In SEO, we're used to talking about rankings. A keyword, a SERP, and a website that appears in a certain position. We can track it over time, compare it to competitors, and analyze whether it's rising or falling.
When we try to apply this same logic to ChatGPT and other AI-based search systems, we run into a problem: an AI-generated response does not function like a traditional SERP.
This directly affects how we should measure a brand's visibility in GEO (Generative Engine Optimization).
Perhaps the question should no longer be “Where does my company rank?” but rather:
How often does my brand appear when a user asks questions relevant to my business?
Why Visibility on ChatGPT Is Different from SEO Ranking
Google's search results aren't completely set in stone. They can vary depending on location, device, the time of the search, and other factors. Even so, traditional SEO provides us with a relatively predictable environment: we can track keywords, URLs, and rankings and analyze how they change over time.
With generative systems, this stability is lower.
Repeating the exact same prompt does not guarantee that you will always get the same answer. The brands mentioned, the recommendations, the structure of the answer, and the sources used may all change.
Factors That Influence AI-Generated Responses
In addition, factors such as the following may influence the outcome:
- the exact wording of the prompt;
- the context of the conversation;
- the language;
- the location;
- the model used;
- and any additional context that the system has available during that interaction.
Therefore, two seemingly identical executions do not necessarily produce the same result.
This variability is one of the main differences we need to take into account when we shift from measuring SEO to measuring GEO.
Citation drift and volatility of cited sources in AI
Variability does not only affect the content of the responses. It also affects the sources that AI systems choose to cite.
This phenomenon is commonly known as citation drift.
In May of this year, SISTRIX published a study based on 82,619 prompts, more than 1.5 million observations, and 17 weeks of tracking across various platforms and countries.
The results show a high turnover rate among the cited sources.
In the case of ChatGPT Search, SISTRIX observed weekly domain turnover rates of up to 74%, although the study itself shows significant differences depending on the country analyzed.
In other words: just because a domain is cited today does not necessarily mean it will still be one of the sources used next week.
At the URL level, according to the same study, the variability is even greater.
This highlights an important difference from SEO: getting a one-time mention does not necessarily mean achieving consistent visibility.
Why a single query to ChatGPT does not demonstrate GEO visibility
This highlights one of the common challenges when analyzing a brand's presence in AI.
We ask ChatGPT a question.
The brand appears.
Let's take a screenshot.
And we conclude that we have visibility.
But that response is merely an observation of a variable system.
The opposite can also happen: we run a prompt once, our company doesn't show up, and we conclude that we don't have GEO visibility.
Both conclusions may be premature.
How to Measure a Company's GEO Visibility in ChatGPT
If we know that the result can vary from one run to the next, a single query is not enough to draw solid conclusions. To obtain a more representative picture, it is necessary to work with multiple observations and analyze the results in aggregate.
One possible approach is to repeat the same prompt a sufficient number of times to identify patterns of occurrence, mentions, and citations. The sample size will depend on the objective of the analysis, the available resources, and the level of variability we wish to observe.
Mention and citation metrics for measuring GEO visibility
For example, in a test with a large number of runs, we could analyze:
- what percentage of responses mention our brand;
- how often the main competitors appear;
- how many times our domain is mentioned;
- Which domains appear most frequently as sources;
- which specific URLs are cited;
- and to what extent those appearances persist across different runs.
The goal is not to review each response in isolation, but to transform a set of results into data that can be used to identify trends.
How to Calculate the Frequency of a Brand's Appearance
If, for example, a company appears in 68 out of 100 runs conducted under the same test conditions, we could say that it has a 68% occurrence rate within that sample.
It wouldn't make much sense to interpret that result as a "GEO position." What we're measuring is something different: how often a brand appears within a set of responses.
Therefore, in this context, it may be more useful to talk about visibility frequency rather than position.
How to Choose Prompts for Monitoring GEO Visibility
There is a second aspect to the problem.
Real users don't all ask the exact same question.
Examples of prompts with a similar search intent
Someone might ask:
What is the best company in X?
Another one:
Which companies in X would you recommend?
And a third one:
I'm looking for X for a small business with a budget of Y. What options do I have?
The intent may be very similar, but the prompts are different.
Therefore, a GEO monitoring strategy should not rely solely on repeating a question many times. It should also work with clusters of prompts that are representative of the user's various intents.
Monitoring prompts across different models over time
SE Ranking, for example, recommends starting by monitoring between 20 and 40 prompts, tracking them across two or three models, and collecting at least 30 days of data before drawing conclusions.
The two approaches actually address different questions.
Repeating a prompt helps us understand the variability within a single query.
Tracking different search queries over time helps us understand the brand's visibility within a specific topic or search intent.
Combining both dimensions can give us a much more representative picture.
GEO visibility is measured using patterns of occurrence over time
As interest in visibility on ChatGPT and other AI systems grows, it's tempting to look for a direct equivalent to SEO rankings.
Traditional SEO has led us to think in terms of rankings. At GEO, it may be more useful to think in terms of probabilities, frequency, and the distribution of visibility.
That is why, when asked:
"Does my company appear in ChatGPT?"
Maybe we should start answering:
"How often does it appear, for what types of questions, compared to which competitors, in which models, and from which sources?"
Because, in a changing environment, a single response is not enough: visibility is best understood by observing patterns of occurrence over time.
Frequently Asked Questions About Visibility in ChatGPT and GEO
GEO stands for Generative Engine Optimization and refers to the optimization of a brand’s presence in the responses of generative AI systems. Measuring its visibility involves analyzing how often it appears, in response to which questions, compared to which competitors, and with which sources.
Select relevant questions that your potential customers might ask, and monitor the responses across various scenarios and time frames. Track mentions of your brand, citations of your domain, and the presence of competitors. A single search query is not enough to assess consistent visibility.
Responses may vary from one run to another. They may also be influenced by the wording of the prompt, the context of the conversation, the language, the location, and the model used. That is why it is advisable to record the conditions of each test and compare results obtained under similar conditions.
Divide the number of responses in which the brand appears by the total number of responses analyzed, and multiply by 100. If it appears in 68 out of 100 trials, its frequency of appearance is 68% in that sample and under those test conditions.
No. A mention occurs when the response names the company. A citation occurs when it uses the company’s domain or a specific URL as a source. It’s a good idea to track both separately, since the brand may appear without its website being cited.
There is no one-size-fits-all number for all businesses. As a starting point, SE Ranking recommends between 20 and 40 prompts, across two or three models, for at least 30 days. The sample size and number of repetitions should be tailored to the objective, available resources, and observed variability.
Citation drift refers to the variation in the sources cited by AI systems over time. A domain or URL may be cited in one response and absent from another. For this reason, a single citation should be interpreted within the context of ongoing monitoring.
Sources and Recommended Reading
SISTRIX (2026). AI Citation Drift: How Stable Are Sources in AI Search Results?
SE Ranking (2026). How to Choose the Prompts Worth Tracking in AI Searches.
