anyone noticing.
API source lists: what an engine retrieved versus what it cited
The AI APIs return the sources behind each answer you request:
- OpenAI's web search tool returns the full list of URLs consulted, separate from citations. OpenAI notes the source list is often longer than the citation list.
- Anthropic's web search tool returns URL, title and
page_age, and cites text spans of up to 150 characters. It costs $10 per 1,000 searches plus tokens. - Perplexity's Agent API returns search results and cites whole sources by index. Web search is opt-in, so code without it gets ungrounded answers.
That split is why Claude and Perplexity citations look different, and it lets you see sources ChatGPT retrieved but didn't cite.
The limit: an API call is not the consumer app. ChatGPT rewrites your question into its own search queries and can use memory. A citation is also not proof of use. Mechanistic research repaired over 90% of missed citations by changing the model's internals, not the sources.
Why prompt trackers only sample
A tracker sends prompts it chose and counts the citations in the answers. That is a sample, not a census.
Otterly's experiment shows the size of the gap. Microsoft's AI Performance report credited one test site with 42,817 citations. Otterly's prompt-based tracker logged 3,934. The time windows may not match, so it is not an exact ratio. It does show how much a first-party count can see that a sample misses.
Trackers still earn their place. They are the only way to compare engines that publish nothing, like ChatGPT. Read their numbers as a snapshot of chosen questions, and expect retrieval to switch on and off between runs. Our guide on what a tracker can't see covers the blind spots. The free checker on this site is that kind of snapshot: up to three buyer questions, scored once.
What to do next
- Verify your site in Bing Webmaster Tools. Export the AI Performance page list and grounding queries every month.
- Open Search Console's generative AI report, filter by page, and mark the August 2026 bug window.
- If you have developer time, run a fixed question set through the APIs and log sources and citations separately.
- Treat tracker output as a sample and note the date and engine for each run.
- If important pages show nothing anywhere, check access first: robots.txt per bot, firewall responses, and content that only appears after JavaScript runs. Start with code-level issues that stop AI indexing.
- If engines cite other sites instead of yours, find the third-party pages AI engines cite.
Can I see ChatGPT citations in my analytics? Partly. ChatGPT adds utm_source=chatgpt.com to referral URLs, so analytics show which pages people clicked through to. That is clicks, not citations. Free tier and mobile app clicks often arrive untagged, so treat the count as a lower bound. Pages cited but never clicked do not appear at all.
Do AI Overviews show up as referral traffic? No. Google AI Overviews emit no referral, so analytics cannot separate those visits from regular Google traffic. Search Console's generative AI report is the only first-party view, and it reports impressions by page with no clicks. You can see that a page appeared, but not whether anyone clicked it.
Why does Bing show citations my tracker never found? Bing counts citations across real Copilot and Bing AI usage. A tracker only sees answers to the prompts it runs. Engines also rewrite questions into their own search queries, so a tracker's prompts may never trigger the retrieval real users trigger. Bing's grounding queries show what the AI actually searched for.
Does a citation mean the engine used my page? Not necessarily. OpenAI's API returns more sources consulted than sources cited, so pages can shape an answer without a citation. Research on citation mechanics found citation output can be changed without changing the sources. Treat citation counts as a noisy signal of influence, not a direct measure of it.
Is there an AI ranking position I can track by page? No. Google's generative AI report has no position metric, and Google says third-party tools lack its internal ranking data. OpenAI discloses only that ChatGPT ranks results using multiple factors. Any tool showing an AI rank has nothing to check it against. Report presence and citation counts instead.
Sources
- 2000 AI Blogs Experiment: Does Google Penalize AI Content?
- AI Features and Your Website
- Web search tool - Claude Platform Docs
- Migrate from Sonar to the Agent API - Perplexity
- Searching the web with ChatGPT
- [2606.28358] How Do LLMs Cite? A Mechanistic Interpretation of Attribution in Retrieval-Augmented Generation
This page is AI-assisted. It was written from a verified evidence base and checked against our editorial policy. Engine behaviour changes often; facts are as of October 21, 2026.