A manager who observes that a generative artificial intelligence mentions his company often draws a hasty conclusion: to be mentioned would already be good standing. Yet visibility in an answer and how that answer describes the business are two separate things. The first indicates whether one appears; the second, how one is perceived once named. The 2026 edition of the CAC 40 AI Reputation Barometer, published for the first time in June 2026 and co-produced by Vectors and Trickstr, makes it possible to objectify this distinction at the scale of CAC 40 companies. This article looks at what the study concretely brings to a communications or digital transformation department.
Being quoted is not being well received
A generative AI model might mention a company in an open response about its industry without describing it favorably. Conversely, he may reserve a balanced and factual description for a company that he cites more rarely. Confusing the two amounts to controlling one’s reputation on a single indicator, the frequency of citations, which provides neither the tone used, nor the facts used, nor the sources used by the model to construct its response.
For a communications department, this nuance has practical consequences. A high volume of mentions can mask an incomplete, dated or unbalanced description, while a more discreet presence can be accompanied by an accurate portrait. Tracking the two dimensions separately avoids drawing conclusions from an indicator that only measures part of the issue.
A large-scale method
To objectivize this distinction, the 2026 edition of the AI Reputation Barometer 2026 administered nearly 39,800 questions to four generative AI models, compared to 4,536 questions covering seven dimensions during the previous edition. Each CAC 40 company is now rated on nine reputation dimensions: vision and strategy, products and services, innovation, employer brand, financial performance, CSR, brand equity, governance, and adoption of AI.
This change in scale allows us to examine more responses and dimensions; the reliability of the comparisons nevertheless depends on the protocol and possible biases. The number of questions alone is not enough to guarantee the solidity of the results: the formulation of the queries, the choice of the models questioned and the collection period are just as important.
What the 2026 ranking shows
According to this edition, Schneider Electric comes at the top of the reputation ranking. This result deserves to be read for what it is: a measure of how the models surveyed describe the company on the nine dimensions retained, and not an objective judgment on the quality of the company itself.
The 2026 edition also describes a correction of the industrial bias which characterized previous rankings, with a notable progression of luxury and service companies. This movement invites us not to consider industrial domination for granted from one edition to another: the composition of the classification evolves with the method and with the available content on which the models are based.
Vectors and Trickstr, two complementary roles
The study combines two organizations with distinct roles. Vectors, a strategic communications consulting agency, provides insight into the reputation and perception issues specific to large companies. Trickstr provides the technical measurement chain: PromptForge, a tool it developed and integrated into its Kosmos platform, orchestrates the generation of questions, their large-scale administration, the structured collection of responses and their analysis.
This combination allows the study to hold both a corporate communication discourse and a measurement method based on a recorded protocol, which can be reused by documenting the conditions of each collection. As models and available information evolve, neither conditions nor results are guaranteed to be identical from one campaign to another.
Observation or causality: what a cited source says
One point deserves particular attention. Source identification describes the references cited in the responses. It does not alone demonstrate their influence on grades. Possible correlations between media coverage and perception require separate analysis and do not prove causality.
Other factors, specific to each company and each sector, may also come into play. Identifying the sources remains a useful step in understanding what content the models are based on, provided that you do not draw conclusions about their real weight in the final grade.
A concrete approach to monitoring your reputation in AI
A company that wishes to appropriate this distinction between visibility and perception can structure its approach in four stages:
- Establish an initial state : how does the company appear today in open questions about its sector, and how is it described when it is named directly?
- Review the responses obtained and the sources that the models cite in support, to understand what content the current description is based on.
- Correct, if necessary, public information available on the company so that they are relevant and up to date, without seeking to artificially guide the models’ responses or guarantee their content.
- Perform a new measurementunder conditions comparable to the first, to observe developments, without automatically attributing them to the corrections made.
This approach guarantees neither a gain in visibility nor a referencing result in the responses of generative AI: the models evolve according to their own training cycles, independently of the actions undertaken by a company. On the other hand, it offers a framework for transforming an intuition, “is our company well perceived by AI? », in structured monitoring, comparable over time and based on data rather than on a one-off impression.
