The fundamental problem with AI tracking
With classical SEO, measurement is direct: Google Search Console gives impressions, clicks, positions. Google Analytics gives traffic by source. Attribution is imperfect but robust enough to make decisions.
AI visibility measurement is fragmented. Google Search Console exposes a dedicated report in limited rollout. Bing Webmaster Tools directly measures citations across Microsoft surfaces. ChatGPT Search, Perplexity and Claude do not provide an equivalent publisher console in the official documentation reviewed.
This does not mean measurement is impossible. It means it is indirect, partial and requires a rigorous methodology to avoid measuring noise.
What is directly measurable
1. Google AI Overviews via GSC
Google announced a separate Generative AI performance report in June 2026 and is rolling it out to a subset of sites. When available, it exposes:
- Impressions
- Pages
- Countries and devices
- Dates
A verified property may not have the dedicated view yet. Report unavailability must not be recorded as zero impressions.
2. Bing Webmaster Tools for Microsoft surfaces
Bing Webmaster Tools AI Performance directly measures citations from Microsoft Copilot, AI-generated summaries in Bing and selected partner integrations. The public preview includes total citations, average cited pages, a sample of grounding queries, page-level citations and trends. It must not be presented as a ChatGPT Search measurement.
3. Server logs
Your server logs record requests that reach your infrastructure. You can measure declared requests from PerplexityBot, OAI-SearchBot, Claude-SearchBot and others, then authenticate agents when the operator publishes a method. A crawl is not an indexing or citation measurement.
What is measurable indirectly (proxies)
Proxy 1 - Branded traffic (the most robust)
If LLMs regularly cite you, your users will search for your brand on Google after seeing your name in an AI response. This "brand halo" phenomenon is measured via:
- Queries containing your brand name in Google Search Console
- Direct traffic in Google Analytics (users remember your URL)
- Brand search volume in Semrush or Ahrefs
This is currently the most defensible proxy signal for measuring overall AI visibility. It is correlated, reliable over time and easily comparable before/after an optimisation action.
Proxy 2 - Referral traffic from AI domains
Perplexity and, to a lesser extent, ChatGPT generate clicks to cited sources. This referral traffic appears in Google Analytics under the domains perplexity.ai and openai.com (or chatgpt.com). It is a direct but partial measurement: not all users click on sources.
Proxy 3 - Regular manual sampling
Define a corpus of 20 to 50 representative target queries for your domain. Regularly query (bi-monthly minimum) ChatGPT, Perplexity, Gemini and Claude on these queries. Record whether you are cited, if it is accurate, if your competitors are cited in your place.
This is laborious but irreplaceable for strategic content decisions. A shared tracking spreadsheet with your editorial team is sufficient.
Proxy 4 - Competitive share of voice
Compare your citation rate with that of your direct competitors on the same queries. If you are cited in 3 out of 10 responses and your competitor in 7 out of 10, you have a measurable visibility gap and a clear improvement objective.
Available third-party tools
| Tool | What it measures | Limitations |
|---|---|---|
| Google Search Console | Search and Discover generative features, depending on the report | Limited rollout, documented dimensions exclude queries, clicks and CTR |
| Bing Webmaster Tools | Citations across Microsoft Copilot, Bing AI and selected partners | Public preview, documented Microsoft and partner scope |
| Semrush | AI Overviews tracking for keywords | Paid, partial coverage |
| SE Ranking | AI Overviews tracking + snippet presence | Paid, Google focus |
| Ahrefs | Classical organic data (indirect proxy) | No direct AI measurement |
| BrandMentions / Mention | Brand mentions on the web (including articles about LLMs) | Indirect, web surface monitoring |
Recommended KPIs - monthly dashboard
Here is the minimum recommended dashboard, ordered by data reliability:
- GSC Generative AI impressions: month-on-month evolution by page, country, device and date when the report is available.
- Bing AI Performance citations: totals, cited pages and available grounding queries in public preview.
- Branded traffic GSC: brand query volume, evolution, new emerging brand queries.
- Referral traffic from perplexity.ai and chatgpt.com in GA: sessions, landing pages, duration.
- Manual sampling results: citation rate per engine, accuracy, competitors present.
- AI bot crawl in logs: PerplexityBot, OAI-SearchBot frequency (interest indicator).
Pitfalls to avoid
- Do not confuse impression and reading: Google defines an impression as a URL appearing in a generative feature. It does not prove that the user saw, read or clicked the link.
- Do not turn report absence into zero: record source availability, rollout status and the observed value separately.
- Do not over-index on manual sampling: LLM responses vary by user, time, phrasing. A sample of 10 queries is not statistically representative. Replicate tests.
- Do not confuse Perplexity referral traffic and Perplexity visibility: the majority of users do not click on sources. Referral traffic is the visible part of the iceberg.
- Do not attribute all branded traffic increases to AI visibility: a PR campaign, a press mention, a viral post can also increase branded searches. Cross-reference data to isolate the effect.
Frequently asked questions
What does Google Search Console's Generative AI report measure?
When available, the Search report shows impressions, pages, countries, devices and dates for generative features. Google does not document clicks, CTR or queries in this dedicated report.
Is there a native tool for measuring AI citations?
For some surfaces: Bing Webmaster Tools measures citations in Microsoft Copilot, Bing AI summaries and selected partner integrations. No equivalent first-party publisher console is documented for Perplexity or ChatGPT Search in the official sources reviewed.
How should an unavailable AI report be interpreted?
Unavailable does not mean zero. Google's report has a limited rollout and Bing's report is in public preview. Record an explicit available, unavailable or preview status before interpreting the observed value.
The expected evolution of AI tracking
The current situation - absence of a unified AI console - is temporary. The ecosystem is moving towards more transparency:
- Google is progressively rolling out dedicated Search and Discover reports
- Bing is expanding citation reporting across its AI surfaces
- First-party consoles remain incomplete and cannot be compared directly across operators
In the meantime, the most solid approach remains: build the fundamentals (content, authority, structure), measure with available proxies, iterate. Do not wait for the perfect tool to optimise.
Official sources
- Google Search Central, Generative AI performance reports, 3 June 2026
- Google Search Central, measuring generative features, updated 10 July 2026
- Bing Webmaster, AI Performance public preview, February 2026