AI Visibility Tracking in 2026: What Google, Bing and GA4 Actually Show You (And What They Don't)

Marketing agencies and business owners have been reaching out to me for a while now, always with the same question: how do we measure a website's or a brand's citability across artificial intelligence platforms?

The honest answer has always been the same: there's still no single, 100% infallible platform for this, because this space is barely consolidating — it's been less than a year since it took on critical weight in digital strategy.

I want to write this without selling anything. This isn't an article meant to get you to hire me. It's so that the next time a client asks you "so how do we know if AI is citing us?", you have the full answer — not the one that sounds good on a call, but the one that's verifiable.

The mistake of looking for a magic wand

With traditional SEO, we never expected a single metric to explain everything. We know a whole website doesn't rank as a block: some pages win, others don't; some keywords bring traffic, others stay buried on page five. We've lived with that fragmentation from the start.

Something similar happens with conversational engines, except we haven't fully accepted it yet. Every console — Google, Bing, Google Analytics 4 — shows you a slice of the picture, not the whole picture. Before building any measurement strategy, it helps to start from that. If you're looking for the one console that gives you everything, you'll be waiting for something this ecosystem, as it's built today, doesn't offer yet.

A point that's often overlooked: the mention without a link

Not every AI citation comes with a hyperlink. Frequently, ChatGPT, Gemini or Copilot name your brand, your company or your content in text form within the answer, without linking to your site.

That doesn't mean nothing happened. That mention builds brand authority and user trust, even though no traditional analytics tool records it as a visit. Counting only the visits that arrive from AI platforms is a legitimate, verifiable metric — but a partial one: it misses the cases where a model already recognized you as a valid answer without the user ever reaching your site. Both signals matter, and they're worth measuring separately instead of treating them as one thing.

What each platform actually tells you (and what it still doesn't)

It's worth being precise here, because a fair amount of content circulating about these tools is based on data that's already changed, or was never entirely accurate to begin with. Here's what I confirmed, platform by platform, against official documentation and real data from an account I manage.

Google Search Console

Google launched its "Generative AI Performance Report" on June 3, 2026, first to a group of UK-based sites, and announced worldwide availability on August 31, 2026. It shows impressions within AI Overviews, AI Mode and generative features in Discover, broken down by page, country, device and date.

There are two important limits to keep in mind. First, the report doesn't include clicks, CTR, position or queries — only impressions. Second, it only covers Google's own ecosystem: it shows visibility within AI Overviews and AI Mode, but gives no data at all on whether ChatGPT, Claude or Perplexity are citing your content. And one more clarification worth having on hand: even though Google announced the global rollout, Search Console's own help documentation states that access still isn't available to every property — it's a progressive rollout, not an instant one. If you don't see the section in your account, that doesn't necessarily mean something is misconfigured.

Bing Webmaster Tools

Microsoft launched its "AI Performance" dashboard on February 9, 2026 — the first major search engine to offer first-party AI citation reporting — and expanded it in June 2026 with four new features: Intents, Topics, Citation Share and Compare.

A real example, from an account I manage, illustrates well what this report delivers:

(This example comes from a Chilean client's tax-and-accounting content — "SII" is Chile's tax authority; "pyme" is the local term for a small/medium business.)

It's a genuinely useful report: it tells you how many citations you got, which query generated them, what topic they belong to, and what percentage of that conversation you captured versus the competition (Citation Share). The example above shows how a single tax-related topic accumulated over 1,300 citations in one period.

But there's an important correction to make here, against a fair amount of content circulating on this: Bing's report doesn't indicate which specific AI generated each citation. Looking at the exact export columns — "Grounding Query," "Intent," "Topic," "Citations," "Citation Share" — none of them distinguish Copilot from an AI summary inside Bing itself or from a partner integration. It's all grouped into a single number.

