How to Track Whether AI Systems Are Citing Your Business

TL;DR: You can measure whether AI systems cite your business today, for free, in tools you already have. Bing Webmaster Tools reports actual citation counts and the queries behind them. Google Search Console reports impressions in AI Overviews and AI Mode but no clicks and no queries. What to do about answers that never produce a click is the subject of our zero-click SEO strategy. GA4 needs one configuration change to stop hiding AI referrals inside Referral traffic.

The one-sentence answer: Turn on the Bing Webmaster Tools AI Performance report for citation counts and grounding queries, check whether your Search Console property has the Generative AI performance report yet, and build a GA4 custom channel group for AI referrers, which applies retroactively so you recover your history the day you create it.

Ask a business owner whether ChatGPT mentions their company and you get a guess. Ask for the number behind the guess and the conversation stops.

For a long time that was the correct place for it to stop, because nobody could tell you. That changed in February 2026, and the tooling now covers enough ground to run a real measurement program. It is uneven, the two major reports disagree about what to even count, and one of them will not show you queries at all. It is still far better than guessing.

This is the setup we run on our own site and on client sites, in the order we run it. If you want the strategy behind the measurement first, start with how to get cited by ChatGPT and Perplexity, or with generative engine optimization for the wider theory.

Prerequisites

Before starting, you need:

• A verified Bing Webmaster Tools property. Verification by DNS record, XML file, or meta tag, same as Search Console.
• A verified Google Search Console property, ideally the domain property rather than a single URL prefix.
• Editor or above on the GA4 account at the property level, which Google states is the requirement to create and edit channel groups.
• Confirmation that you are not blocking AI crawlers. This is the step people skip, and it invalidates everything downstream.

That last one deserves thirty seconds now rather than a confused month later, and it is the check almost everyone gets wrong.

Open yourdomain.com/robots.txt. The critical distinction is that each vendor runs separate crawlers for training and for search retrieval, and only the second kind governs whether you can be cited:

Two consequences worth internalizing. Blocking GPTBot and ClaudeBot is a defensible content-licensing position that costs you nothing in citations, because neither is the retrieval agent. And Google-Extended is not the lever people think it is: Google states that it "does not impact a site's inclusion in Google Search nor is it used as a ranking signal in Google Search," so blocking it does not remove you from AI Overviews.

In our site audits, a security plugin or CDN toggle switched on by default is the single most common silent cause of AI invisibility, and the owner almost never chose it. The blanket toggles are the dangerous ones, because they take the search agents down with the training agents. Check 2 of our AI SEO audit checklist is the crawler-by-crawler walkthrough.

Step 1: Turn on the Bing AI Performance report

Bing Webmaster Tools has reported AI citations since its AI Performance report entered public preview on February 10, 2026. It is the most complete free citation data available, and it covers Microsoft Copilot, AI-generated summaries in Bing, and select partner integrations.

Open Bing Webmaster Tools, select your property, and find AI Performance in the left navigation. Data begins accruing from the date the property is verified, so verify today even if you do nothing else this week.

Failure mode: the report is empty. Usually this means the property was verified recently and there is no history yet, or your site genuinely is not being cited.

Escalation: confirm the site is indexed in Bing at all by searching site:yourdomain.com in Bing. If the index is empty, the citation problem is an indexing problem and belongs in Bing's URL submission tool first. If the index is healthy and citations are still zero after three weeks, the content is being crawled and not selected, which is a content-format problem rather than a measurement problem.

Step 2: Read the four numbers that actually matter

The report gives you total citations, average cited pages, grounding queries, and page-level citation counts.

Read them as a pair, not individually. Total citations is volume. Average cited pages is breadth, meaning how many distinct URLs from your site got quoted. Volume rising while breadth stays flat means one post is carrying you, which is fragile. On our own site through May 2026, a single article accumulated 932 citations across query variants while most of the library had none.

Grounding queries are the phrases the AI used when retrieving your content, and they are the closest thing to keyword data that exists on this surface. Bing states plainly that grounding query data "represents a sample of overall citation activity," and that the metrics reflect citation frequency and "not page importance, ranking, or placement." Treat them as directional.

Failure mode: you read a citation spike as a ranking win and go looking for what you did right.

Escalation: before attributing any movement to your own work, check a second property you also manage. In our fourth month, citations fell from a 66.0 per day average to 14.4, a 78% drop, and stayed on the floor for eleven days before recovering to 60.7. The single-day move on July 27, 2026 was 50 citations across 12 pages down to 2 across 1. The obvious cause was a bulk metadata change we had shipped days earlier. It was wrong: the same collapse and recovery hit three unrelated domains on the same dates, and one of them had received no edits at all. If you have only one property, ask a peer in a different industry to check theirs before you rewrite anything.

