Can Anyone See This? Delivery Confirmation as a Community-Bound Sequence in Algorithmically Curated Comment Spaces

In public comment spaces, visibility is individuated. A contribution may be present for its author and absent for everyone else. This produces an asymmetry that established models of interaction do not anticipate: a speaker cannot perceive their own invisibility, because their own comment is always visible to them. Only another participant can establish that it is not visible to anyone else. This report describes the sequence participants have developed in response. A first turn reports possible invisibility or requests proof of receipt, typically without a designated addressee. A second turn, produced by a self-selecting respondent, confirms receipt. In roughly a third of cases the confirmer asks the same question about themselves, because they are equally uncertain of their own visibility. The analysis draws on 165,527 comments from eight YouTube videos in two conditions. In four videos whose topic is comment deletion, 25 hand-validated instances occur (1.35 per 1,000 comments, 95% CI 0.91–1.98). In four heated dispute videos without that topic — including one with 99,983 comments and a 79% reply rate — none occur (Fisher exact p = 1.8 × 10⁻²⁴). Hand inspection of all 52 lexical candidates in the control condition found two genuine reports of invisibility, neither of which received a confirmation. The sequence is therefore not a general structural response to uncertain delivery but a practice bound to a community that has come to treat invisibility reports as answerable. The theoretical interest lies in the inversion: where a classical summons secures the recipient's attention, this one requests evidence of the speaker's own existence — evidence the speaker is structurally unable to produce alone. Private messaging provides delivery and read receipts; public comment sections do not. Participants produce socially what the platform withholds. Provenance. This study was carried out by a researcher without formal training in linguistics, working with a large language model (Claude, Anthropic) across a single week. The model proposed the framing, wrote the retrieval and detection code, conducted the literature search and drafted the text; the human collected the corpora, made all substantive decisions, and hand-validated every instance. Section 10 records the division of labour and, in particular, the errors each party caught in the other. The author places this on record deliberately: as AI-assisted research by non-specialists becomes common, the field will need documented cases — including their failure modes — to assess what such work is and is not worth. This paper is offered as one such case, and readers are invited to use it accordingly.