GA4

Google Analytics 4 is, so far, the tool that distinguishes most precisely which AI platform a visitor came from — as long as that visitor actually clicked through. With channels properly configured, domains like chatgpt.com, claude.ai, perplexity.ai, copilot.microsoft.com, gemini.google.com or notebooklm.google.com show up as distinct referral sources — but with an important caveat. When someone clicks a link inside a Google AI Overview, the referrer that reaches the site is identical to a normal organic search — google / organic — and GA4 has no technical way to tell an AI Overview click apart from a traditional search result click. A study across more than 50,000 AI Overview events measured an average misattribution rate of 22.4%. Even the most precise of the three tools has a blind spot, and it happens to sit inside the largest search engine in the world.

Third-party tools: a wider picture, with their own limits

There's an additional layer of tools — Ahrefs with its "Brand Radar," Semrush with its AI Visibility Toolkit, Ubersuggest and other well-structured platforms — that do let you break down mentions and citations by individual AI platform: ChatGPT, Google AI Overviews, Google AI Mode, Gemini, Perplexity, Copilot, Claude and Grok, depending on the tool.

These platforms meaningfully complement what Google, Bing and GA4 don't cover on their own. But it's worth understanding how they arrive at those numbers, because they aren't infallible either. Each one builds a set of prompts — some drawn from their own database of real user queries, others combining keywords with AI-generated questions — preconfigured by niche and topic, and in most cases adjustable to your own criteria. In other words: the result depends on how well that prompt strategy covers the real angles of your content. If the question set doesn't account for the exact way your audience actually asks, the tool simply won't detect that citation, even if it exists. They're a solid complement, not a replacement for manual auditing.

A snapshot of the pieces available today

No single column tells the whole story. Together, they offer a reasonably reliable picture.

Why knowledge graphs matter more than they seem to

AI models don't read a website the way a person does. They read knowledge graphs and semantic connections: who said what, with what authority, connected to which other trusted entities.

A personal example: an article I authored, along with content I developed for a company, ended up cited organically in financial press publications in the United States, with no PR spend involved. Out of the millions and millions of articles published in the United States, having mine chosen for citation shows the level of authority that had been built over time. It wasn't chance — it was the result of structuring information as verifiable primary data, with real authoritative sources, in self-explanatory paragraphs that a journalist or a language model can cite without fear of getting it wrong. When you build that way, citability isn't something you chase: IT COMES TO YOU.

How I measure this in practice: three layers, with GA4 as the validator

With everything above in mind, here's how I organize measurement with my clients, without relying on a single source:

Layer 1 — GA4 as validator. This is the piece that empirically confirms a user clicked an AI citation and reached the site. With its limits (the AI Overviews blind spot), it's still the hardest data point available.

Layer 2 — A "Prompt Universe" as a radar for what GA4 doesn't see. I define between 30 and 50 representative prompts, split into informational ("what is…"), transactional ("best tool for…") and comparative ("X vs Y"), simulating how an ideal client would actually ask. I run that set periodically across ChatGPT, Perplexity and Google AI Overviews/AI Mode to audit exactly where a citation shows up, with or without a link.

Layer 3 — Entity authority. I review whether the data structure, the cited sources, and the way answers are written meet the bar a knowledge graph recognizes as a trustworthy source.

The content dilemma

There's a real tension here: write a guide detailed enough to answer everything, and AI models will prefer it as a source — but the reader may also walk away with the full answer they needed, with no reason to hire anyone.

The solution isn't to hide information or write worse on purpose. It's to be transparent about the limits of theory: explain the what and the why with precision, and be clear about the point where something stops being general information and becomes a specific case that requires human judgment. That's the point that can lead to a possible conversion — possible, because it's not a guaranteed outcome.

Conclusion

Measuring citability requires building a strategy specific to this topic, grounded in real, well-founded knowledge, that combines prompt audits, GA4 validation, an understanding of knowledge graphs, and a content architecture designed with conversion in mind — as part of a larger strategy aimed at real citability results, not just isolated numbers.

In the next article, I'll walk through other platforms and updates worth knowing about, including Microsoft Clarity — a tool I feel privileged to have early access to for testing, and one I've genuinely enjoyed using.

I'm listening, I appreciate you, and let's keep building a better future, step by step, on solid, verifiable ground.