For decades, corporate reputation has been managed mainly in three spaces: the media, search engines and social networks. The expansion of ChatGPT, Gemini, Perplexity, AI Overviews and other conversational models has added a fourth territory: the answers that artificial intelligence offers when users search for information, compare alternatives or directly ask for a recommendation.
The issue introduces a new layer of complexity for communications departments. A brand can appear frequently and yet not be recommended; stand out in one model and be practically invisible in another; or discovering that the information with which an AI builds its reputation comes from sources that the company barely controls.
A brand may discover that the information with which an AI builds its reputation comes from sources it does not control
It is a reality that Eduard Corral, Founder and Head of Innovation at Buzz, already addressed during the AIBC 2026 event, by distinguishing three levels of presence of a brand in the models: recall, reasoning and recommendation. The first involves the model remembering the brand when asked about its category; the second, what attributes and associations it builds around it; and the third, the most valuable, is that it is recommended as the first option without the user having previously mentioned it. “Appear” It is not equivalent, therefore, to occupying a relevant position. “It’s not how much, it’s how.”Corral then explained, also pointing out that measuring only ChatGPT or only Gemini would be a mistake, since the models have different behaviors.
Precisely the fragmentation, the lack of methodologies and the difficulty in understanding the sources now appear among the main concerns included in the first “Decalogue of concerns of directors regarding LLMs”, prepared by Asesores and Good Rebels.
The document is based on two focus groups held between the end of June and the beginning of July with 30 communications directors and media relations managers. Its conclusions have been contrasted with four sectoral analyzes of banking, insurance, hotels and telecommunications carried out on more than 40,000 responses from ChatGPT, Gemini, Perplexity and AI Overviews, with more than 10,000 prompts per model.
The ten concerns of Dircoms regarding LLMs
The decalogue shows that the concern of those responsible for communication goes far beyond getting mentions. It ranges from media selection and information expiration to paid content, reviews or the ability to justify new investments.
1. Without proven methodologies. The models are still opaque regarding how they construct their responses and there is no consolidated methodology to achieve early positions in the mentions. Furthermore, what we learned today can change in a few months.
2. Integrate PR, SEO and GEO without reputational risk. The question arises about how to coordinate these disciplines and what the strategy should determine. The Dircoms fear an over-optimization oriented towards commercial models and objectives that ends up damaging the global corporate reputation.
3. Redefine the map of relevant media. The analyzes have identified that some media considered secondary have a significant weight in the responses, while certain reference headings barely appear in some cases. This forces us to review the traditional criteria with which relations with the media are prioritized.
4. Maintain the coherence and validity of the messages. Old content, both your own and published by third parties, may continue to feed incorrect or outdated answers. Reputational management thus requires thinking not only about what to publish now, but also what historical information is still available.
5. Fragmentation between models. Each LLM consults, cites and links to sources differently. A brand can lead in one and be almost invisible in another, which raises the question of where to concentrate resources and with what time horizon.
6. The paradox of coverage. Having extensive coverage does not guarantee a dominant position. A brand can accumulate impacts and receive late or diluted mentions, while more specialized companies can gain relevance in specific territories with less coverage.
7. The limited weight of own channels. Depending on the sector, the corporate website can represent between 2% and 25% of the sources used in the analyzed queries. Comparators, aggregators and third parties that the company does not control thus acquire increasing importance.
8. The opacity of paid content. According to the report, as far as is known, the models do not detect sponsored content when it is not flagged. The Dircoms observe a possible tactical opportunity, but also uncertainty about its regulatory sustainability and how the media will manage it.
9. Reviews as a critical and vulnerable front. The models place a high weight on user opinion, which raises questions about fake reviews, coordinated negative review campaigns, and internal responsibility for this new reputational risk.
10. Budgetary limitations in the face of a still uncertain ROI. The methodologies are poorly contrasted and it is still difficult to attribute an improvement to specific actions. The test and learn mechanics become necessary, but it also makes it difficult to justify budgets without a proven return.
“The directors convey to us great concern and an enormous need for judgment. They know that the reputation of their companies is already at stake in AI responses and they want to decide with data where to invest their resources”explains Carlos Hergueta, Partner and Director of Business Development at Asesores.
For Hergueta, this scenario also confirms a historical trend in public relations: “Credibility is built through third parties, and now it must also be built to be read by machines.”
This point connects with another of the ideas raised by Corral during AIBC. The corporate website continues to have value, but the models contrast companies’ claims with external sources. “The models read from your website, but they confirm what you tell them”he explained. Hence, not all content has the same value. Generic statements such as “we are the first” either “we are the best” They provide little information if they cannot be verified. Data, figures, recognitions, external references and content that allow verification of a statement offer, on the other hand, materials that are more likely to be used by models.
The media can be almost twice as influential for LLMs as other sources
The new report from Asesores and Good Rebels provides data to this issue. Their main finding is that the media can be almost twice as influential for LLMs as other sources. Thus, when a medium appears as a source, brand mentions are multiplied by 1.78 in banking and by 1.64 in insurance. In hotels the multiplier reaches 1.38 and in telecommunications it stands at 1.05, with a weight equivalent to that of websites and comparators.
Comparators have, however, a particularly high presence: between 70% and 90% of responses mentioning a brand mention them. Their high penetration means that the report considers them a necessary condition, but not necessarily a differential one.
In banking, analyzes also indicate that an architecture made up of between six and eight diverse sources, combining its own website, three or four comparators and one or two editorial references, raises the mention rate above 61%-68%.
The results reinforce one of the ideas also raised during AIBC: general notoriety does not guarantee recommendation when the query becomes specific. Corral then used the example of sports shoes: Nike or Adidas can dominate a general question, while specialized brands such as Asics or Hoka can appear ahead when the user asks specifically about running. For brands challenger, This opens opportunities to build authority in specific territories.
Measuring how much a brand appears is no longer enough
The transformation also affects the metrics. Compared to the Share of Voice, which allows us to observe the frequency with which a company appears, Buzz also proposes the Share of Recommendation, which analyzes how many times it occupies the first recommendation. Added to this are issues such as the depth of the mention, the semantic tone and the sources used.
This distinction is especially relevant for communication departments: one brand may appear repeatedly in the responses and always do so in secondary positions, while another may accumulate fewer mentions but be preferentially recommended for certain needs.
Advisors and Good Rebels propose four recommendations in this regard:
- Understand the particularities of each model
- Build a diverse font architecture
- Make editorial content a differential element
- Continually monitor and adjust brand presence
Asesores has also developed an algorithmic reputation method structured in seven phases that starts from identifying the weight of each type of source and comparing the citability of the brand against its competitors. Then the means that construct the responses are isolated, recurring patterns and approaches are detected, communication territories are defined, these learnings are incorporated into the PR strategy and the measurement is periodically repeated to adjust the actions.
The proposal largely coincides with the five steps that Corral proposed at AIBC: measure, diagnose sources, produce citable content, influence those sources and monitor on a recurring basis.
And reputation in LLMs poses a new technical scenario for brands and therefore represents an expansion of traditional communication work, which now also involves understanding what the AI remembers about a brand, how it reasons about it, what sources it uses to build that perception and, finally, under what circumstances it decides to recommend it.
