We propose AdaptDel, a randomized smoothing technique with adaptable deletion rates, achieving higher certified accuracy for sequence classification under edit distance perturbations, outperforming previous methods on diverse NLP tasks.