Brand recommendations: a study reveals the inconsistency of artificial intelligence

In recent years, the rise of artificial intelligence has revolutionized how recommendations are formulated by increasingly sophisticated algorithms. Yet, a recent study highlights a worrying paradox: the glaring inconsistency of these systems in recommending brands and products. By 2026, the promise of reliable technology to guide consumers seems to be fading as AI tools reveal their fragility, particularly their tendency to produce fluctuating and sometimes inconsistent lists. The reliability of these recommendations is now at the heart of the debate in a context where companies are investing hundreds of millions of dollars in visibility tracking and data collection to refine their strategies. However, the question remains: at what cost are these algorithms truly effective or, more simply, credible? With the widespread adoption of these new methods, specialists are questioning the true usefulness of a system that, despite all its innovation, remains subject to random variability that is difficult to control. The rapidly growing market for visibility tracking in the artificial intelligence sphere is at a crossroads between technological advancement and the need for greater rigor in data interpretation. Trust in these recommendations is becoming a major strategic issue for brands, which must now navigate a landscape where the consistency and accuracy of algorithms do not always go hand in hand with optimized results.

How algorithms are revolutionizing the reliability of brand recommendations.

The artificial intelligence systems that manage brand recommendations today use algorithms fueled by massive amounts of data. Yet, behind this sophisticated facade lies a far more chaotic reality. Originally, these systems were designed to analyze consumer behavior, index preferences, and provide relevant lists. But in practice, their reliability is undermined by the probabilistic nature of their operation, where each query can elicit a radically different response. According to a study by specialists, for every question asked identically 100 times, 99 different answers can be generated, rendering any attempt at standardization futile. This phenomenon is explained by the machine learning method of these agents, where each result depends on a specific context, often invisible even to experts. The direct consequence: the list of recommended brands, their order, and the number of items can vary considerably, which calls into question the very notion of ranking. In a sector where brand loyalty is based precisely on consistency, this variability becomes a major challenge, especially for marketers. By going beyond traditional metrics and attempting to assess visibility through a percentage of appearances, we find that these trends fluctuate depending on the prompts, sectors, and even usage contexts. To address this instability, some specialists advocate an approach based on massive query repetition to obtain a more reliable metric, but this remains a precarious solution given the probabilistic nature of algorithms.

→ À lire aussi AI Performance: Microsoft Bing Unveils First Official GEO Measurement Tool Data · 11 Feb 2026

The risks of inconsistency in brand recommendation for modern scraping/la-polyvalence-du-scraping-un-outil-mille-possibilites/">marketing

→ À lire aussi the best bots to follow to assess your visibility on language models Data · 25 Jul 2025

Discover common inconsistencies in artificial intelligence and how to identify them to improve the reliability of AI systems.

A groundbreaking study reveals the complexity of the “recommendation” variable in automated search systems.

The originality of this study, conducted over several months, lies in its scope and rigor. Confused by the extreme diversity of prompts formulated by thousands of users, the analysis was entrusted to a platform capable of normalizing nearly 3,000 queries covering various economic sectors. The result: astonishing variability in recommendations, both in the list of brands suggested and in their positioning. Tests show that for each query, the number of different brands mentioned can range from dozens to several hundred. The result: total confusion, as it becomes impossible for a brand to consistently appear in the same order or even in the same list. According to the researchers, this inconsistency clearly illustrates the fundamental limitation of current systems, which merely generate random or semi-random lists but offer no guarantee of their accuracy. However, in certain sectors, such as connected devices or niche products, maximum visibility often reaches a ceiling that is difficult to surpass, illustrating AI’s ability to capture a degree of consistency amidst chaos. Nevertheless, analyzing the variation in human prompts also reveals a diversity created by the originality of the forms, further complicating the work of specialists. Recommendations in this context thus become data to be handled with care, each result requiring precise interpretation.

Cost of tracking.

Multiple queries required to achieve stability.

Significant increase in investment.

Why are AI-driven brand recommendations so variable?

The probabilistic behavior of algorithms, which generate random or semi-random lists, often leads to extreme variability in results, even for identical queries, complicating their reliability.

How can brand visibility be effectively measured in an unstable context?

Recommend repeating queries to obtain a percentage of appearance across a large volume of attempts, resulting in a more stable and representative metric.

Are existing tools for tracking these recommendations reliable?

Écrit par

Kevin Grillot

Consultant Webmarketing & Expert SEO.

Voir tous les articles →