Why estimates of hallucinated citations are probably low