AI Citation and Referral Tracking: What Is Measurable

View this page

Five ways to measure AI visibility today: Search Console, Bing Webmaster Tools, analytics referrals, server logs and repeated prompt runs. What each shows and what it cannot.

Organisation
WavX Solutions
Telephone
+919310079927

Description

Service AI citation and referral tracking

Five things can be measured today: impressions in Google's AI features (Search Console), citations in Microsoft's (Bing Webmaster Tools), visits referred by AI tools (analytics, a floor and not a total), pages fetched for a user's question (server logs), and how often you appear across repeated runs of a fixed prompt set. WavX sets these up and reports them monthly, with their limits stated.

Discuss your SEO/GEO/AEO strategy Last updated 2 October 2026

What can be measured today

There is no single number for "AI visibility", and no report that shows every time an AI tool named your business. There are five partial measures. Four come from first-party data you already own or can get free. The fifth you produce yourself.

Source

What it tells you

Main limit

Google Search Console, generative AI performance report

How often links to your pages were shown in AI Overviews and AI Mode

Impressions only. No clicks, no queries

Bing Webmaster Tools, AI Performance

How often your pages were cited in Copilot and Bing's AI answers, and a sample of the retrieval phrases

Citations, not visits. Microsoft surfaces and unnamed partners only

Analytics (GA4)

Visits whose referrer is an AI tool

A floor. Many AI-influenced visits carry no such referrer

Server or CDN logs

Pages fetched by an AI tool on behalf of a user

A fetch is not a citation

Repeated prompt runs

The share of runs in which you are named, for questions you choose

A sample you designed, true for the day it was run

Each is described below as its vendor documents it in October 2026. WavX's tracking service sets up all five and reads them together, as part of generative engine optimisation .

Google Search Console: generative AI performance report

Google announced the report on 3 June 2026 and states that it has been available to all websites since 31 August 2026.

It counts impressions: how many times links to your site were shown to a user in AI Overviews or AI Mode.

It breaks them down by page, country, device and date, and separates text searches from image-based ones.

It has no click figure and no query dimension. Google's announcement says it is considering additional metrics.

It may be missing if your site has had too few AI impressions, or if the site has been excluded under the Search generative AI setting.

It excludes experiments in Search Labs.

Traffic from AI features is also included in the ordinary Performance report under the Web search type, where it cannot be separated from other results. So Google will tell you that a page was shown in an AI answer, and separately how many clicks the page received in total, but not how many clicks came from the AI answer. Optimising for Google AI Overviews and AI Mode covers what affects these numbers.

Bing Webmaster Tools: AI Performance

Microsoft released this report in public preview on 10 February 2026. It covers Microsoft Copilot, AI-generated summaries in Bing, and what Microsoft calls select partner integrations.

Metric

Microsoft's definition

Total citations

How many times your pages were displayed as sources in AI-generated answers in the period

Average cited pages

The average number of your unique pages shown as sources per day

Grounding queries

The key phrases the AI used when retrieving content that was cited. Microsoft says the data is a sample

Page-level citation activity

Citation counts for each URL

Citation Share (preview, June 2026)

Your share of all citations shown for a given grounding query

Microsoft is careful about what these mean. Citation counts reflect "how often pages are cited, not page importance, ranking, or placement". Citation Share "does not expose competitor domains, represent traffic share, or assign quality scores". None of the metrics is a count of visits.

The grounding queries are the most useful part. They are the system's retrieval phrases, which are different from what the person typed, but they show which subjects your pages are being used for.

Analytics: visits referred by AI tools

Google Analytics 4 has a default channel named AI Assistant. Its help page describes it as visits from "sources like ChatGPT, Gemini, Deepseek, Copilot, or Grok", assigned when the referrer matches Google's list of AI assistants. The same page states that the channel "excludes Google's AI Overviews and AI Mode".

Three things keep this number low:

A visit is only attributed when the browser passes an AI tool as the referrer. A link opened from an app, or copied and pasted, often arrives with no referrer and is counted as Direct.

A person who reads an AI answer and later searches for your name arrives as organic or direct traffic. The AI answer did the work and gets no credit.

Clicks from Google's own AI features are counted as Google organic search.

So the AI Assistant channel is a floor. It is still worth watching, because it shows which pages AI-referred visitors land on and whether they go on to enquire. WavX checks which referrers your property actually receives, adds any that Google's list does not cover to a custom channel group, and ties the visits to your enquiry or purchase events. The method is written up in how to measure AI referral traffic in GA4, and the setup is part of the Google Analytics 4 integration .

Server logs: pages fetched for a user

When someone asks an AI tool a question and the tool opens your page to answer it, the request appears in your server or CDN logs under a documented name.

User agent

Vendor's description

robots.txt

ChatGPT-User

Used "for certain user actions in ChatGPT and Custom GPTs". Not used for automatic crawling

OpenAI says the rules "may not apply", because a user started the action

Perplexity-User

Visits a page when a user's question needs it

Perplexity says it "generally ignores robots.txt rules"

Claude-User

Accesses websites when a Claude user asks a question

Anthropic says its bots honour robots.txt

These are different from the search crawlers (OAI-SearchBot, PerplexityBot, Claude-SearchBot) and the training crawlers (GPTBot, ClaudeBot), which visit on their own schedule. The AI crawler directory lists them all.

Limits to state with every log figure:

A fetch is not a citation. The tool read the page. The logs do not show whether the answer quoted or linked it.

User agents can be forged. OpenAI and Perplexity publish the IP ranges their agents use, and Anthropic publishes a list of its addresses. Counts should be checked against them.

Google is invisible here. Google documents that Google-Extended has no user agent of its own, so use of your pages by Gemini cannot be told apart from ordinary Googlebot crawling.

