Track ChatGPT, Gemini, Claude & AI Overviews

Six surfaces, not one

"AI visibility" is usually discussed as though there were a single place answers come from. There is not. A buyer asking a question in Google's AI Mode, in the Gemini app, in ChatGPT and in Perplexity is using four different systems, with different indexes, different retrieval behaviour and different publisher controls. A result from one is not evidence about another.^1,2,3

That distinction decides what you can measure, what you can control, and what a number in a report actually means. This page goes surface by surface and says plainly which are observable, which are not, and what Robot Visible does and does not check.

The API surface is not the consumer app

This is the most important caveat on the page, and almost nobody states it. Tools that report on "ChatGPT visibility" are, with very few exceptions, querying OpenAI's API with web search enabled, not the ChatGPT product a customer uses. The same is true of Claude and Gemini: a measurement runs against the Messages API with web search, or the Gemini API with Google Search grounding.

These are genuine, useful measurement surfaces. They share retrieval infrastructure with the consumer products and they answer real questions with real citations. But they are proxies, and the consumer app may personalise, use a different model, apply memory, or route through a different retrieval path. A tool that says "you were cited in ChatGPT" when it means "the OpenAI web-search API cited you" is overstating what it saw.

Robot Visible records the surface and model on every result for exactly this reason. A citation with no surface recorded is not evidence, because you cannot tell what was actually asked or what answered.

Surface by surface

Google AI Overviews, and why an absent overview is not a failure

AI Overviews is the largest generative answer surface in existence and the one most likely to sit between your page and a buyer. It is also the surface with the most misleading measurement, because a query that returns no overview looks identical in a bad report to a query where a competitor won the citation.

Those are completely different findings. Robot Visible treats "this query produced no AI Overview" as its own recorded outcome rather than a negative result, in the same way a provider error is never counted as evidence against you. If your monitoring cannot distinguish the two, its trend line is measuring Google's overview trigger rate as much as your visibility.

AI Overviews is one of the two surfaces frozen into our public AI Visibility Index methodology, alongside Perplexity, which is why the distinction is enforced in the data rather than described in prose.

Google AI Mode, and the honest limit

AI Mode is a separate conversational surface rather than a block on the results page, and it is the one people most often ask how to optimise for. The honest answer is that there is no separate optimisation for it. Google's own documentation states there are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary.

Eligibility for both comes through ordinary Googlebot crawling, and the levers are the ordinary ones: be crawlable, be indexable, and be the page that answers the question. The snippet controls apply too, so `nosnippet`, `data-nosnippet`, `max-snippet` and `noindex` all restrict how your content can appear in AI features on Search.

Measurement is the real limit. Google documents no public API for AI Mode, so there is no supported interface for automated per-query collection. A person can still inspect AI Mode directly and preserve one answer with its query, visible source URLs, surface, locale, state, and timestamp. Search Console reports aggregate impressions with no query dimension. We do not measure AI Mode, and any tool claiming automated per-query tracking is worth asking how it accesses and preserves the surface.

What Search Console's generative AI report can and cannot prove

Google-Extended does not control what you think it controls

This is the most consequential misunderstanding in publisher crawler policy, and it costs sites visibility in both directions. Google-Extended is not a crawler. Google's documentation is explicit: it doesn't have a separate HTTP request user agent string, crawling is done with existing Google user agent strings, and the robots.txt token is used in a control capacity only.

What it controls is the use of content for training future Gemini models and for grounding in Gemini applications. What it does not control is Search. Google states that Google-Extended does not impact a site's inclusion in Google Search nor is it used as a ranking signal in Google Search, and eligibility for AI Overviews and AI Mode is governed by Googlebot, not by this token.

So the two common mistakes are opposite and both real. Blocking Google-Extended in the belief that it removes you from AI Overviews does not do that; it removes you from Gemini grounding while leaving the Search AI features untouched. And allowing it in the belief that it is required for AI Overviews is equally wrong, because that eligibility never depended on it.

What Robot Visible measures, and what it does not

Stating this plainly is more useful than a longer list of logos. The paid product checks your questions against OpenAI's web-search surface, Anthropic's web-search surface, Perplexity Sonar, and Gemini with Google Search grounding, recording the answer, the cited URLs, the provider, the surface, the model and the timestamp for every run. The public AI Visibility Index is frozen on two surfaces, Perplexity and Google AI Overviews.

We do not measure Google AI Mode, Microsoft Copilot, or the consumer apps directly, and we would rather say so than let a logo grid imply otherwise. A surface we did not check is not a negative result about your site, and it is never reported as one.

Run a free scan to see what these surfaces can currently read

Choosing which surfaces to track

• Start with where your buyers actually are. For most consumer and local questions that is Google's AI features; for developer and research questions the assistants carry more weight.
• Track a surface only if you can record per-query evidence from it. A surface you can only see in aggregate belongs in a different column of your reporting, not blended into a citation count.
• Treat each surface as its own series. Combining four providers into one visibility score hides the case where you gained on one and lost on another, which is the finding worth acting on.
• Record the model and surface alongside every result. Providers change models frequently, and a shift you cannot attribute to a model change will get attributed to your content instead.
• Do not read an absent answer as a loss. No AI Overview for a query, an unchecked provider and a provider error are three different things, and none of them is evidence against your page.