Before you write a single new page for AI search, find out which of your existing pages AI platforms already cite. Most sites have a handful of pages doing almost all the work: a pricing page Copilot keeps referencing, a guide that appears in AI Overviews, a comparison post Perplexity links to. Those pages are your proof of what works, and the pages that never get cited are your gap list.
An AI citation audit is a structured way to collect cited URLs across AI platforms, map them back to your site and decide what to do next. In 2026 it is far easier than it was a year ago, because both Google and Microsoft now report AI data directly to site owners.
This guide shows where to collect citation data, how to build the audit sheet, how to classify pages and how to turn the results into a short list of quick wins.
Key Takeaways
- Combine three evidence types: first-party reports (Google Search Console, Bing Webmaster Tools), referral data (GA4) and prompt sampling (manual or with a tool).
- Google's Generative AI performance report shows which pages got impressions in AI Overviews and AI Mode, but not the queries behind them.
- Bing's AI Performance report shows citations, cited pages and sampled grounding queries for Copilot and Bing AI summaries.
- Classify every important page as Cited, Visited-not-cited, Competitor-cited or Invisible, then prioritise.
- Quick wins are usually pages that are close to being cited: indexed, relevant and visible in classic search, but missing a clear, quotable answer.
Where Citation Data Comes From
| Source | What you get | What you do not get |
|---|---|---|
| Search Console: Generative AI performance report | Impressions for pages shown in AI Overviews and AI Mode, by page, country, date, device; exportable | Queries, clicks, position |
| Bing Webmaster Tools: AI Performance | Total citations, cited pages, sampled grounding queries, citation share per query; exportable | Clicks or views |
| GA4 (AI Assistant channel, landing pages) | Pages that received visits from ChatGPT, Gemini, Copilot, Claude and similar assistants | Citations that were never clicked |
| Manual prompt sampling | Cited URLs (yours and competitors') for prompts you choose, on any platform | Scale; results vary between runs |
| Prompt-tracking tool | Cited domains and URLs across many prompts and platforms over time | Certainty; data is sampled |
“Impressions are how many times links to your site were shown to a user in a generative AI feature on Google Search.”
— Google Search Console Help, Generative AI performance report (Search)
Note what that definition says: links were shown. It is the closest thing Google offers to a citation count, and it is enough to tell you which pages Google's AI features rely on. For more on this report, see our post on Search Console AI data.
Step-by-Step: Run the Audit
- Export Google AI pages. In Search Console, open the Generative AI performance report, choose the last three months and export the Pages table. Remember the 1,000-row limit; large sites should filter by folder and export several times.
- Export Bing citations. In Bing Webmaster Tools, open AI Performance and export cited pages and grounding queries. Since a June 2026 update, you can click a grounding query to see which of your pages were cited for it.
- Export AI referral landing pages. In GA4, filter the Landing page report to the AI Assistant channel and export landing pages with sessions and key events. Our guide to AI referral landing pages explains how to read this report.
- Sample prompts manually. Pick 30–50 prompts across your main topics. Run each on ChatGPT, Perplexity, Google AI Mode, Gemini and Copilot. Record every cited URL, yours and competitors'. Run each prompt at least twice, because answers vary.
- Normalise URLs. Strip parameters such as
utm_source=chatgpt.comand any text fragments (#:~:text=) so variants of the same URL are counted once; see text fragment links. Map each URL to its canonical version. - Merge into one sheet. One row per page, with columns for Google AI impressions, Bing citations, AI referral sessions, manual citations and competitor URLs cited for the same prompts.
- Classify and prioritise using the table below.
Classify Every Page
| Class | Signal | What it usually means | Action |
|---|---|---|---|
| Cited | AI impressions or citations, plus some referrals | Page answers a real question clearly | Protect: keep facts fresh, do not restructure without reason |
| Cited, low clicks | Many impressions or citations, few referrals | The answer is complete in the AI response | Add next-step value: tools, pricing, booking, downloads |
| Competitor-cited | Competitor URL cited for prompts where your page exists | Their page is clearer, more specific or better supported | Compare side by side; fill the gap |
| Third-party-cited | A review site, directory or forum cited instead of any brand | AI prefers independent sources for this question | Earn presence on those sources |
| Invisible | No AI data at all, despite organic rankings | No quotable passage, thin content or not indexed | Check indexing, then rewrite the key section |
The "third-party-cited" class is common for comparison and "best" prompts. Our guide on why AI cites other sites instead of yours covers how to respond.
Spotting Quick Wins
Quick wins are pages that should be cited but are not. Look for pages that tick most of these boxes:
- Indexed and ranking on page one in classic search for the related query.
- Relevant to prompts where competitors are cited.
- Answer buried below a long introduction, or spread across several sections.
- Key facts (prices, dates, specifications, steps) shown only in images or scripts rather than text.
- Out-of-date numbers that a competitor has refreshed.
The fix is usually editorial, not technical. Move the direct answer to the top of the relevant section (see answer-first writing), add a short table or list where it helps, update facts and make sure the page is internally linked. Google's own guidance is that no special AI markup is needed.
“You don't need to create new machine readable files, AI text files, or markup to appear in these features.”
— Google Search Central, AI features and your website
Common Mistakes
- Relying on one source. Google, Bing and prompt sampling each show a different slice.
- Running each prompt once. A single run can mislead; repeat and record how often each URL appears.
- Counting syndicated copies as your citations. If a partner republishes your article, the citation may go to them; see syndicated content and AI citations.
- Rewriting cited pages heavily. If a page is already cited, change it carefully and measure the effect.
- Ignoring the low-click pattern. A page cited often but rarely clicked is still valuable for brand visibility; judge it accordingly.
Repeat the audit quarterly, and fold it into your broader AI search audit and regular content audit. For background on how platforms pick sources in the first place, read where AI gets its answers.
Related Guides
- AI Search KPIs: The Metrics to Report Each Month
- Detecting AI Agent Visits in Your Analytics
- AI Search for Local Businesses: A Practical Starter Plan
Frequently Asked Questions
What is an AI citation audit?
It is a review of which pages on your site AI platforms such as Google AI Overviews, AI Mode, Copilot, ChatGPT and Perplexity cite or link to, compared with competitors, so you can protect what works and fix gaps.
Can Search Console show which queries trigger AI citations?
No. Google's Generative AI performance report shows impressions by page, country, date and device, but not queries. Bing Webmaster Tools does show a sample of grounding queries for Copilot and Bing AI summaries.
How often should I run a citation audit?
Quarterly works for most sites, with a lighter monthly check of first-party reports. Run one after major site changes or content updates too.
Do I need a paid tool to audit AI citations?
No. You can start with Search Console, Bing Webmaster Tools, GA4 and a manual prompt sample. Tools help when you need hundreds of prompts, several markets or regular competitor tracking.
Why does a competitor get cited instead of my page?
Usually their page answers the specific question more directly, has more specific facts, is more up to date or is backed by more independent sources. Compare the two pages side by side.
Start With What Already Works
An AI citation audit shows you which pages already earn trust from AI platforms and which pages are close. Protect the first group, improve the second and earn third-party mentions where AI prefers independent sources. If you would like us to run the audit and turn it into a content plan, explore our SEO services or get a quote.
References
- Generative AI performance report (Search) – Search Console Help
- Introducing AI Performance in Bing Webmaster Tools Public Preview – Bing Webmaster Blog (February 2026)
- Bing's improved AI Performance section – Practical Ecommerce (June 2026)
- [GA4] Default channel group – Google Analytics Help
- AI features and your website – Google Search Central
- How to Track LLM Prompts in 3 Steps – Semrush Blog
