GEO and AI Visibility: How to Monitor ChatGPT, Gemini and Claude

▶

For some time, measuring a website's organic visibility was relatively simple. We had a keyword, a SERP, and a position to monitor. If our website ranked third on Google for a specific search query, we could track that position over time, understand whether we were gaining or losing ground, and compare ourselves with competitors.

With the arrival of ChatGPT, Gemini, Claude, Perplexity, Google AI Mode and other generative systems, this logic has not disappeared, but it is no longer enough. Today, a brand can be cited by ChatGPT for a specific question and disappear from the same answer a few weeks later. It can be highly visible on Perplexity and almost absent from Claude or Gemini. It can be mentioned without its own website being used as a source, or appear through content published on completely different platforms.

This is where GEO, Generative Engine Optimization, comes into play, and above all where a problem emerges that, in my opinion, will become increasingly important in the coming months: how do we actually measure a brand's visibility within generative engines?

From SEO to GEO: the concept of ranking is changing

In traditional SEO, we are used to thinking through a fairly linear structure. A user performs a search, Google returns a SERP, and our website occupies a specific position. Of course, SEO has always been volatile too. Algorithms change, new competitors enter the market, content gets updated, and rankings move up and down. But the concept of ranking remains easy to understand.

Generative systems work differently.

The user does not necessarily receive ten ordered results. Instead, they receive an answer generated by the model, which may use different sources to produce it and may change even when very similar prompts are used. For this reason, simply trying to turn the old concept of SEO ranking into a hypothetical "ChatGPT ranking" risks leading us in the wrong direction. We need to start asking ourselves:

"How often does it appear? For which topics? On which models? Through which sources? And how stable is this presence over time?"

40-60% of AI sources can change from one month to the next

One of the figures that makes this issue particularly clear concerns citation volatility. An analysis based on monitoring 2,500 prompts across ChatGPT and Google AI Mode found that between 40% and 60% of cited sources can change from one month to the next. (Source: Search Engine Land)

This is an important figure because it completely changes the value of a single check. Let's imagine asking ChatGPT today:

"What are the best platforms for doing X?" Our brand appears in the answer. We take a screenshot, publish it on LinkedIn and say: "We're ranking on ChatGPT." It is certainly a positive signal, but it is not yet a metric.

That screenshot captures what happened at that particular moment. It does not tell us whether the brand will continue to appear, how frequently it will be mentioned, or whether other models will return the same result. In other words:

a screenshot is a snapshot. We need to start measuring GEO continuously.

Volatility does not mean GEO is useless

At this point, an objection might naturally arise. If sources change so frequently, what is the point of working on GEO? The real issue is not that we cannot influence visibility within generative engines. The problem is thinking that we can evaluate it through a single answer.

AI systems retrieve and use information from an extremely broad ecosystem. Index updates, new content, changes to retrieval systems, model variations, and even differences in how a question is phrased can change the sources that are selected.

The goal, therefore, should not be to achieve some kind of "permanent number-one ranking on ChatGPT". It should be to progressively increase the likelihood that our brand is recognized, mentioned, and used as a source for the topics that truly matter to our business. And, above all, to measure how this evolves over time.

ChatGPT, Gemini and Perplexity do not necessarily use the same sources

There is another issue that makes GEO monitoring even more interesting: there is no single AI ecosystem. Meltwater analyzed thousands of prompts and approximately 7.3 million citations across eight platforms, including ChatGPT, Claude, Copilot, Google AI Mode, Google AI Overviews, Gemini, Perplexity and Grok. (Source: Meltwater)

The results show significant differences. In August 2026, for example, YouTube was the most cited individual source overall in the dataset, followed by Reddit and LinkedIn. But when looking at individual models, the picture changed completely. Perplexity showed a particularly strong presence of LinkedIn, with more than 69,000 citations in the analyzed sample. Gemini and Google AI Mode, on the other hand, made extensive use of YouTube.

Among the main sources analyzed in the report, ChatGPT showed a stronger presence of institutional sources such as NIH. This means that talking generically about "AI visibility" risks becoming increasingly imprecise. A brand may have a strong presence on one system and a much weaker presence on another. This is why I increasingly believe it is important to monitor each platform separately.

