How do health and social care professionals deal with undecipherable talk produced by adults with intellectual disabilities (ID)? Some of their practices are familiar from the other-initiated repair canon. But some practices seem designed for, or at least responsive to, the needs of the institutional task at hand, rather than those of difficult-to-understand conversational partners. One such practice is to reduce the likelihood of the person with ID issuing any but the least repair-likely utterances, or indeed having to speak at all. If they do produce a repairable turn, then, as foreshadowed by earlier work on conversations with people with aphasia, their interlocutors may overlook its deficiencies, respond only minimally, simply pass up taking a turn, or deal with it discreetly with an embedded repair. When the interlocutor does call for a repair, they will tend to offer candidate understandings built from comparatively flimsy evidence in the ID speaker’s utterance. Open-class repair initiators are reserved for utterances with the least evidence to go on, and the greatest projection of a response from the interlocutor. We reflect on what this tells us about the dilemma facing those who support people with intellectual disabilities.