Unlike at lower redshift , where there is a 40 % detection rate , surveys for 21-cm absorption arising within the hosts of z \stackrel { > } { { } _ { \sim } } 1 radio galaxies and quasars have been remarkably unsuccessful . Curran et al . ( 15 ) suggest that this is due to the high redshift selection biasing towards the most optically bright objects ( those most luminous in the ultra-violet in the rest-frame ) , where the gas is ionised by the active galactic nucleus . They therefore argue that there must be a population of fainter objects in which the hydrogen is not ionised and which exhibit a similar detection rate as at lower redshifts . In order to find this “ missing ” gas at high redshift , we have therefore undertaken a survey of z \stackrel { > } { { } _ { \sim } } 2 radio sources , selected by optical faintness . Despite having optical magnitudes which indicate that the targets have ultra-violet luminosities below the threshold where all of the gas is ionised , there were no detections in any of the eight sources for which useable data were obtained . Upon an analysis of the spectral energy distributions , ionising photon rates can only be determined for three of these , all of which suggest that the objects are above the highest luminosity of a current 21-cm detection . The possibility that the other five could be located at lower photon rates can not be ruled out , although zero detections out of five is not statistically significant . Another possible cause of the non-detections is that our selection biases the sample towards sources which are very steep in the radio band , with a mean spectral index of \left < \alpha \right > = -1.0 , cf . -0.3 for both the 21-cm detections and UV luminous non-detections . This adds the further possibility that the sources have very extended emission , which would have the effect of reducing the coverage by the putative absorbing gas , thus decreasing the sensitivity of the observation .