GEO does not replace SEO

Another mistake I often see is presenting GEO as if it were destined to completely replace SEO. I do not believe this is the right way to interpret it. The foundations remain largely the same: useful content, a clear technical structure, authority, brand presence, structured data where appropriate, reliable sources, and information that is easy to interpret.

What changes is the ecosystem in which this information is used. We are no longer optimizing a page solely to help it achieve a strong position in a SERP. We are trying to build a digital presence that is sufficiently clear, authoritative, and distributed so that different systems can understand who we are, what we do, and when we are relevant to a particular question.

It is also interesting to observe the growing importance of external sources. In the same August Meltwater report, LinkedIn grew by 24.8% in a single month, reaching 100,200 citations and becoming the third most cited individual source in the sample. YouTube remained in first place with 212,500 citations. This reminds us that working on GEO does not simply mean adding a few FAQs to our website. It means thinking about the brand's overall digital presence.

Which metrics should we monitor in GEO?

If we can no longer rely solely on traditional rankings, we also need to change how we measure results. Personally, I would start with at least a few fundamental metrics. The first is the mention rate, meaning how frequently the brand is actually mentioned within responses to the prompts we have chosen to monitor.

Then we have citations, which are different from simple mentions. A model may know or mention the brand without directly using our website as a source. It therefore becomes important to understand:

1. which URLs are cited;
2. which external sources contribute to our visibility;
3. which competitors are mentioned alongside or instead of us;
4. on which platforms we have the strongest presence;
5. how this presence changes over time.

I would therefore not look for a single magic number to replace SEO rankings. GEO probably requires a combination of visibility, mention rate, citation rate, share of voice and trends over time. Above all, it requires continuous measurement.

We need to start with the prompts that truly matter to the business

Here too, I think it is easy to fall into the trap of monitoring everything. There is no need to add hundreds of prompts simply to fill a dashboard with numbers. We need to start with the questions that actually matter to the business.

If I sell CRM software, for example, I might want to understand what happens when a user asks "What is the best CRM for a small business?" or "Which CRM can I integrate with WooCommerce?" or even "What alternatives are there to [competitor]?".

From there, I can build a consistent set of prompts and repeat them over time. It is precisely this repetition that turns a simple check into usable data. Meltwater also highlights how prompt wording, category, location, timing of the measurement, and the specific retrieval system can influence the sources used, which is why repeated measurement is more useful than an isolated snapshot.

GeoTracking V3: why I started building a GEO monitoring system

It was precisely these issues that led me, over the past few months, to work on the development of GeoTracking V3, a proprietary Webita tool dedicated to SEO, GEO and AI Visibility monitoring.

The initial need was very simple. I did not want to limit myself to opening ChatGPT, asking a question, checking whether a specific brand appeared, and saving a screenshot. I needed to be able to create a project, define the important queries and prompts, monitor different systems and, above all, maintain a historical record of the results.

This makes it possible to start answering much more interesting questions:

1. Is the brand being cited more frequently than last month?
2. For which topics?
3. Which competitors appear?
4. Which sources are being used?
5. Is the presence stable, or does it depend on a few isolated occurrences?

This, in my opinion, is the direction in which organic visibility monitoring will need to evolve. Not replacing Search Console, Analytics, or traditional SEO tools, but adding a new layer of observation dedicated to generative search.

GEO is a process, not a screenshot

We are still at a very early stage. ChatGPT, Gemini, Perplexity, Google AI Mode and other systems will continue to change rapidly. Models will change, retrieval systems will change, the sources used will change, and the metrics we use to measure this ecosystem will probably change too. But one thing already seems quite clear to me.

Being cited once does not mean being consistently visible.

SEO continues to build content, authority, and a clear digital presence. GEO broadens the scope and forces us to observe how that presence is interpreted, retrieved, and used by generative systems. I therefore do not believe we are witnessing the end of SEO. Rather, I believe we are adding a new layer to organic search. And, as always, before optimizing something, we need to learn to understand it and find the right way to measure it