One of the challenges faced by the analyst of digital communication consists on the establishment of a corpus that preserve the representativeness and which responds to the nature of these samples of language, in particular as regards two parameters: multimodality and the multi-simultaneity. The article has three objectives. On the one hand, we review the situation of samples of digital speech at international level and specifically for the Spanish. Then, we show a methodological reflection on the problems of collecting and setting of data in the digital speech, from four cornerstones (communicative situation, nature of the data, representativeness of the sample and ethical issues). Finally, we propose guidelines to collect and transcript samples of digital speech for the repository feed; textual or enriched with multimodal data to the repository code.