A CDN may hide requests. If cached pages are served at the edge, the origin server never sees the request. The CDN's logs are needed.

The first four sources show what the engines did with your pages. None shows whether an engine names your business when a buyer asks about your category. For that, you ask, many times.

Fix a prompt set. Questions a buyer would type, agreed once and then left unchanged so that months can be compared.

Run each prompt several times on each tool, in a fresh conversation each time.

Record the raw answer, the date, the tool, whether web search was used, and the sources linked.

Report the share of runs in which your business and each named competitor appeared, and how you were described.

Repetition is the whole method. SparkToro and Gumshoe had 600 volunteers run 12 prompts 2,961 times across ChatGPT, Claude and Google's AI features in January 2026. For ChatGPT and Google's AI, a prompt asked 100 times had a less than 1 in 100 chance of returning the same list of brands in any two responses, and the same order turned up roughly once in 1,000 runs. Their conclusion was that position in an AI answer is not worth tracking, and that the percentage of runs in which a brand appears is a reasonable measure when taken over many runs.

WavX therefore never reports an "AI rank". It reports a percentage, with the number of runs beside it.

What the monthly report contains

Section

Contents

Google AI impressions

Total and by page, against the previous month, with the pages gained and lost

Microsoft AI citations

Total citations, cited pages, top grounding queries

AI-referred visits

Sessions, landing pages and enquiries from AI referrers, labelled as a floor

User-triggered fetches

Requests by ChatGPT-User, Perplexity-User and Claude-User, by page

Prompt results

Share of runs per prompt and per tool, for you and named competitors, with the number of runs

Changes made

What was changed on the site or off it during the month, with dates

Reading

What moved, what probably did not move because of the work, and what cannot be told

The report is delivered as a spreadsheet or a Looker Studio dashboard. A business that wants the data in its own system can have it built as a custom dashboard.

What tracking cannot do

It cannot show the questions people typed.

It cannot attribute a sale to an AI answer when the buyer arrived later by another route.

It cannot prove cause. If prompt share rises after a change, the change may be the reason. Models and indexes also change on their own.

It does not improve anything by itself. A report that nobody acts on is a cost.

When you should not buy this

You are not yet changing anything. Measure once, in an AI visibility audit , and start tracking when there is work to track.

Your volumes are small. Opening Search Console and Bing Webmaster Tools once a month takes twenty minutes and costs nothing.

You want a ranking position. No honest source can give one.

It is worth buying when money is being spent on content, listings or PR for the purpose of AI visibility, and someone needs to know whether the spending is doing anything.

Frequently asked questions

Can I see what people asked ChatGPT before they visited my site?

No. None of the sources available to a site owner shows the question a person typed into an AI tool. Bing Webmaster Tools shows a sample of the phrases its system used to retrieve your pages, which is the nearest thing. Search Console's generative AI report has no query breakdown as of October 2026.

Why does my analytics show so few visits from AI tools?

Because analytics only counts a visit as AI-referred when the browser reports an AI tool as the referrer. Someone who reads an answer and later types your name into a search engine is counted as search or direct traffic. Google Analytics also states that its AI Assistant channel excludes Google's AI Overviews and AI Mode. Treat the figure as a floor.

Is a crawler request in my server logs the same as a citation?

No. A request from ChatGPT-User, Perplexity-User or Claude-User means the tool fetched your page while working on someone's question. It does not show that the page was quoted or linked in the answer, or that anyone visited. It is evidence of use, and it is worth counting for that reason.

How often should the prompt set be run?

Monthly is enough for most businesses, provided each prompt is run several times on each tool. A single run tells you almost nothing: in a January 2026 experiment by SparkToro and Gumshoe, a prompt asked of ChatGPT a hundred times had a less than 1 in 100 chance of returning the same list of brands twice.

Do I need a paid AI visibility tool?

Not to begin with. Search Console, Bing Webmaster Tools and Google Analytics are free and are first-party data. A paid tracker automates the prompt runs. If you buy one, check how many times it runs each prompt, and remember Google's warning that no third-party tool has access to its internal systems.

What does tracking cost?

WavX does not publish a price for GEO work. Tracking is quoted after an audit, because the cost depends on the number of prompts, tools and competitors, and on whether log access needs engineering work.

Sources

Search Console Help: Generative AI performance report (Search) · read 2 October 2026

Google Search Central Blog: Introducing Search Generative AI performance reports in Search Console (3 June 2026) · read 2 October 2026

Google Search Central: Optimizing your website for generative AI features on Google Search · read 2 October 2026

Google Search Central: AI features and your website · read 2 October 2026

Bing Webmaster Blog: Introducing AI Performance in Bing Webmaster Tools (10 February 2026) · read 2 October 2026

Bing Blogs: New AI visibility insights in Bing Webmaster Tools: Intents, Topics, Citation Share, Compare (16 June 2026) · read 2 October 2026

Analytics Help: default channel group (AI Assistant channel) · read 2 October 2026

OpenAI: overview of OpenAI crawlers · read 2 October 2026

Perplexity: Perplexity crawlers · read 2 October 2026

Anthropic: does Anthropic crawl data from the web, and how can site owners block the crawler? · read 2 October 2026

Google: common crawlers, including the Google-Extended token · read 2 October 2026

SparkToro: AIs are highly inconsistent when recommending brands or products (28 January 2026) · read 2 October 2026

Related

Google Analytics 4 integration

AI visibility audit

Optimising for Google AI Overviews and AI Mode

Generative engine optimisation (GEO) services

Build your own software — your way, your pricing.

WavX Solutions is here to create your own software in a fully custom way, built exactly how you work — with a pricing model that fits your business. Connect now and let's build it.

Contact Now helpwavx@gmail.com