Research Summary

Abstract:

Since 2022, the widespread adoption of conversational agents (ChatGPT, Claude, Gemini, Copilot, Le Chat) has been reshaping the way travelers access regional information. These algorithmic systems are gradually replacing search engines and tourist offices as the primary sources for trip planning, without the internal logic behind the generation of their narratives or the structural biases that permeate them being truly documented. This paper offers an exploratory analysis of the representations of Agadir and the Souss-Massa region produced by five large language models (LLMs), queried according to a standardized protocol in French using thirteen prompts covering the entire tourism spectrum (description, itineraries, history, Amazigh culture, hinterland, authenticity, gastronomy). The resulting corpus comprises 65 verbatim responses, totaling 38,402 words of automated tourism discourse. The analysis employs a theoretical framework that integrates the theory of devices (Peraya, Jeanneret), algorithmic mediation (Cardon, Rouvroy), and critical AI studies (Bender, Crawford), and combines quantitative methods (lexicometry, thematic coding, TF-IDF cosine similarity) to measure inter-model convergence, spatial coverage, and the depth of cultural treatment. The results show a marked spatial asymmetry favoring the coastline, significant inter-model lexical convergence that peaks in prompts with strong encyclopedic markers (history, Amazigh), a spontaneous erasure of the Amazigh dimension, corrected only when explicitly requested, and a systematic reconfiguration of Agadir’s authenticity in contrast to the imperial cities (Marrakech, Fes), revealing an interpretive framework shared by all five models. The study proposes a conceptualization of chatbots as devices of territorial mediation that produce regimes of algorithmic visibility, and discusses their implications for regional tourism communication.