Step 3: Check for the Search Console generative AI report

Google now reports its own version, and it is important to know what it does and does not give you.

The Generative AI performance report in Search Console shows impressions for AI Overviews and AI Mode combined. Google defines an impression as how many times links to your site were shown to a user in a generative AI feature. There are no clicks in the report, no query data, and no split between AI Overviews and AI Mode.

So Bing tells you what you were cited for. Google tells you that you appeared, somewhere, in something. Plan your reporting around Bing and use Google as corroboration, and read the impression count against how AI Overviews actually assemble an answer.

Failure mode: the report is empty in your property and you assume you have no AI visibility.

Escalation: absence of data is not absence of visibility, because the report needs sufficient impressions in generative features before it populates and it excludes sites that have opted out of those features. Fall back to Bing citations and GA4 referrals for the period, and note in your reporting that Google data was unavailable rather than reporting zero.

Step 4: Stop GA4 from hiding your AI referrals

By default, visits from ChatGPT, Perplexity, and Copilot land in Referral traffic, mixed in with every forum link and directory listing. GA4's default channel group cannot be edited, so the fix is a custom channel group.

In GA4: Admin → Data display → Channel groups → Create new channel group. Create a channel named AI Assistants, matching source against the hostnames your visitors actually arrive from, and drag it above Referral in the channel order. Ordering is the whole trick. Google's documentation states that "traffic is included in the first channel whose definition it matches given the current order of channels in the group," so a channel sitting below Referral will never receive anything.

The useful part: custom channel groups apply to your reports retroactively. You are not starting a clock, you are recovering history you already collected.

Two limits worth knowing before you build it. Standard properties allow two custom channel groups plus the default, 360 properties allow five, and any group is capped at 50 channels.

Failure mode: the channel exists but stays at zero sessions.

Escalation: check ordering first, since that accounts for most cases. If ordering is correct, the referrer may be stripped entirely, because some assistants pass no referrer on click. Confirm by looking at Unassigned traffic and direct traffic for a matching spike on days when citations moved, and if that pattern holds, treat GA4 as a floor on AI referrals rather than a count.

Step 5: Run a manual citation check

Automated reports miss the surfaces that do not report at all. ChatGPT, Claude, Perplexity, and Gemini publish no webmaster console.

Once a month, take your ten highest-intent queries, run them in each assistant, and record whether you were cited, which URL was cited, and which competitors appeared alongside you. Keep it in a spreadsheet with the date. Twenty minutes, and it is the only visibility data available for those surfaces.

Failure mode: results differ every time you run the same prompt, and you conclude the exercise is meaningless.

Escalation: variance is expected, so record presence or absence rather than position, and run each query three times, scoring two out of three as cited. You are tracking a trend across months, not a rank on a given afternoon.

Step 6: Set a baseline before you change anything

Record one row per month: total citations, average cited pages, AI referral sessions, generative AI impressions if available, and your manual check score out of ten.

Do this for a full month before making changes. Without a baseline, every subsequent number is uninterpretable, and you will end up attributing normal platform volatility to whatever you happened to ship that week.

What "working" looks like

The success criterion is not a citation count. It is this: breadth rising alongside volume, meaning more distinct URLs cited month over month, plus AI referral sessions that convert at or above your organic baseline.

Volume alone can be one lucky post. Breadth is the compounding signal, and it is the one that survives a platform-side event like the eleven-day drop above.

Where this fails

This measurement setup has three real limits, and pretending otherwise makes the reporting worse.

Bing is not ChatGPT. The AI Performance report covers Copilot and Bing's own surfaces. It is a strong proxy for how retrieval-based assistants see you, and it is not a direct measurement of OpenAI's products. On this site in August 2026, Bing counted 1,402 citations in 22 days while GA4 attributed 7 sessions to ChatGPT and none to Copilot, a gap our month five SEO report breaks down.

None of it measures influence without a click. Someone reads your name in an AI answer, does not click, and calls you two weeks later. That is the outcome the whole strategy is aimed at, and no report above captures it. The only instrument is asking on your intake form how the caller heard about you, which is why that field belongs on your forms before any of this.

The surfaces change faster than the reporting. Bing's report added intents, topics, citation share, and period-over-period comparison in June 2026, four months after launch. Anything documented here is a snapshot, and re-checking the interface is part of the monthly cadence rather than a one-time setup cost.

