Formal experimental tests of mediumship are designed to control for cold reading. Standard practice is to keep the medium separate from the sitter so that it is not possible to monitor their nonverbal behaviour or to solicit verbal responses to subtle or implied questions during the reading. The medium remains ignorant of (‘blind to’) the identity of the sitter beyond perhaps being given their first name, so that they are not able to use hot reading or pigeon-holing techniques. The researcher who interacts with the medium is also blind to the identity of the sitter (called a double-blind design) so that they cannot unwittingly react to the reading in a way that could confirm or disconfirm statements. Typically, the sitter is given a number of readings, one intended for them but also others that are intended for other sitters so that they act as decoys or controls. Any Barnum-type effect might still inflate accuracy ratings, but would not allow the sitter to recognise their own reading relative to the others. Information provided by mediums is broken down into distinct items; sitters or independent judges rate each item (e.g. “totally wrong” to “totally correct”), often via Likert scales. Global ratings of overall readings are also used. Comparisons are made between intended readings (“correct” ones) vs control readings.

For example, the 2015 study by Julie Beischel and colleagues, “Anomalous Information Reception by Research Mediums Under Blinded Conditions II: Replication and Extension” aimed to replicate and extend earlier positive tests of mediumship. They recruited twenty mediums who had previously been tested and accredited by the Windbridge Institute. They produced 96 telephone readings under conditions that the researchers described as ‘triple blind’. This involved giving the mediums only the first name of the deceased person (the “discarnate”) and not telling them anything about the living person requesting the reading (the sitter). The mediums and sitters never spoke or interacted directly. Experimenters who interacted with the mediums were also blinded, meaning that they had no information about the sitter or the discarnate aside from the first name, so as to prevent the possibility of unintentional cueing. Thirdly, Sitters were blinded to which reading was intended for them. After the readings were transcribed, each sitter received two anonymized readings: the target reading (about their own deceased loved one) and a decoy reading (intended for someone else). They did not know which was which and were asked to evaluate both for accuracy.

The results showed that target readings were consistently rated more accurate than decoy readings. For example, in one key section involving five specific questions, sitters rated the accuracy of target readings at about 52.8%, compared to only 36.6% for decoy readings. Other metrics—such as the number of “hits,” global scores, and forced-choice selections—also supported the conclusion that mediums provided information more applicable to the intended recipient than chance would predict.

The authors concluded that these findings support the existence of “anomalous information reception” (AIR), a term used to describe the apparent acquisition of information by non-ordinary means. They acknowledged that while the mechanisms remain unknown, the results warrant continued investigation. Limitations included the subjective nature of sitter ratings and the potential for minimal identifying cues (such as first names), though these were carefully controlled. The study reinforces earlier findings suggesting that, under rigorously controlled conditions, some mediums demonstrate access to information that current scientific models cannot fully explain.

Of course, one significant study is not sufficient to provide compelling evidence for a claim. In the last 25 years there have been a number of such experiments, some of which have reported that sitters have been able to correctly identify their intended reading to a degree that is highly unlikely by chance alone. However, others have not been able to replicate these effects. This suggests that, much like other effects in the social sciences, this is not a phenomenon that can be reproduced ‘on demand’, but is sensitive to the particulars of the participants and the circumstances of testing. Nevertheless, when findings across all studies have been combined in a summary review and meta-analysis, Sarraf, Woodley and Tressoldi (2020) reported a statistically significant cumulative effect, which led them to conclude “the results of this meta-analysis support the hypothesis that some mediums can retrieve information about deceased persons through unknown means” (p. 1). While this does not provide definitive evidence that mediums’ claims are warranted, it is sufficiently promising to suggest that further, more intensive, investigations of mediumship would be worthwhile and could potentially tell us something important about human nature.