Generalising, this is one of the lessons of the past ~2 centuries in which we've had for the first time reasonably objective analogic recording capabilities: photography, phonography, cinema, and the like. Until their emergence, human testimony
across a wide range of phenomena was the only way to transmit information and, due to its low fidelity, low information density, unreliable interpretation and unreliable reproduction, that was at best only modestly reliable. A fantastic example of this (in numerous senses) is Albrecht Dürer's woodcut of a rhinoceros (1515), made from second-hand reports and sketches. On the one hand, it doesn't look true to life, but on the other, specific features of the animal are recorded with remarkable accuracy --- the segmentation of body plates, horns, toes, and aspect of the eyes for example. See: https://en.wikipedia.org/wiki/D%C3%BCrer%27s_Rhinoceros>.
And whilst analogue recordings have long been subject to manipulation, most of the time that took effort and expertise to accomplish smoothly, and independent recordings could be compared to detect edits and alterations. Following the emergence of digital image manipulation with photoshop, photographic "evidence" has become increasingly less evidentiary, with the spread of AI and smartphones, virtually all still and video images are at least somewhat processed, and with AI we can generate lifelike fabulations in realtime in multiple modes (still image, video, audio), including speech and background sounds, which can confound pretty much anyone, layperson or expert.
Which means that we're back in the realm of low-reliability fabulated reporting even or exspecially when mediated by our technologies, which had previously offered a solution to that problem.