Conversation analysis researchers insist on the use of audio- or video recordings of episodes of 'naturally occurring', that is non-experimental, interaction as their basic data. Having said that, as we shall see below (Section 6.2.7), some researchers have analysed conversations in written, fictional texts (Muirhead, 2000).
Recording of conversations results in data that can be examined and re-examined and provides a safeguard against analytic conclusions that are the result of selective attention on the part of the researcher, or idiosyncratic intuition or presupposition about what has been said. The analysis can, in effect, be checked against the recorded conversation in a way that, for example, a reported interview, reliant on the researchers' notes and memory, cannot.
At what point exactly this preference for tape-recorded data became invested with an actively-developed suspicion towards the use of 'unsatisfactory data sources' in language description (e.g. interview data, observational data obtained through field notes, invented examples and experimental elicitation) is a matter of hindsight interpretation. (Slembrouck, 2006)