Taken from Hills et al (2019):

Participants undertook the same experimental procedure (an old/new recognition experiment with three consecutive phases: learning, distractor, and test) with the exception of the instructions provided to the participants. If taking part in an “observed” trial, the participant was informed of the webcam and that it would be recording their behaviour during the trial in order for the Experimenter to establish how people react when viewing faces - participants were not told that their performance was being judged by the camera. A vital element to the success of this study was the belief of observation. In order to make this manipulation believable to participants, the researcher reinforced the realism of the “working” camera by adjusting the zoom, fixing the location and by turning it on in front of the participant. The researcher also showed the participant their face on the secondary computer screen.
Participants in the recorded at learning conditions were informed of the recording at the start of the experiment. Participants in the recorded at learning only condition were told immediately after viewing the learning set of faces that the webcam was no longer recording - the researcher then disconnected the webcam to solidify belief of authenticity. Participants in the recorded at test phase only condition were informed of the recording before the test phase and not before.
During the learning phase, participants viewed nineteen 5 second videos of volunteers counting from 1 to 5 sequentially, in a random order. Participants were required to count aloud from 1 to 5 at the same time as the person in the video. This was to ensure adequate attention was paid to the stimulus. Between each face there was an inter-stimulus interval of 5 seconds; during which the participants were presented with the following participant number and a statement requesting to press any key to continue.
Immediately following this, the distractor phase began, consisting of two 5 minute activities, irrelevant from the task measured. In the first phase, participants viewed the screen populated with multiple (1-15), randomly positioned squares and triangles and were required to indicate using the keyboard (S-square, T-triangle) which was the majority shape. In activity two, simple maths problems were displayed requiring the participant to indicate whether the answer provided was true or false, again by pressing the appropriate key (Z-true, M-false). The order of trials within these tasks was randomised. Immediate feedback was provided for both activities, where a green (for correct) and red (for incorrect) screen with the words “correct” or “incorrect” in the top left corner of the screen was presented; added to this display was also the overall percentage score that the participant had achieved in the task.
Following this, a two-alternative forced-choice recognition test was completed in which all 38 still images of faces were presented in a random order. Nineteen were target images (faces displayed as videos in the learning phase) and 19 were non-targets (new images). Participants were required to indicate whether they had seen the image before or not, using the keypad (S-seen, N-not seen). Test images persisted on screen until a response was made. Between each face there was a plain white screen for 150ms.