The ability to improvise a speech extemporaneously was considered by ancient treatises to be the apex of an orator's preparation. However, the prestige attributed to the ability to improvise led, paradoxically, to the habit of inserting, even in speeches prepared in writing, something that mimicked, with greater or lesser effectiveness, elements that were instead characteristic of the impromptu performance. Through the examination of the ancient treatises on improvised speeches, from Alcidamas to Quintilian, this contribution will first of all try to examine the formation of a theoretical tradition relating to the techniques of fictitious extemporaneousness (those that Pliny the Younger will call figurae extemporales), and then analyze the presence of such techniques in orations that were certainly never uttered: epidictic orations that pretended to be judicial orations, such as the Platonic Apology of Socrates and the Antidosis of Isocrates in the Greek oratory, and orations published but never actually pronounced, as, in the Latin oratory , the Actio Secunda in Verrem, the Pro Milone and the Philippica Secunda. The results obtained from this analysis have methodological consequences that go beyond the texts we have explicitly dealt with: the probability that analogous figurae extemporales were commonly inserted even in orations normally passed through the performance represents a powerful incentive to exercise a methodical doubt in face to elements that we would otherwise be led to consider historical traces of what actually happened during the execution of the oration. In texts other than those considered here it will certainly be difficult - if not often impossible - to distinguish between authentic traces of extemporaneousness and artificial effects of improvisation: however, methodical caution in front of supposed improvisation must represent an unavoidable part of the exegetical baggage of the scholar of ancient oratory.