As a consequence of the large (and growing) number of near-Earth objects discovered, some of them are lost before their orbit can be firmly established to ensure long-term recovery. A fraction of these present non-negligible chances of impact with the Earth. We present a method of targeted observations that allowed us to eliminate that risk by obtaining deep images of the area where the object would be, should it be on a collision orbit. 2006 QV89 was one of these objects, with a chance of impact with the Earth on 2019 September 9. Its position uncertainty (of the order of 1 degree) and faintness (below V$sim$24) made it a difficult candidate for a traditional direct recovery. However, the position of the virtual impactors could be determined with excellent accuracy. In July 2019, the virtual impactors of 2006 QV89 were particularly well placed, with a very small uncertainty region, and an expected magnitude of V$<$26. The area was imaged using ESOs Very Large Telescope, in the context of the ESA/ESO collaboration on Near-Earth Objects, resulting in very constraining a non-detection. This resulted in the elimination of the virtual impactor, even without effectively recovering 2006 QV89, indicating that it did not represent a threat. This method of deep non-detection of virtual impactors demonstrated a large potential to eliminate the threat of other-wise difficult to recover near-Earth objects