AI Parasocial Collapse Index (APCI) Dataset: YouTube Consumer Discourse on Parasocial Relationship Collapse Following AI Disclosure in Influencer Contexts (n=2,997, April 2026)

This dataset contains 2,997 YouTube comments collected viathe YouTube Data API v3 (April 2026) across five discoursedomains targeting naturalistic consumer responses to AIdisclosure in influencer parasocial relationship contexts: (1) influencer caught using AI to write replies comments(2) youtuber admits using AI chatgpt to respond fans(3) influencer AI ghostwriting exposed followers reaction(4) creator using AI to fake personal stories posts fans(5) AI generated influencer content disclosed audience The corpus is the primary empirical dataset for the AIParasocial Collapse Index (APCI) — the first computationalmetric measuring the ratio of Parasocial Collapse toParasocial Continuity following AI disclosure in influencercontexts, providing the first naturalistic discoursemeasurement of parasocial relationship breakdown dynamicsfollowing AI mediation revelation. APCI = Parasocial Collapse / (Parasocial Collapse + Parasocial Continuity)1.0 = pure Parasocial Collapse0.5 = contested zone — equal collapse and continuity0.0 = pure Parasocial Continuity Three signal types grounded in PSR theory: PARASOCIAL COLLAPSE (PC): APCI numerator → "I feel betrayed", "unfollowing", "you lied to us" → "I thought you were real", "heartbroken" → "don't know you anymore", "can't trust you" Grounded in: Horton & Wohl (1956) PSI theory; Cohen (2003) parasocial breakup; Weiner (1985) attribution theory; Expectancy Violation Theory PARASOCIAL CONTINUITY (PCo): APCI denominator partner → "still love you", "AI or not, content is great" → "doesn't matter", "who cares", "no big deal" Grounded in: Sheng et al. (2025) PSR resilience; Dibble et al. (2016) PSR maintenance IDENTITY DISSOLUTION (ID): novel modifier signal → "was any of this real?", "what is authenticity?" → "who even are you?", "is anything real?" Theoretically novel — not in any prior PSR model Grounded in: Cohen (2003) parasocial grief; Burgoon (1978) EVT — violation collapses identity KEY FINDINGS:- Total corpus: n=2,997 comments, ~50 videos- Categorised (APCI computed): 20 (0.7%)- Mean APCI: 0.5000 — CONTESTED ZONE Equal Parasocial Collapse and Continuity discourse- PC cluster: 10 (50% of categorised)- PCo cluster: 10 (50%)- Identity Dissolution: 0 detected at lexical level- PC density: 0.0235/100 tokens- PCo density: 0.0133/100 tokens DOMAIN-LEVEL APCI:- "creator using AI to fake personal stories": APCI=0.7000 — Collapse dominant PC=4.13/100 | PCo=1.77/100- "youtuber admits using AI to respond fans": APCI=0.6000 — Collapse dominant PC=5.84/100 | PCo=1.89/100- "AI generated content disclosed": APCI=0.0000 — pure Continuity PC=0.00 | PCo=2.35/100- "influencer caught using AI to write replies": APCI=0.0000 — pure Continuity CRITICAL FINDING — FRAMING COLLAPSE GRADIENT:Faking personal stories → APCI=0.7000 (Collapse)Replying with AI → APCI=0.0000 (Continuity)Same AI use — opposite parasocial outcomes.The degree of PERSONAL INTIMACY VIOLATION determinescollapse, not AI use per se. Consistent withExpectancy Violation Theory (Burgoon, 1978). Most-liked PC comment (1,545 likes):"Man PMdamiann literally made a video called'Inseparable Duo'. Now he cut contact with h..."→ Cohen (2003) parasocial breakup grief confirmed in naturalistic discourse. Most-liked PCo comment (3,863 likes):"The fact you need to analyze videos with attentionto know if they are AI or not really speaks..."→ Acceptance when AI is sophisticated/invisible Files:- apci_videos.csv: ~50 unique videos metadata- apci_comments.csv: 2,997 raw comments- apci_results.csv: APCI scores + signal annotation Method: YouTube Data API v3. Python 3.12, April 2026.