Frequently Asked Questions

How long until citations appear for a new site?

There is no published figure and it varies with indexing. The more useful expectation is the order of events: our own citation volume climbed while Google organic clicks were still near zero, so AI citations can precede conventional search authority rather than following it.

Can I track which competitors get cited instead of me?

Only on the manual check, which is why that column is the highest-value one in the spreadsheet. Bing's Citation Share, added June 16, 2026, reports the percentage of citations on a grounding query that went to your site, and Bing states plainly that it "does not expose competitor domains, represent traffic share, or assign quality scores to content." It tells you how much of a query you are winning without telling you who took the rest. The Compare feature shipped alongside it overlays an earlier time period on your own data, not on anyone else's.

Does the number of citations matter more than which pages get cited?

Which pages, in almost every case. A citation on a money page reaches someone deciding, while a citation on a definition post reaches someone learning. Segment your citation counts by page type or the total will flatter you.

Should I report AI citations to clients as a KPI?

Report it as a leading indicator with the volatility stated. Handing a client a citation number without the caveat that platform-side events can cut it 78% overnight sets up a conversation you will not enjoy having.

We publish our own numbers monthly, including the months they go the wrong way, because the failure cases are where the method is actually tested. If you want this measurement built on your property with the reporting attached, see ClickWerxs AI SEO or get in touch.

Sources

1. Bing Webmaster Blog, "Introducing AI Performance in Bing Webmaster Tools Public Preview" — total citations, average cited pages, grounding queries, page-level citation activity; coverage of Microsoft Copilot, Bing AI summaries and select partner integrations; grounding queries "represents a sample of overall citation activity"; metrics reflect frequency, "not page importance, ranking, or placement." blogs.bing.com (February 10, 2026; retrieved 2026-08-22)
2. Bing Search Blog, "New AI Visibility Insights in Bing Webmaster Tools: Intents, Topics, Citation Share, Compare" — Citation Share is the percentage of citations on a grounding query attributed to your site, and "does not expose competitor domains, represent traffic share, or assign quality scores to content"; Compare overlays a previous time period on the current reporting view, comparing the current 30 days against the prior 30 or a custom range. blogs.bing.com (June 16, 2026; retrieved 2026-08-22)
3. Google Search Console Help, "Generative AI performance report (Search)" — impressions only; AI Overviews and AI Mode combined; requires sufficient impressions to populate and excludes sites opted out of generative features. support.google.com (retrieved 2026-08-22)
4. Google Analytics Help, "Custom channel groups" — "traffic is included in the first channel whose definition it matches given the current order of channels in the group"; retroactive application; two custom groups on standard properties, five on 360; 50-channel cap. support.google.com (retrieved 2026-08-22)
5. OpenAI, "Bots" — OAI-SearchBot "used to surface websites in search results in ChatGPT's search features"; GPTBot "used to crawl content that may be used in training our generative AI foundation models." developers.openai.com (retrieved 2026-08-22)
6. Perplexity, "PerplexityBot" — PerplexityBot "is designed to surface and link websites in search results on Perplexity"; not used for training foundation models. docs.perplexity.ai (retrieved 2026-08-22)
7. Anthropic, "Does Anthropic crawl data from the web?" — Claude-SearchBot "navigates the web to improve search result quality"; ClaudeBot collects content that "could potentially contribute to their training." support.claude.com (retrieved 2026-08-22)
8. Google Search Central, "Google crawlers" — "Google-Extended does not impact a site's inclusion in Google Search nor is it used as a ranking signal in Google Search." developers.google.com (retrieved 2026-08-22)
9. ClickWerxs first-party analytics — 932 Bing AI citations on a single post (May 2026), reported across our AI SEO cluster; citation volume rising while Google organic clicks were near zero; citations 66.0/day (July 18-26, 2026) falling to 14.4/day (July 27-August 6) and recovering to 60.7/day (August 7-9), replicated across three unrelated domains, published in full in our month four results. Operator data, on record.

ClickWerxs sells AI SEO services built on the methodology described in this post. First-party citation figures are our own past results, not a promise of outcomes for any engagement. Platform interfaces and report availability cited here reflect the publication date and change frequently.

Kaleb Dickhaut — Founder, ClickWerxs. Kaleb built ClickWerxs from the ground up, from payment processing ISO to the Command Center platform to the AI SEO methodology the blog runs on. He has onboarded hundreds of small businesses onto payment and CRM systems. linkedin.com/in/kaleb-dickhaut