ChatGPT, Gemini, Perplexity: The Algorithm That Changes Everything

The Impact of Algorithms on Brand Visibility

When discussing visibility on platforms like ChatGPT, Gemini, or Perplexity, one often thinks simply of the mention of a brand's name by these tools. However, this view is reductive. In fact, a brand's visibility depends not only on the interface used but also on the underlying AI model, such as GPT-4.1 or GPT-5.4. Each model can produce different results depending on the query, context, freshness of information, and whether web search is activated or not.

It is crucial to understand the distinction between fast models, which prioritize response speed, and thinking models, which focus on reasoning and in-depth analysis. The latter are capable of exploring a wider range of sources and perspectives, which can influence how a brand is cited. Thus, the choice of model can alter the hierarchy of mentioned brands and the number of visible players.

To effectively measure a brand's visibility, it is essential to consider not only the engine used but also the model, type of query, search intent, phrasing, and even the targeted persona. Brand Score AI, an internally developed tool, allows for mapping this shifting visibility within LLMs, demonstrating that visibility is not a fixed state but a constantly evolving dynamic.

Variations in Citations Across Different Models

The differences in visibility between AI models can be striking. A brand may be very well represented by one model and much less so by another, even for similar queries.

This disparity has been observed across various sectors, particularly in banking. By comparing the performance of traditional banks, neobanks, credit agencies, and comparison sites across several ChatGPT models, it became clear that visibility curves can reverse depending on the model used.

Some players who are less visible on a fast model may gain visibility on a more advanced or thinking model, and vice versa. Additionally, some models tend to cite a limited number of players, while others broaden their responses by mentioning more brands, categories of players, or alternatives.

Understanding the Gaps Between Models

The differences between two models of the same family can be attributed to several factors. First, they do not necessarily rely on the same training datasets, weighting mechanisms, or response guidelines. This includes the level of caution, the tendency to recommend a brand, cite sources, or broaden the scope of the response.

Furthermore, some models excel in reasoning, others in synthesis, and still others in retrieving recent information. For example, fast models may produce responses very close to the initial prompt, limiting themselves to the explicitly requested players. In contrast, thinking models may broaden their analysis, integrate other categories of players, and provide a more comprehensive response, thereby altering GEO visibility.

The Rise of Locally Hosted Models

The emergence of locally hosted AI models or those in private environments could further fragment brand visibility. Two companies using the same base model can yield very different responses if they have connected it to specific data, documents, or business rules.

For brands, this means that AI visibility is no longer solely determined by large public models but also within private ecosystems, such as supplier databases, partner content, product documentation, structured data, and industry reference sources.

Thus, a brand may be visible in a public model but absent from an internal AI assistant if it is not present in the sources used by that environment.

Conducting a Reliable Presence Audit

To conduct a reliable audit of visibility, it is essential to cover multiple dimensions: engines, models, search intents, simple or complex formulations, targeted personas, competitors, etc. It is also crucial to repeat these tests regularly, as responses can vary significantly from one session to another.

At Origine, we use Brand Score AI, our proprietary tool, to track brand visibility within LLMs. This tool allows monitoring brands across specific models and analyzing multiple models within the same AI, according to needs and prompt types.

The goal is to measure not only citations but also the context of mentions, associated competitors, sentiments, sources used, and the stability of mentions over time.

Strategies for Advertisers in a Changing Environment

There is no one-size-fits-all solution to being more visible on one model than another, although some models have specific characteristics. Responses are constantly recomposed.

The priority is first to allow access to AI bots when relevant, and then to strengthen the fundamental signals already known in SEO: expert content, structured data, authority, popularity, and semantic coherence.

It is also essential to develop visibility and awareness outside of one's own site. This involves obtaining citations in authoritative media sourced by AIs, identified through GEO audits. It is also necessary to work on the synergy between GEO and public relations, identifying semantic fields to strengthen in order to increase mentions within specialized sites or authoritative media.

Finally, it is crucial to think omnisearch: Google, YouTube, Reddit, LinkedIn, media, marketplaces, comparison sites, forums, and LLMs are now interconnected in new search journeys. Visibility must be built across all these touchpoints.

Pitfalls to Avoid in GEO

Certain mistakes should be avoided. The first is to believe that there is a single technical shortcut: an "AI-optimized" page, a file dedicated to LLMs, or a few tags added to the site will not be enough to sustainably boost a brand's visibility. LLMs rely on a much broader set of signals: authority, notoriety, quality of sources, semantic coherence, and mentions.